Compare commits

...

27 Commits

Author SHA1 Message Date
byGalax d5c54166a4 fix(migrate): restore storage object xattrs lost in the server move
CI / verify (push) Has been cancelled
After the netralax.cloud -> netralax.de migration every storage object GET
returned HTTP 500 (ENODATA "The extended attribute does not exist"). Root
cause: the object bytes were copied but Supabase Storage (file backend,
v1.48.26) keeps each object's response metadata in Linux xattrs
(user.supabase.{content-type,cache-control,etag}); the copy did not preserve
them. The old server is gone, but the values survive in storage.objects.metadata,
so this script reconstructs the xattrs from the DB. Verified: public avatar GETs
went 500 -> 200 after running it; all 18 objects (avatars, banner, attachments)
restored, 0 files genuinely missing. Idempotent, touches no object bytes.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-02 23:59:46 +02:00
byGalax f438018400 chore(desktop): release v0.21.11 2026-06-02 23:07:34 +02:00
byGalax 89003f71a4 fix(desktop): remove chat-switch reveal flicker (decouple from reactions)
The residual flicker on chat switch was a loading/reveal artifact, not scroll.
listReady gated the MessageList reveal on reactionsReady OR a 300ms timeout, so
on a cache-hit switch (messages already present from the first render) the list
sat at opacity:0 for up to 300ms and then popped in. Drop the reactions/timeout
gate: reveal as soon as messages exist. Reaction chips stream in a beat later;
because the list is pinned to the bottom their height growth re-pins with no
visible jump, and MessageList still defers its own reveal a few frames until the
row-height measurement settles so it appears already at the final bottom.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-02 23:02:07 +02:00
byGalax b364c53c61 fix(desktop): degrade missing avatar image to letter circle
Avatar rendered a bare <img> with no error handling, so an avatar_url whose
storage object is unreachable (e.g. a 404 after the server move) showed a
broken image instead of the coloured letter-circle fallback. Track an onError
flag and fall back to the circle; reset it when the URL changes so a fresh
valid avatar is retried. This is client-side resilience only — it does not
restore a genuinely missing storage object.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-02 23:02:07 +02:00
byGalax 9c5456b492 chore(desktop): release v0.21.10 2026-06-02 22:32:52 +02:00
byGalax b057795735 fix(desktop): stop chat opening at top + flicker on switch
Make stick-to-bottom intent the single source of truth in MessageList and
drive onAtBottomChange from intent, not raw scroll position. A measurement
reflow can no longer flip the intent off (RC1), the second scrollPositions
writer no longer persists a drifting topmost index while stuck (RC2), and a
pin-on-rows layout effect re-pins through the two-phase data swap (RC3).

- scrollController: add tested nextStickIntent() state machine
- MessageList: input-event-based unstick (wheel/key/touch + scrollbar drag),
  reveal after 2 stable frames, tabIndex for keyboard nav, remove debug overlay
- ConversationPage: reuse resolveInitialAnchor; harden handleRangeChanged

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-02 22:31:39 +02:00
byGalax c81a036c4e chore(desktop): release v0.21.9 2026-06-02 21:48:43 +02:00
byGalax bdc017e609 fix(desktop): MessageList sticks to bottom via ResizeObserver + scroll guard
Data from the on-screen overlay showed SCROLLABLE=YES but scrollTop=84/0 and atBottom=false: the initial pin happened before rows finished measuring, then a measurement reflow fired onScroll with the stale (top) scrollTop, flipping atBottom=false and disabling re-pinning, so the list never reached the bottom. Fix: a ResizeObserver re-pins to the true bottom as the content measures/grows; a programmatic-scroll guard makes onScroll ignore the scrolls we cause (so measurement reflows no longer flip the stick intent); overflow-anchor:none so the browser doesn't fight us; reveal waits for the height to settle. Overlay kept for one more verification pass.
2026-06-02 21:47:15 +02:00
byGalax 27160145f9 chore(desktop): release v0.21.8 2026-06-02 21:37:29 +02:00
byGalax 8ea2cb48e9 debug(desktop): on-screen scroll-metrics overlay (temporary) 2026-06-02 21:34:04 +02:00
byGalax c3ef995404 chore(desktop): release v0.21.7 2026-06-02 21:04:37 +02:00
byGalax b4ed3aced0 fix(desktop): MessageList anchors via direct scrollTop + flex-1 height
scrollToIndex raced the virtualizer's own layout effect and depended on size estimates, leaving the list pinned at the top on open (and flickering as it settled). Drive scrollTop = scrollHeight directly for the bottom case (order-independent, true bottom) and re-pin on measure; switch the scroll root from h-full to flex-1 min-h-0 so it always has a bounded, scrollable height.
2026-06-02 21:00:19 +02:00
byGalax 30d00194be chore(desktop): release v0.21.6 2026-06-02 20:46:53 +02:00
byGalax 31b394a6e0 chore(desktop): drop react-virtuoso + scroll debug instrumentation 2026-06-02 20:44:56 +02:00
byGalax 271d6fff5c feat(desktop): use MessageList in ConversationPage (replace react-virtuoso) 2026-06-02 20:32:08 +02:00
byGalax 8b8d71bc4d feat(desktop): TanStack-Virtual MessageList with deferred reveal 2026-06-02 20:27:31 +02:00
byGalax 40f36cb182 refactor(desktop): export VirtuosoRow type for MessageList 2026-06-02 20:26:21 +02:00
byGalax 2372731504 feat(desktop): expose reactions reveal-gate flag (ready) 2026-06-02 20:26:00 +02:00
byGalax 43a99a8d6d feat(desktop): pure scroll-decision logic for new message list 2026-06-02 20:25:06 +02:00
byGalax f73abbd860 build(desktop): add @tanstack/react-virtual 2026-06-02 20:24:15 +02:00
byGalax e822f6f58f docs(plan): message-list / scroll rewrite implementation plan
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-02 20:22:28 +02:00
byGalax 255dbdc712 docs(spec): message-list / scroll rewrite design (TanStack Virtual + deferred reveal)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-02 20:16:42 +02:00
byGalax 51a8630114 chore(desktop): release v0.21.5 2026-06-02 19:41:46 +02:00
byGalax d539656535 fix(conversation): anchor message list to bottom on chat switch
initialTopMostItemIndex was a plain index (top-aligned), so react-virtuoso painted with estimated row heights then corrected scrollTop after measuring the real (taller) dynamic bubbles — a visible jump on every chat switch. Use { index: 'LAST', align: 'end' } to pin the bottom edge instead, matching react-virtuoso's canonical chat pattern; the restore-to-saved-row path stays align: 'start'.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-02 19:39:04 +02:00
byGalax 5bc30c950c feat(infra): migrate self-hosted backend to netralax.de
Move Supabase + LiveKit from the netralax.cloud VPS to a new netralax.de server. Adds the migration runbook (docs/), one-time move scripts (scripts/migrate/), and prod Caddy/LiveKit config templates (infra/). Repoints the desktop publish/changelog URLs and prod ops config to .de. JWT_SECRET + VAPID copied identically so already-installed clients keep working; the new server also serves the legacy .cloud hostnames.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-02 19:39:04 +02:00
byGalax 588b843904 chore(desktop): release v0.21.4 2026-05-21 23:12:38 +02:00
byGalax dbf8030e93 fix(conv-key): rotate-instead-of-share + cache invalidation; fire-first realtime inserts (no sound-vs-text gap)
Friend-DM "Nachricht nicht lesbar" recurred even after v0.21.3 because the
proactive sweep called shareConvKeyToUser, which reads the module-level
conv-key cache first. After a server-side cleanup the cache still held the
stale locally-bootstrapped key, so each side wrapped its own different key
for the peer and the bundles diverged anew.

Switch the sweep to rotate_conv_key when any peer's user-id bundle is
missing at the active version: a fresh symmetric key is generated, wrapped
for every member at their CURRENT pubkey, and the active version is bumped
under a row-level FOR UPDATE lock. Concurrent rotations are race-safe — the
loser sees "new version must be greater" and bails; the winner's bundles
propagate via realtime.

Realtime conversation_keys subscription now invalidates the cache for the
affected (conversationId, key_version) on any INSERT/UPDATE/DELETE — so
admin cleanups, peer rotations, or device wraps can no longer leave a
stale entry in this client's session cache.

queueInsert now fires the first event of a quiet period immediately and
only collapses follow-up bursts. BATCH_WINDOW_MS dropped 250 → 80 ms.
This closes the ~250 ms gap between the notification sound and the
message body appearing.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-21 23:10:36 +02:00
27 changed files with 3494 additions and 177 deletions
+5 -1
View File
@@ -7,7 +7,11 @@
# stopped publishing Tauri releases to this host. # stopped publishing Tauri releases to this host.
# Host serving latest.yml + installer artifacts over HTTPS. # Host serving latest.yml + installer artifacts over HTTPS.
UPDATE_HOST=update.netralax.cloud # NOTE: during the .cloud→.de transition the new VPS must ALSO serve the same
# artifacts under update.netralax.cloud (point its DNS at the new IP) so that
# already-installed clients — which have update.netralax.cloud baked in — can
# still pull the release that switches them over to .de.
UPDATE_HOST=update.netralax.de
# SSH user on UPDATE_HOST with write access to UPDATE_REMOTE_PATH. # SSH user on UPDATE_HOST with write access to UPDATE_REMOTE_PATH.
UPDATE_SSH_USER=chatapp-deploy UPDATE_SSH_USER=chatapp-deploy
+3 -3
View File
@@ -1,6 +1,6 @@
{ {
"name": "@chat-app/desktop", "name": "@chat-app/desktop",
"version": "0.21.3", "version": "0.21.11",
"private": true, "private": true,
"description": "Electron desktop client (Windows / macOS / Linux)", "description": "Electron desktop client (Windows / macOS / Linux)",
"type": "module", "type": "module",
@@ -24,6 +24,7 @@
"@livekit/components-react": "^2.9.0", "@livekit/components-react": "^2.9.0",
"@livekit/track-processors": "^0.7.2", "@livekit/track-processors": "^0.7.2",
"@supabase/supabase-js": "^2.46.0", "@supabase/supabase-js": "^2.46.0",
"@tanstack/react-virtual": "^3.10.0",
"better-sqlite3": "^11.3.0", "better-sqlite3": "^11.3.0",
"canvas-confetti": "^1.9.4", "canvas-confetti": "^1.9.4",
"electron-updater": "^6.3.0", "electron-updater": "^6.3.0",
@@ -35,7 +36,6 @@
"react-easy-crop": "^5.5.7", "react-easy-crop": "^5.5.7",
"react-i18next": "^15.1.1", "react-i18next": "^15.1.1",
"react-router-dom": "^6.28.0", "react-router-dom": "^6.28.0",
"react-virtuoso": "^4.18.7",
"zustand": "^5.0.1" "zustand": "^5.0.1"
}, },
"devDependencies": { "devDependencies": {
@@ -95,7 +95,7 @@
"publish": [ "publish": [
{ {
"provider": "generic", "provider": "generic",
"url": "https://update.netralax.cloud/windows/" "url": "https://update.netralax.de/windows/"
} }
] ]
} }
+10 -1
View File
@@ -1,6 +1,8 @@
// Reusable avatar that prefers an uploaded image and falls back to a coloured // Reusable avatar that prefers an uploaded image and falls back to a coloured
// letter circle. Use this everywhere the app needs to render a profile. // letter circle. Use this everywhere the app needs to render a profile.
import { useEffect, useState } from 'react';
import { useCachedAvatarUrl } from '../lib/avatarCache'; import { useCachedAvatarUrl } from '../lib/avatarCache';
interface Props { interface Props {
@@ -27,7 +29,13 @@ export function Avatar({
loading = 'lazy', loading = 'lazy',
}: Props) { }: Props) {
const effectiveUrl = useCachedAvatarUrl(url); const effectiveUrl = useCachedAvatarUrl(url);
if (effectiveUrl) { // If the image URL is non-empty but unreachable (e.g. the storage object is
// missing / 404s), the bare <img> would render broken with no fallback.
// Track a load error and degrade to the letter circle instead. Reset on URL
// change so a fresh, valid avatar is retried.
const [failed, setFailed] = useState(false);
useEffect(() => setFailed(false), [effectiveUrl]);
if (effectiveUrl && !failed) {
return ( return (
<img <img
src={effectiveUrl} src={effectiveUrl}
@@ -35,6 +43,7 @@ export function Avatar({
className={'shrink-0 rounded-full object-cover ' + className} className={'shrink-0 rounded-full object-cover ' + className}
draggable={false} draggable={false}
loading={loading} loading={loading}
onError={() => setFailed(true)}
/> />
); );
} }
+325
View File
@@ -0,0 +1,325 @@
import { useVirtualizer } from '@tanstack/react-virtual';
import {
forwardRef,
useCallback,
useEffect,
useImperativeHandle,
useLayoutEffect,
useRef,
useState,
type ReactNode,
} from 'react';
import { isNearBottom, isNearTop, nextStickIntent } from '../lib/scrollController';
import type { VirtuosoRow } from '../pages/ConversationPage';
export interface MessageListHandle {
scrollToBottom(behavior?: ScrollBehavior): void;
scrollToRow(index: number, align?: 'center' | 'end', behavior?: ScrollBehavior): void;
}
export interface MessageListProps {
rows: VirtuosoRow[];
renderRow: (index: number, row: VirtuosoRow) => ReactNode;
computeKey: (row: VirtuosoRow) => string;
/** Initial scroll target for a freshly-mounted list. */
initialAnchor: { type: 'bottom' } | { type: 'row'; index: number };
/** Reveal gate — the list stays hidden until reactions/heights are loaded, so
* the post-paint height cascade is never visible. */
ready: boolean;
estimateRowHeight?: number;
atBottomThreshold?: number;
onReachTop?: () => void;
onAtBottomChange?: (atBottom: boolean) => void;
onTopRowChange?: (topIndex: number) => void;
}
export const MessageList = forwardRef<MessageListHandle, MessageListProps>(function MessageList(
{
rows,
renderRow,
computeKey,
initialAnchor,
ready,
estimateRowHeight = 64,
atBottomThreshold = 64,
onReachTop,
onAtBottomChange,
onTopRowChange,
},
ref,
) {
const scrollElRef = useRef<HTMLDivElement>(null);
const [revealed, setRevealed] = useState(false);
// THE single source of truth: should the view stay pinned to the bottom?
// Only a genuine user up-input (wheel / key / touch / scrollbar drag) turns
// this OFF; only reaching the bottom turns it ON. A measurement reflow must
// never flip it — that was the root cause of the chat-switch bug.
const stickRef = useRef(true);
// Debounce for onAtBottomChange — fire the parent only on a real transition.
const lastReportedAtBottomRef = useRef<boolean | null>(null);
// Guard: scrolls WE cause (pin / measure re-pin / scrollToIndex) fire onScroll
// a tick later. Within this window we don't treat a scrollTop decrease as the
// user dragging up.
const programmaticRef = useRef(0);
// Previous scrollTop, to detect a genuine scrollbar/keyboard up-drag.
const lastScrollTopRef = useRef(0);
// Load-older preservation: remember the first row key + scrollHeight so a
// prepend can be detected and the viewport restored.
const prevFirstKeyRef = useRef<string | null>(null);
const prevScrollHeightRef = useRef(0);
// Latest onAtBottomChange, read through a ref so the input-listener effect
// can stay mounted once (deps []) without capturing a stale callback.
const onAtBottomChangeRef = useRef(onAtBottomChange);
onAtBottomChangeRef.current = onAtBottomChange;
const virtualizer = useVirtualizer({
count: rows.length,
getScrollElement: () => scrollElRef.current,
estimateSize: () => estimateRowHeight,
overscan: 8,
getItemKey: (index) => computeKey(rows[index]!),
});
const readMetrics = useCallback(() => {
const el = scrollElRef.current;
return el
? { scrollTop: el.scrollTop, scrollHeight: el.scrollHeight, clientHeight: el.clientHeight }
: { scrollTop: 0, scrollHeight: 0, clientHeight: 0 };
}, []);
const pinToBottom = useCallback(() => {
const el = scrollElRef.current;
if (!el) return;
programmaticRef.current = performance.now();
el.scrollTop = el.scrollHeight;
lastScrollTopRef.current = el.scrollTop;
}, []);
// Report at-bottom to the parent only on a true transition, always driven by
// the INTENT (stickRef) — never the raw position. This is what kills the
// feedback loop: a transient "not at bottom" mid-reflow is never persisted.
const reportAtBottom = useCallback((atBottom: boolean) => {
if (lastReportedAtBottomRef.current === atBottom) return;
lastReportedAtBottomRef.current = atBottom;
onAtBottomChangeRef.current?.(atBottom);
}, []);
// A genuine user up-input: drop the stick intent immediately.
const markUserMovedUp = useCallback(() => {
if (!stickRef.current) return;
stickRef.current = false;
reportAtBottom(false);
}, [reportAtBottom]);
// Re-pin to the true bottom whenever the content (or viewport) resizes while
// sticking. ResizeObserver fires after layout / before paint, so as rows
// measure and the list grows the bottom stays pinned with no stale frame.
useEffect(() => {
const el = scrollElRef.current;
if (!el) return;
const ro = new ResizeObserver(() => {
const e = scrollElRef.current;
if (stickRef.current && e) {
programmaticRef.current = performance.now();
e.scrollTop = e.scrollHeight;
lastScrollTopRef.current = e.scrollTop;
}
});
ro.observe(el);
const inner = el.firstElementChild;
if (inner) ro.observe(inner);
return () => ro.disconnect();
}, []);
// Genuine-user-intent listeners. These are the ONLY way (besides reaching the
// bottom) the stick intent turns off, so a reflow can never unstick the list.
useEffect(() => {
const el = scrollElRef.current;
if (!el) return;
const onWheel = (e: WheelEvent) => {
if (e.deltaY < 0) markUserMovedUp();
};
const onKeyDown = (e: KeyboardEvent) => {
if (e.key === 'PageUp' || e.key === 'Home' || e.key === 'ArrowUp') markUserMovedUp();
};
let touchStartY = 0;
const onTouchStart = (e: TouchEvent) => {
touchStartY = e.touches[0]?.clientY ?? 0;
};
const onTouchMove = (e: TouchEvent) => {
const y = e.touches[0]?.clientY ?? 0;
// Finger dragged DOWN (content scrolls up toward older messages). Guard on
// scrollTop>0 so an overscroll bounce at the bottom doesn't unstick.
if (y - touchStartY > 8 && (scrollElRef.current?.scrollTop ?? 0) > 0) markUserMovedUp();
};
el.addEventListener('wheel', onWheel, { passive: true });
el.addEventListener('keydown', onKeyDown);
el.addEventListener('touchstart', onTouchStart, { passive: true });
el.addEventListener('touchmove', onTouchMove, { passive: true });
return () => {
el.removeEventListener('wheel', onWheel);
el.removeEventListener('keydown', onKeyDown);
el.removeEventListener('touchstart', onTouchStart);
el.removeEventListener('touchmove', onTouchMove);
};
}, [markUserMovedUp]);
// Deferred reveal: when ready, pin to the anchor and keep pinning each frame
// until the list height has SETTLED over two consecutive frames, THEN reveal —
// so what appears is already at its final position with no top-then-jump.
useLayoutEffect(() => {
if (!ready || revealed || rows.length === 0) return;
const el = scrollElRef.current;
if (!el) return;
const rowIdx = Math.max(
0,
Math.min(initialAnchor.type === 'row' ? initialAnchor.index : 0, rows.length - 1),
);
if (initialAnchor.type === 'bottom') {
stickRef.current = true;
pinToBottom();
} else {
stickRef.current = false;
programmaticRef.current = performance.now();
virtualizer.scrollToIndex(rowIdx, { align: 'start' });
}
reportAtBottom(stickRef.current);
let prevSH = -1;
let stableFrames = 0;
const settle = (attempts: number): void => {
const e = scrollElRef.current;
if (!e) {
setRevealed(true);
return;
}
programmaticRef.current = performance.now();
if (stickRef.current) {
e.scrollTop = e.scrollHeight;
lastScrollTopRef.current = e.scrollTop;
} else {
virtualizer.scrollToIndex(rowIdx, { align: 'start' });
}
const sh = e.scrollHeight;
// Require TWO consecutive stable-height frames: a single stable frame can
// land mid-cascade (between reactions and the unread divider measuring)
// and reveal a not-yet-final layout that then jumps.
stableFrames = sh === prevSH ? stableFrames + 1 : 0;
prevSH = sh;
if (stableFrames >= 2 || attempts <= 0) {
setRevealed(true);
} else {
requestAnimationFrame(() => settle(attempts - 1));
}
};
requestAnimationFrame(() => settle(12));
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [ready, rows.length]);
// Load-older preservation: if rows were prepended (first key changed and the
// user is near the top), restore scrollTop by the height delta so the viewport
// stays put instead of jumping.
useLayoutEffect(() => {
const firstKey = rows.length > 0 ? computeKey(rows[0]!) : null;
const el = scrollElRef.current;
if (el && revealed && prevFirstKeyRef.current && firstKey !== prevFirstKeyRef.current) {
const delta = el.scrollHeight - prevScrollHeightRef.current;
if (delta > 0 && el.scrollTop < atBottomThreshold * 4) {
el.scrollTop += delta;
lastScrollTopRef.current = el.scrollTop;
}
}
prevFirstKeyRef.current = firstKey;
prevScrollHeightRef.current = el?.scrollHeight ?? 0;
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [rows]);
// Re-pin on any rows change while sticking. Covers the two-phase data swap
// (the cached array is replaced by the freshly-decrypted one ~100ms after
// reveal) which the ResizeObserver can miss when the new content happens to
// measure to the same height.
useLayoutEffect(() => {
if (revealed && stickRef.current) pinToBottom();
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [rows]);
const handleScroll = useCallback(() => {
const m = readMetrics();
const programmatic = performance.now() - programmaticRef.current < 120;
const nearBottom = isNearBottom(m, atBottomThreshold);
// A scrollbar drag or keyboard scroll surfaces here as a scrollTop decrease.
// Suppress it inside the programmatic window so our own re-pin / settle is
// never mistaken for the user moving up. 2px deadzone absorbs sub-pixel jitter.
const userMovedUp = !programmatic && m.scrollTop < lastScrollTopRef.current - 2;
lastScrollTopRef.current = m.scrollTop;
stickRef.current = nextStickIntent(stickRef.current, { nearBottom, userMovedUp });
reportAtBottom(stickRef.current);
if (isNearTop(m, atBottomThreshold * 4)) onReachTop?.();
const first = virtualizer.getVirtualItems()[0];
if (first) onTopRowChange?.(first.index);
}, [atBottomThreshold, onReachTop, onTopRowChange, readMetrics, reportAtBottom, virtualizer]);
useImperativeHandle(
ref,
() => ({
scrollToBottom: () => {
stickRef.current = true;
reportAtBottom(true);
pinToBottom();
},
scrollToRow: (index, align = 'center') => {
// The user is jumping to a specific row — drop the stick intent first so
// the ResizeObserver doesn't immediately drag the target back to the bottom.
stickRef.current = false;
reportAtBottom(false);
programmaticRef.current = performance.now();
virtualizer.scrollToIndex(index, { align });
},
}),
// eslint-disable-next-line react-hooks/exhaustive-deps
[virtualizer, rows.length, reportAtBottom, pinToBottom],
);
const items = virtualizer.getVirtualItems();
return (
<div
ref={scrollElRef}
onScroll={handleScroll}
tabIndex={0}
className="min-h-0 flex-1 overflow-y-auto"
style={{
opacity: revealed ? 1 : 0,
position: 'relative',
overflowAnchor: 'none',
outline: 'none',
}}
>
<div style={{ height: virtualizer.getTotalSize(), position: 'relative', width: '100%' }}>
{items.map((vi) => (
<div
key={vi.key}
data-index={vi.index}
ref={virtualizer.measureElement}
style={{
position: 'absolute',
top: 0,
left: 0,
width: '100%',
transform: `translateY(${vi.start}px)`,
}}
>
{renderRow(vi.index, rows[vi.index]!)}
</div>
))}
</div>
{/* 12px bottom breathing space (matches the old Virtuoso Footer). */}
<div style={{ height: 12 }} />
</div>
);
});
+6 -2
View File
@@ -1,11 +1,15 @@
// In-app changelog feed. // In-app changelog feed.
// //
// The release script (`scripts/release.mjs`) maintains a single // The release script (`scripts/release.mjs`) maintains a single
// `changelog.json` file alongside `latest.json` on update.netralax.cloud. // `changelog.json` file alongside `latest.json` on update.netralax.de.
// The list is newest-first, capped at 200 entries server-side, and rewritten // The list is newest-first, capped at 200 entries server-side, and rewritten
// after every release. // after every release.
//
// NOTE: already-installed clients still fetch this from update.netralax.cloud
// (baked into their bundle), so Caddy on the new VPS must keep serving the
// update.netralax.cloud vhost from the same directory during the transition.
const CHANGELOG_URL = 'https://update.netralax.cloud/windows/changelog.json'; const CHANGELOG_URL = 'https://update.netralax.de/windows/changelog.json';
export interface ChangelogEntry { export interface ChangelogEntry {
version: string; version: string;
@@ -0,0 +1,73 @@
import { describe, expect, it } from 'vitest';
import { isNearBottom, isNearTop, nextStickIntent, resolveInitialAnchor } from './scrollController';
const m = (scrollTop: number, scrollHeight: number, clientHeight: number) => ({
scrollTop,
scrollHeight,
clientHeight,
});
describe('isNearBottom', () => {
it('true exactly at the bottom', () => {
expect(isNearBottom(m(900, 1000, 100), 64)).toBe(true);
});
it('true within threshold', () => {
expect(isNearBottom(m(860, 1000, 100), 64)).toBe(true);
});
it('false beyond threshold', () => {
expect(isNearBottom(m(800, 1000, 100), 64)).toBe(false);
});
});
describe('isNearTop', () => {
it('true at top', () => {
expect(isNearTop(m(0, 1000, 100), 64)).toBe(true);
});
it('false past threshold', () => {
expect(isNearTop(m(200, 1000, 100), 64)).toBe(false);
});
});
describe('resolveInitialAnchor', () => {
it('anchors to last row at end by default (no saved position)', () => {
expect(resolveInitialAnchor(null, 50)).toEqual({ index: 49, align: 'end' });
});
it('anchors to bottom when saved position stuck to bottom', () => {
expect(resolveInitialAnchor({ topmostIndex: 10, stickToBottom: true }, 50)).toEqual({
index: 49,
align: 'end',
});
});
it('restores the saved row at the top when scrolled up', () => {
expect(resolveInitialAnchor({ topmostIndex: 12, stickToBottom: false }, 50)).toEqual({
index: 12,
align: 'start',
});
});
it('clamps a stale saved index to the current row count', () => {
expect(resolveInitialAnchor({ topmostIndex: 999, stickToBottom: false }, 50)).toEqual({
index: 49,
align: 'start',
});
});
it('handles an empty list', () => {
expect(resolveInitialAnchor(null, 0)).toEqual({ index: 0, align: 'end' });
});
});
describe('nextStickIntent', () => {
it('turns ON when the bottom is reached', () => {
expect(nextStickIntent(false, { nearBottom: true, userMovedUp: false })).toBe(true);
});
it('turns OFF when the user genuinely moves up', () => {
expect(nextStickIntent(true, { nearBottom: false, userMovedUp: true })).toBe(false);
});
it('keeps the previous intent on a neutral scroll (measurement reflow)', () => {
expect(nextStickIntent(true, { nearBottom: false, userMovedUp: false })).toBe(true);
expect(nextStickIntent(false, { nearBottom: false, userMovedUp: false })).toBe(false);
});
it('reaching the bottom wins over a simultaneous move-up signal', () => {
expect(nextStickIntent(false, { nearBottom: true, userMovedUp: true })).toBe(true);
});
});
+62
View File
@@ -0,0 +1,62 @@
// Pure, DOM-free scroll-decision logic for MessageList. Unit-tested so the
// tricky math is verified without a browser (jsdom has no layout).
export interface ScrollMetrics {
scrollTop: number;
scrollHeight: number;
clientHeight: number;
}
/** Distance from the bottom edge is within `threshold` px. */
export function isNearBottom(m: ScrollMetrics, threshold: number): boolean {
return m.scrollHeight - (m.scrollTop + m.clientHeight) <= threshold;
}
/** Scroll offset is within `threshold` px of the top. */
export function isNearTop(m: ScrollMetrics, threshold: number): boolean {
return m.scrollTop <= threshold;
}
export interface SavedPosition {
topmostIndex: number;
stickToBottom: boolean;
}
export interface Anchor {
index: number;
align: 'start' | 'end';
}
/**
* Where a freshly-opened chat should start.
* - default / "left at bottom" → last row, aligned to the viewport bottom.
* - "left scrolled up" → the saved top-most row, aligned to the viewport top
* (clamped in case the cached row count shrank).
*/
export function resolveInitialAnchor(saved: SavedPosition | null, rowCount: number): Anchor {
if (rowCount <= 0) return { index: 0, align: 'end' };
if (saved && !saved.stickToBottom) {
const index = Math.max(0, Math.min(saved.topmostIndex, rowCount - 1));
return { index, align: 'start' };
}
return { index: rowCount - 1, align: 'end' };
}
/**
* The next stick-to-bottom intent given the current intent and the latest
* scroll signal. Intent only flips on a *definitive* signal:
* - reaching the bottom turns it ON,
* - a genuine user move-up turns it OFF.
* A neutral scroll — e.g. a measurement reflow that grows the content while
* rows settle — leaves the intent unchanged. This is the core fix for the
* chat-switch bug: a reflow must never be mistaken for the user scrolling up
* and so must never silently unstick the list.
*/
export function nextStickIntent(
prev: boolean,
signal: { nearBottom: boolean; userMovedUp: boolean },
): boolean {
if (signal.nearBottom) return true;
if (signal.userMovedUp) return false;
return prev;
}
+123 -66
View File
@@ -1,6 +1,6 @@
import { fetchPeerPublicKeys } from '@chat-app/shared/auth';
import { import {
type AttachmentHandle, type AttachmentHandle,
clearConvKeyCache,
type DecryptedMessage, type DecryptedMessage,
decryptMessages, decryptMessages,
encryptAndUploadAttachment, encryptAndUploadAttachment,
@@ -9,8 +9,8 @@ import {
insertAttachmentRow, insertAttachmentRow,
MAX_ATTACHMENT_BYTES, MAX_ATTACHMENT_BYTES,
type MessageWithCipher, type MessageWithCipher,
rotateConvKey,
sendEncryptedMessage, sendEncryptedMessage,
shareConvKeyToUser,
} from '@chat-app/shared/chat'; } from '@chat-app/shared/chat';
import { bytesToPgHex, pgBytesToBytes } from '@chat-app/shared/supabase'; import { bytesToPgHex, pgBytesToBytes } from '@chat-app/shared/supabase';
import { useCallback, useEffect, useMemo, useRef, useState } from 'react'; import { useCallback, useEffect, useMemo, useRef, useState } from 'react';
@@ -125,12 +125,25 @@ export function useConversationMessages({ conversationId, userId, deviceId }: Ar
}); });
}, [userId]); }, [userId]);
// Proactive rewrap sweep: when a conversation opens, walk every accepted // Proactive rewrap sweep: when a conversation opens, ensure the active
// member and ensure the active conv-key has a `recipient_user_id` bundle // conv-key has a `recipient_user_id` bundle for every accepted member.
// for them. Members who are missing one (typically peers who haven't yet //
// migrated to the per-user key model) get a best-effort wrap from the // If any peer is missing a bundle at the active version, the previous
// local conv-key handle. Closes the legacy migration gap so peer B can // implementation called `shareConvKeyToUser` for each missing peer
// read on first unlock without manual intervention from A. // that helper reads from the module-level conv-key cache first, and if
// the cache held a STALE locally-generated key (from a buggy bootstrap
// race in an earlier app version), the stale key got propagated to the
// peer's row. Both sides then encrypt with mutually un-mergeable keys
// and every message is "Nachricht nicht lesbar" forever (incident:
// conv aae12d84).
//
// The replacement: when any peer is missing, call `rotateConvKey` once.
// Rotation generates a fresh symmetric key locally, fetches each member's
// CURRENT pubkey, wraps the fresh key for everyone, and atomically bumps
// `active_key_version` via the `rotate_conv_key` RPC (FOR UPDATE lock
// serialises concurrent rotations). This bypasses the cache entirely:
// the new version's cache entry is the just-rotated key, and the stale
// entry at the old version is irrelevant because nobody reads it any more.
useEffect(() => { useEffect(() => {
if (!conversationId || !userId) return; if (!conversationId || !userId) return;
let cancelled = false; let cancelled = false;
@@ -154,72 +167,76 @@ export function useConversationMessages({ conversationId, userId, deviceId }: Ar
// db-types snapshot predates the active_key_version column; cast via unknown. // db-types snapshot predates the active_key_version column; cast via unknown.
const version = (convRow as unknown as { active_key_version: number }).active_key_version; const version = (convRow as unknown as { active_key_version: number }).active_key_version;
// Use `getOrCreateConvKey` rather than `tryGetConvKey` so that if we // First, make sure we have a usable handle for the active version
// can't unwrap our bundle at the active version (we lost the device // (this auto-rotates if we're locked out of our own bundle — the
// key, or the bundle was wiped by the 0.18.0 reset_user_key bug, or // recovery path added in v0.21.1/v0.21.2).
// we only ever had a legacy `recipient_device_id` row), the helper
// auto-rotates the conv-key to version+1 and wraps fresh bundles
// for every member with a `user_keys` row. This is the only path
// that recovers stuck-legacy conversations on the receive side —
// `tryGetConvKey` just returned null and left the chat permanently
// un-decryptable for the locked-out party.
const handle = await getOrCreateConvKey(supabase, conversationId, { const handle = await getOrCreateConvKey(supabase, conversationId, {
userId, userId,
privateKey: priv, privateKey: priv,
}); });
if (cancelled) return; if (cancelled) return;
// If the helper rotated, all current members with a `user_keys` if (handle.keyVersion > version) return; // already rotated by helper
// public key were already wrapped by `rotateConvKey`. No further
// sweep work is needed.
if (handle.keyVersion > version) return;
// Check membership state on the server.
const { data: members, error: mErr } = await supabase const { data: members, error: mErr } = await supabase
.from('conversation_members') .from('conversation_members')
.select('user_id, accepted') .select('user_id, accepted')
.eq('conversation_id', conversationId); .eq('conversation_id', conversationId);
if (mErr || !members) return; if (mErr || !members) return;
const memberIds = (members as Array<{ user_id: string; accepted: boolean }>) const peerIds = (members as Array<{ user_id: string; accepted: boolean }>)
.filter((m) => m.accepted && m.user_id !== userId) .filter((m) => m.accepted && m.user_id !== userId)
.map((m) => m.user_id); .map((m) => m.user_id);
if (memberIds.length === 0) return; if (peerIds.length === 0) return;
const peers = await fetchPeerPublicKeys(supabase, memberIds); // Count how many of the peers have a recipient_user_id bundle at
for (const peer of peers) { // the active version. If any are missing, rotate to V+1 — the
if (cancelled) return; // rotation will wrap a fresh key for every accepted member with a
const { count, error: cntErr } = await ( // user_keys row.
supabase as unknown as { const { data: existingRows, error: rowsErr } = await (
from: (t: string) => { supabase as unknown as {
select: (s: string, o?: object) => { from: (t: string) => {
eq: (...a: unknown[]) => { select: (s: string) => {
eq: (...a: unknown[]) => { eq: (c: string, v: string) => {
eq: ( eq: (c: string, v: number) => {
...a: unknown[] in: (c: string, v: string[]) => Promise<{
) => Promise<{ count: number | null; error: unknown }>; data: Array<{ recipient_user_id: string }> | null;
}; error: unknown;
}>;
}; };
}; };
}; };
} };
)
.from('conversation_keys')
.select('recipient_user_id', { count: 'exact', head: true })
.eq('conversation_id', conversationId)
.eq('recipient_user_id', peer.userId)
.eq('key_version', version);
if (cntErr) continue;
if ((count ?? 0) === 0) {
try {
await shareConvKeyToUser(
supabase,
conversationId,
peer.userId,
peer.publicKey,
{ userId, privateKey: priv },
);
} catch (err) {
console.warn('proactive rewrap failed for', peer.userId, err);
}
} }
)
.from('conversation_keys')
.select('recipient_user_id')
.eq('conversation_id', conversationId)
.eq('key_version', version)
.in('recipient_user_id', peerIds);
if (rowsErr) return;
const wrappedPeerIds = new Set(
(existingRows ?? []).map((r) => r.recipient_user_id),
);
const missing = peerIds.filter((id) => !wrappedPeerIds.has(id));
if (missing.length === 0) return;
// At least one peer is missing a bundle — rotate. We deliberately do
// NOT use the cached conv-key here. The rotation generates a fresh
// key wrapped to every current member's CURRENT pubkey, so any
// staleness in the local cache for the OLD version is irrelevant
// going forward.
try {
await rotateConvKey(supabase, conversationId, {
userId,
privateKey: priv,
});
} catch (err) {
// Most likely cause: a concurrent peer also called rotate and
// won the race; their bumped active_key_version makes our
// `p_new_version <= cur_version` and the RPC raises. That's fine —
// the next chat-open / send will fetch the new active version and
// unwrap the bundle that peer wrapped for us.
console.warn('proactive rotate failed (likely concurrent rotation)', err);
} }
} catch (err) { } catch (err) {
console.warn('proactive rewrap sweep failed', err); console.warn('proactive rewrap sweep failed', err);
@@ -490,11 +507,14 @@ export function useConversationMessages({ conversationId, userId, deviceId }: Ar
void refresh(); void refresh();
// Batch INSERT bursts so a paste / backfill doesn't fire N parallel // Batch INSERT bursts so a paste / backfill doesn't fire N parallel
// refetches + decrypts. If more than BATCH_BURST_THRESHOLD ids arrive // refetches + decrypts. The first event in a quiet period fires
// within BATCH_WINDOW_MS, collapse to a single refresh() which pulls // `handleInsert` immediately so single incoming messages don't sit
// the last 100 in one query — cheaper and keeps order stable. For // behind a debounce timer (previous behaviour: 250 ms blank between
// lone inserts the per-id path stays so latency is unchanged. // notification-sound and message body). Subsequent events arriving
const BATCH_WINDOW_MS = 250; // within BATCH_WINDOW_MS of the first are buffered; if the burst grows
// past BATCH_BURST_THRESHOLD the buffered tail collapses into one
// `refresh()` instead of N individual refetches.
const BATCH_WINDOW_MS = 80;
const BATCH_BURST_THRESHOLD = 3; const BATCH_BURST_THRESHOLD = 3;
let burstBuffer: Array<Record<string, unknown>> = []; let burstBuffer: Array<Record<string, unknown>> = [];
let burstTimer: number | null = null; let burstTimer: number | null = null;
@@ -513,6 +533,15 @@ export function useConversationMessages({ conversationId, userId, deviceId }: Ar
} }
}; };
const queueInsert = (row: Record<string, unknown>) => { const queueInsert = (row: Record<string, unknown>) => {
if (burstBuffer.length === 0 && burstTimer === null) {
// First event in a quiet period — fire immediately so the user sees
// the message right when they hear the notification sound. Arm a
// short window in case a burst follows; follow-ups go through the
// buffer and may collapse into a refresh.
void handleInsert(row);
burstTimer = window.setTimeout(flushBurst, BATCH_WINDOW_MS);
return;
}
burstBuffer.push(row); burstBuffer.push(row);
if (burstTimer === null) { if (burstTimer === null) {
burstTimer = window.setTimeout(flushBurst, BATCH_WINDOW_MS); burstTimer = window.setTimeout(flushBurst, BATCH_WINDOW_MS);
@@ -539,18 +568,46 @@ export function useConversationMessages({ conversationId, userId, deviceId }: Ar
} }
}, },
) )
// When a peer device wraps the conversation-key for us (e.g. we just // Any conversation_keys change for this conv invalidates the cached
// registered a fresh device), re-decrypt the visible messages. // conv-key for the affected version. The module-level cache in
// shared/chat/convKeys.ts otherwise holds the previously-unwrapped key
// forever within a session — which is exactly what propagated the
// stale local bootstrap key in conv aae12d84, recreating divergent
// bundles after a server-side cleanup. Clearing on any INSERT/UPDATE/
// DELETE for the conv forces the next `getOrCreateConvKey` /
// `tryGetConvKey` call to re-fetch the canonical bundle from the
// server. Cheap (a single Map.delete), defensive, and avoids stale-
// cache propagation across all of {peer rotation, device wrap, admin
// cleanup}.
//
// We also keep the historical "device wrap → refresh" trigger so a
// freshly-registered device of our own re-decrypts in place.
.on( .on(
'postgres_changes', 'postgres_changes',
{ {
event: 'INSERT', event: '*',
schema: 'public', schema: 'public',
table: 'conversation_keys', table: 'conversation_keys',
filter: 'conversation_id=eq.' + conversationId, filter: 'conversation_id=eq.' + conversationId,
}, },
(payload: { new: { recipient_device_id?: string } }) => { (payload: {
if (payload.new?.recipient_device_id === deviceId) { eventType: 'INSERT' | 'UPDATE' | 'DELETE';
new: { recipient_device_id?: string; key_version?: number };
old: { recipient_device_id?: string; key_version?: number };
}) => {
const v =
payload.eventType === 'DELETE'
? payload.old?.key_version
: payload.new?.key_version;
if (typeof v === 'number') {
clearConvKeyCache(conversationId, v);
} else {
clearConvKeyCache(conversationId);
}
if (
payload.eventType === 'INSERT' &&
payload.new?.recipient_device_id === deviceId
) {
void refresh(); void refresh();
} }
}, },
+11 -1
View File
@@ -19,6 +19,10 @@ export interface UseMessageReactionsResult {
byMessage: Map<string, AggregatedReaction[]>; byMessage: Map<string, AggregatedReaction[]>;
toggle: (messageId: string, emoji: string) => Promise<void>; toggle: (messageId: string, emoji: string) => Promise<void>;
voteExclusive: (messageId: string, emoji: string, exclusiveEmojis: string[]) => Promise<void>; voteExclusive: (messageId: string, emoji: string, exclusiveEmojis: string[]) => Promise<void>;
// True once the reactions for the current message-id set have been fetched
// (or there are no messages). Drives MessageList's deferred reveal so the
// chat opens already showing reaction chips — no post-paint height jump.
ready: boolean;
} }
// Batch-fetches reactions for the given message ids + subscribes to the // Batch-fetches reactions for the given message ids + subscribes to the
@@ -29,10 +33,12 @@ export function useMessageReactions(
): UseMessageReactionsResult { ): UseMessageReactionsResult {
const idsKey = useMemo(() => messageIds.join(','), [messageIds]); const idsKey = useMemo(() => messageIds.join(','), [messageIds]);
const [rows, setRows] = useState<MessageReaction[]>([]); const [rows, setRows] = useState<MessageReaction[]>([]);
const [readyKey, setReadyKey] = useState<string | null>(null);
const refresh = useCallback(async () => { const refresh = useCallback(async () => {
if (messageIds.length === 0) { if (messageIds.length === 0) {
setRows([]); setRows([]);
setReadyKey(idsKey);
return; return;
} }
try { try {
@@ -40,6 +46,8 @@ export function useMessageReactions(
setRows(data); setRows(data);
} catch (err: unknown) { } catch (err: unknown) {
console.error('listReactionsForMessages failed', err); console.error('listReactionsForMessages failed', err);
} finally {
setReadyKey(idsKey);
} }
// eslint-disable-next-line react-hooks/exhaustive-deps // eslint-disable-next-line react-hooks/exhaustive-deps
}, [idsKey]); }, [idsKey]);
@@ -129,5 +137,7 @@ export function useMessageReactions(
[byMessage, myId, refresh], [byMessage, myId, refresh],
); );
return { byMessage, toggle, voteExclusive }; const ready = readyKey === idsKey;
return { byMessage, toggle, voteExclusive, ready };
} }
+55 -81
View File
@@ -3,7 +3,7 @@ import { extractErrorCode } from '@chat-app/shared/i18n';
import { lazy, Suspense, useCallback, useEffect, useMemo, useRef, useState } from 'react'; import { lazy, Suspense, useCallback, useEffect, useMemo, useRef, useState } from 'react';
import { useTranslation } from 'react-i18next'; import { useTranslation } from 'react-i18next';
import { useParams } from 'react-router-dom'; import { useParams } from 'react-router-dom';
import { Virtuoso, type VirtuosoHandle } from 'react-virtuoso'; import { MessageList, type MessageListHandle } from '../components/MessageList';
import { ComposerActionsMenu } from '../components/ComposerActionsMenu'; import { ComposerActionsMenu } from '../components/ComposerActionsMenu';
import { ConversationHeader } from '../components/ConversationHeader'; import { ConversationHeader } from '../components/ConversationHeader';
@@ -54,6 +54,7 @@ import {
createWatchTogetherPayload, createWatchTogetherPayload,
createGamePayload, createGamePayload,
} from '../lib/conversationFeatures'; } from '../lib/conversationFeatures';
import { resolveInitialAnchor } from '../lib/scrollController';
const WhiteboardModal = lazy(() => const WhiteboardModal = lazy(() =>
import('../components/WhiteboardModal').then((m) => ({ default: m.WhiteboardModal })), import('../components/WhiteboardModal').then((m) => ({ default: m.WhiteboardModal })),
); );
@@ -84,7 +85,7 @@ import { clearDraft, getDraftSync, setDraft } from '../lib/composerDraftStore';
// pending bubbles and the "load older" tile inside the same Virtuoso // pending bubbles and the "load older" tile inside the same Virtuoso
// instance means scroll-to-bottom / followOutput stay coherent across both // instance means scroll-to-bottom / followOutput stay coherent across both
// (we don't need a sibling scroll container for pending items). // (we don't need a sibling scroll container for pending items).
type VirtuosoRow = export type VirtuosoRow =
| { kind: 'loader'; key: string } | { kind: 'loader'; key: string }
| { kind: 'message'; key: string; message: DecryptedMessage; idx: number } | { kind: 'message'; key: string; message: DecryptedMessage; idx: number }
| { kind: 'pending'; key: string; item: OutboxItem }; | { kind: 'pending'; key: string; item: OutboxItem };
@@ -145,6 +146,17 @@ export function ConversationPage() {
voteExclusive: votePoll, voteExclusive: votePoll,
} = useMessageReactions(messageIds, session?.user.id); } = useMessageReactions(messageIds, session?.user.id);
// Reveal gate for MessageList: as soon as messages exist (cache hit = first
// render, so no spinner and no wait), let the list reveal. We deliberately do
// NOT gate on reactions readiness: on a cache-hit chat switch the messages are
// already present, and gating on the async reactions fetch held the list at
// opacity:0 for up to 300ms and then "popped" it in — that was the residual
// chat-switch flicker. Reaction chips stream in a beat later; because the list
// is pinned to the bottom, their height growth re-pins with no visible jump.
// MessageList still defers its own reveal a few frames until the row-height
// measurement settles, so the list still appears already at the final bottom.
const listReady = !loading && messages.length > 0;
const myId = session?.user.id; const myId = session?.user.id;
const ownMessageIds = useMemo( const ownMessageIds = useMemo(
@@ -303,7 +315,7 @@ export function ConversationPage() {
}, },
[send], [send],
); );
const virtuosoRef = useRef<VirtuosoHandle>(null); const listRef = useRef<MessageListHandle>(null);
const topmostIndexRef = useRef<number>(0); const topmostIndexRef = useRef<number>(0);
const fileInputRef = useRef<HTMLInputElement>(null); const fileInputRef = useRef<HTMLInputElement>(null);
const composerRef = useRef<HTMLTextAreaElement>(null); const composerRef = useRef<HTMLTextAreaElement>(null);
@@ -471,18 +483,14 @@ export function ConversationPage() {
// previous visit to this chat AND the user wasn't sticking to the // previous visit to this chat AND the user wasn't sticking to the
// bottom, restore the saved row index (clamped to the current row // bottom, restore the saved row index (clamped to the current row
// count in case the cache was trimmed). // count in case the cache was trimmed).
const initialTopMostIndex = useMemo(() => { // Reuses the unit-tested resolveInitialAnchor so the "where do I open" rule
const saved = savedPositionRef.current; // lives in one tested place. MessageList only reads this at reveal time (when
if (saved && !saved.stickToBottom) { // rows are loaded), so depending on the row count clamps a stale saved index
return Math.max(0, Math.min(saved.topmostIndex, virtuosoRows.length - 1)); // correctly without freezing a mount-time count of 0.
} const initialAnchor = useMemo<{ type: 'bottom' } | { type: 'row'; index: number }>(() => {
return virtuosoRows.length - 1; const anchor = resolveInitialAnchor(savedPositionRef.current, virtuosoRows.length);
// virtuosoRows.length changes when the conversation loads — that's the return anchor.align === 'end' ? { type: 'bottom' } : { type: 'row', index: anchor.index };
// intentional trigger so a freshly-loaded chat anchors to the bottom }, [virtuosoRows.length]);
// on first paint. We deliberately don't re-derive this on every row
// append; Virtuoso owns scroll position from that point on.
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [virtuosoRows.length > 0]);
const jumpToMessage = useCallback( const jumpToMessage = useCallback(
(targetId: string) => { (targetId: string) => {
@@ -509,11 +517,7 @@ export function ConversationPage() {
// resolve the target row. Without this, the scroll either no-ops or // resolve the target row. Without this, the scroll either no-ops or
// lands on a stale row. // lands on a stale row.
requestAnimationFrame(() => { requestAnimationFrame(() => {
virtuosoRef.current?.scrollToIndex({ listRef.current?.scrollToRow(rowIndex, 'center', 'smooth');
index: rowIndex,
align: 'center',
behavior: 'smooth',
});
}); });
setHighlightedId(targetId); setHighlightedId(targetId);
window.setTimeout( window.setTimeout(
@@ -709,23 +713,26 @@ export function ConversationPage() {
const handleRangeChanged = useCallback( const handleRangeChanged = useCallback(
(range: { startIndex: number; endIndex: number }) => { (range: { startIndex: number; endIndex: number }) => {
topmostIndexRef.current = range.startIndex; topmostIndexRef.current = range.startIndex;
if (id) { if (!id) return;
if (stickToBottom) {
// Pinned to the bottom: the topmost-visible row drifts as the
// virtualizer mounts/unmounts rows. Persisting it would later reopen the
// chat at that arbitrary row (the second writer behind the scrollTop=0
// bug). Keep the last real reading position and only refresh the flag.
const prev = scrollPositions.get(id); const prev = scrollPositions.get(id);
scrollPositions.set(id, { scrollPositions.set(id, {
topmostIndex: range.startIndex, topmostIndex: prev?.topmostIndex ?? range.startIndex,
stickToBottom: prev?.stickToBottom ?? true, stickToBottom: true,
}); });
} else {
scrollPositions.set(id, { topmostIndex: range.startIndex, stickToBottom: false });
} }
}, },
[id], [id, stickToBottom],
); );
const jumpToBottom = useCallback(() => { const jumpToBottom = useCallback(() => {
virtuosoRef.current?.scrollToIndex({ listRef.current?.scrollToBottom('smooth');
index: 'LAST',
align: 'end',
behavior: 'smooth',
});
setStickToBottom(true); setStickToBottom(true);
setNewMessagesWhileAway(0); setNewMessagesWhileAway(0);
}, []); }, []);
@@ -745,11 +752,7 @@ export function ConversationPage() {
const lastPendingCountRef = useRef(pending.length); const lastPendingCountRef = useRef(pending.length);
useEffect(() => { useEffect(() => {
if (pending.length > lastPendingCountRef.current) { if (pending.length > lastPendingCountRef.current) {
virtuosoRef.current?.scrollToIndex({ listRef.current?.scrollToBottom('auto');
index: 'LAST',
align: 'end',
behavior: 'auto',
});
} }
lastPendingCountRef.current = pending.length; lastPendingCountRef.current = pending.length;
}, [pending.length]); }, [pending.length]);
@@ -764,11 +767,7 @@ export function ConversationPage() {
// targets the correct bottom edge. // targets the correct bottom edge.
const snapToBottom = useCallback(() => { const snapToBottom = useCallback(() => {
window.requestAnimationFrame(() => { window.requestAnimationFrame(() => {
virtuosoRef.current?.scrollToIndex({ listRef.current?.scrollToBottom('auto');
index: 'LAST',
align: 'end',
behavior: 'auto',
});
}); });
}, []); }, []);
@@ -1053,49 +1052,24 @@ export function ConversationPage() {
/> />
</div> </div>
) : ( ) : (
<Virtuoso <MessageList
ref={virtuosoRef} ref={listRef}
className="flex-1" rows={virtuosoRows}
style={{ height: '100%' }} // Deferred reveal: the list stays hidden until messages + reactions
data={virtuosoRows} // + the unread divider are loaded, then anchors and reveals — so the
computeItemKey={(_idx, row) => row.key} // post-paint height cascade is never visible (no chat-switch flicker).
// Initial position: either restored from per-conv memory, or ready={listReady}
// pinned to the bottom for fresh entry. Virtuoso applies this computeKey={(row) => row.key}
// synchronously before its first paint so the user doesn't see initialAnchor={initialAnchor}
// a "loaded at top, then jumped" flicker (matches the layout- // 250px at-bottom tolerance — a tall appended row (image, voice note,
// effect behavior we used in the non-virtualized version). // grouped attachments) shouldn't push the user out of the at-bottom zone.
initialTopMostItemIndex={initialTopMostIndex}
// followOutput auto-scrolls only when the user was already at
// the bottom; returning `false` from the callback when they're
// scrolled up preserves their reading position when realtime
// messages arrive (critical UX: do NOT jerk the user).
//
// We deliberately use 'auto' (instant) rather than 'smooth':
// with a smooth scroll animation, atBottomStateChange fires
// `false` mid-animation (scrollTop is briefly above the new
// bottom) and then `true` after settle — that flips
// stickToBottom twice, flashing the "Zum neuesten" pill and
// re-rendering the whole list. Instant scroll has zero
// mid-animation state so the cascade never happens.
followOutput={(isAtBottom) => (isAtBottom ? 'auto' : false)}
atBottomStateChange={handleAtBottomStateChange}
// 250 px tolerance — large enough that appending a tall row
// (image, voice note, grouped attachments) doesn't push the
// user out of the at-bottom zone. The previous 80 px flipped
// stickToBottom on nearly every typical message arrival.
atBottomThreshold={250} atBottomThreshold={250}
rangeChanged={handleRangeChanged} onReachTop={handleStartReached}
startReached={handleStartReached} onAtBottomChange={handleAtBottomStateChange}
// Render rows just outside the viewport so fast scrolling onTopRowChange={(topIndex) =>
// doesn't briefly flash empty space. handleRangeChanged({ startIndex: topIndex, endIndex: topIndex })
increaseViewportBy={400} }
// Visual breathing space below the last message so a bubble renderRow={(_index, row) => {
// bottom doesn't sit flush against the composer top — matches
// Discord's chat-pane bottom padding.
components={{
Footer: () => <div style={{ height: '12px' }} />,
}}
itemContent={(_index, row) => {
if (row.kind === 'loader') { if (row.kind === 'loader') {
return ( return (
<div className="flex items-center justify-center py-2 text-xs text-fg-muted"> <div className="flex items-center justify-center py-2 text-xs text-fg-muted">
+733
View File
@@ -0,0 +1,733 @@
# Migrations-Runbook: Self-Hosted Backend von `*.netralax.cloud` auf `*.netralax.de` (neuer VPS)
> **Zweck:** Vollständiger Umzug des selbstgehosteten Chat-Backends (Supabase + LiveKit/coturn + Update-Host) vom ALTEN VPS (`46.225.156.249`, `*.netralax.cloud`) auf einen FRISCHEN, leeren NEUEN VPS, der danach `*.netralax.de` UND während der Übergangsphase weiterhin `*.netralax.cloud` ausliefert.
>
> **Lesbar als:** Copy-paste-Runbook. Überschriften und Erklärungen sind deutsch; alle Befehle, Pfade, Variablennamen und Konfig-Snippets bleiben wörtlich/literal.
---
## ⚠️ Zwei nicht verhandelbare Kontinuitäts-Garantien (vor allem anderen lesen)
Die bereits installierten Desktop- (Vite/electron) und Mobile- (Expo) Clients tragen die ALTEN Hostnamen **und** den anon-JWT **fest im Bundle einkompiliert** (`SUPABASE_URL`, `LIVEKIT_URL`, `SUPABASE_ANON_KEY`, `VITE_VAPID_PUBLIC_KEY`, Update-Host). Daraus folgen zwei Garantien, deren Verletzung **alle bestehenden Installationen sofort und lautlos zerstört**:
1. **`JWT_SECRET` (und damit `ANON_KEY`, `SERVICE_ROLE_KEY`) MÜSSEN byte-für-byte vom ALTEN Server übernommen werden.** Der anon-JWT in den Bundles ist mit dem alten `JWT_SECRET` signiert. Ein anderes Secret → Kong/PostgREST/GoTrue verwerfen **jedes** Token → **alle** Sessions fallen aus, niemand kann sich mehr anmelden. Es gibt keine Fehlermeldung, die das offensichtlich macht.
2. **Das VAPID-Schlüsselpaar (`VAPID_PUBLIC_KEY` + `VAPID_PRIVATE_KEY`) MUSS identisch übernommen werden.** Bestehende Web-Push-Subscriptions sind an den öffentlichen VAPID-Key gebunden. Ändert er sich, brechen **alle** vorhandenen Push-Abos Benachrichtigungen verstummen lautlos.
Zusätzlich: Der NEUE Caddy **muss die Legacy-Vhosts `*.netralax.cloud` mitbedienen** und die `.cloud`-DNS-A-Records müssen auf die NEUE IP zeigen, sonst sterben alte Clients in dem Moment, in dem der alte VPS abgeschaltet wird.
---
## 0. Voraussetzungen & Übersicht
### 0.1 Architektur (unverändert auf beiden Servern)
| Komponente | Verzeichnis | Intern | Öffentlich (neu) | Öffentlich (Legacy, weiter bedient) |
|---|---|---|---|---|
| Supabase (Postgres 17, GoTrue, PostgREST, Realtime, Storage, Kong, edge-runtime, Mailpit) | `/opt/supabase` | Kong `127.0.0.1:8000` | `supabase.netralax.de` | `supabase.netralax.cloud` |
| LiveKit SFU (Signaling-WS) | `/opt/livekit` | `127.0.0.1:7880` | `livekit.netralax.de` | `livekit.netralax.cloud` |
| coturn (TURN/TURNS) | `/opt/livekit` | `:3478`, `:5349` (TLS) | `turn.netralax.de:5349` | `turn.netralax.cloud:5349` |
| Update-Host (electron-updater) | `/var/www/updates/windows` | `file_server` | `update.netralax.de` | `update.netralax.cloud` |
TLS-Terminierung für Supabase/LiveKit/Update via **Caddy** (automatisches Let's Encrypt). **TURNS auf `5349` läuft NICHT über Caddy** und braucht ein eigenes Zertifikat auf der Platte.
### 0.2 Was du brauchst
- SSH-Zugang: User `prox` auf dem **alten** (.cloud) VPS, User `debian` auf dem **neuen** (.de) VPS. Der neue VPS ist leer.
- Die NEUE öffentliche IP des `.de`-VPS: **`141.95.34.204`** (bereits in `scripts/migrate/config.sh``NEW_HOST` und `scripts/prod/config.sh``PROD_SERVER` eingetragen). Login-User: `debian`.
- Lese-Zugriff auf die ALTE `/opt/supabase/.env` (enthält alle zu kopierenden Secrets).
- DNS-Verwaltung für `netralax.de` **und** `netralax.cloud`.
- Entwickler-Laptop mit Bash (Linux/macOS/WSL), `ssh`, `rsync`, `openssl`.
- Ein Wartungsfenster (Schreibstopp auf der App), siehe Abschnitt 5.
### 0.3 Reihenfolge der Arbeit (Überblick)
```
1. DNS vorbereiten (niedrige TTL setzen, noch NICHT umbiegen)
2. Neuen VPS bootstrappen → scripts/migrate/01-bootstrap-new-server.sh
3. Secrets 1:1 in /opt/supabase/.env übernehmen (JWT_SECRET/VAPID identisch!)
4. Stacks LEER hochfahren (init der Rollen)
5. Wartungsfenster: DB + Storage migrieren → scripts/migrate/02-migrate-data.sh
6. LiveKit/coturn Prod-Config + Firewall + TURNS-Zertifikat
7. Caddy mit BEIDEN Domain-Sätzen (.de + .cloud)
8. Edge-Functions + deren Secrets deployen
9. Update-Host migrieren + Dual-Publish (.de UND .cloud)
10. Cutover: DNS scharf schalten (alle .de + Repoint aller .cloud)
11. Smoke-Tests (inkl. ALTER .cloud-Client)
12. Repo-Edits + neues Desktop-/Mobile-Release ausliefern
13. Rollback-Plan (bereithalten)
14. Aufräumen / .cloud später abschalten
```
### 0.4 Konventionen der Migrate-Skripte (Interface-Contract)
Alle `scripts/migrate/*`-Skripte sourcen `scripts/migrate/config.sh`. Dieses kennt **beide** Hosts und ist bewusst **unabhängig** von `scripts/prod/config.sh` (das bereits auf den End-Zustand `.de` zeigt). `config.sh` exportiert:
```bash
OLD_HOST="46.225.156.249"
OLD_USER="prox"
NEW_HOST="141.95.34.204"
NEW_USER="debian" # neuer .de-Server: User debian (alt: prox)
SUPABASE_DIR="/opt/supabase"
LIVEKIT_DIR="/opt/livekit"
OLD_SSH="${OLD_USER}@${OLD_HOST}"
NEW_SSH="${NEW_USER}@${NEW_HOST}"
SSH_OPTS="-o StrictHostKeyChecking=accept-new"
```
…und die Helfer `old_remote()` / `new_remote()`, die per `ssh ${SSH_OPTS}` zum jeweiligen Host verbinden.
---
## 1. DNS-Plan
> **Wichtig:** In diesem Schritt wird DNS **noch nicht** umgebogen (außer der TTL-Absenkung). Das eigentliche Scharfschalten passiert erst im **Cutover (Abschnitt 10)**, wenn der neue VPS vollständig steht und getestet ist.
### 1.1 Jetzt (Vorbereitung): TTL absenken
Setze auf **allen** unten genannten A-Records die TTL auf **300 Sekunden (5 min)**, mindestens 2448 h vor dem geplanten Cutover. So wird die spätere Umstellung schnell wirksam.
### 1.2 Beim Cutover (Abschnitt 10): A-Records auf `141.95.34.204`
**Neue `.de`-Records (anlegen):**
| Record | Typ | Ziel |
|---|---|---|
| `supabase.netralax.de` | A | `141.95.34.204` |
| `livekit.netralax.de` | A | `141.95.34.204` |
| `turn.netralax.de` | A | `141.95.34.204` |
| `update.netralax.de` | A | `141.95.34.204` |
**Legacy `.cloud`-Records (REPOINT von alter IP `46.225.156.249` auf neue IP):**
| Record | Typ | Neues Ziel |
|---|---|---|
| `supabase.netralax.cloud` | A | `141.95.34.204` |
| `livekit.netralax.cloud` | A | `141.95.34.204` |
| `turn.netralax.cloud` | A | `141.95.34.204` |
| `update.netralax.cloud` | A | `141.95.34.204` |
> **⚠️ Den Repoint der `.cloud`-Records NICHT vergessen.** Alle bereits installierten Clients sprechen `*.netralax.cloud` an. Bleiben diese Records auf der alten IP, brechen sämtliche Installationen, sobald der alte VPS abgeschaltet wird. Der neue Caddy bedient die `.cloud`-Vhosts mit (Abschnitt 7), und Let's Encrypt stellt für `.cloud` erst dann gültige Zertifikate aus, **wenn** die `.cloud`-A-Records auf die neue IP zeigen.
### 1.3 Verifikation nach dem Cutover
```bash
for h in supabase livekit turn update; do
echo "== $h.netralax.de =="; dig +short $h.netralax.de
echo "== $h.netralax.cloud =="; dig +short $h.netralax.cloud
done
```
Alle acht müssen `141.95.34.204` zurückgeben.
---
## 2. Neuen Server bootstrappen
Das Skript **`scripts/migrate/01-bootstrap-new-server.sh`** wird **auf den neuen VPS kopiert und dort als root** ausgeführt. Es ist idempotent, erfindet **keine** Secrets und gibt am Ende klare NEXT-STEP-Hinweise.
### 2.1 Skript übertragen und ausführen
```bash
# Vom Laptop aus:
scp -o StrictHostKeyChecking=accept-new \
scripts/migrate/01-bootstrap-new-server.sh \
debian@141.95.34.204:/tmp/
ssh -o StrictHostKeyChecking=accept-new debian@141.95.34.204 \
'sudo bash /tmp/01-bootstrap-new-server.sh'
```
### 2.2 Was das Bootstrap-Skript tut
- Installiert **Docker Engine + compose-plugin**.
- Installiert + aktiviert **ufw** und öffnet die Ports (siehe Abschnitt 6 für die vollständige Liste): `22/tcp`, `80/tcp`, `443/tcp`, `7880/tcp`, `7881/tcp`, `50000:50100/udp`, `3478/tcp`, `3478/udp`, `5349/tcp`, `50200:50300/udp`.
- Klont `https://github.com/supabase/supabase` und kopiert `supabase/docker/` nach **`/opt/supabase`** (inkl. `docker-compose.yml`, `volumes/`, `.env.example`). Hinweis: Wir vendoren die Supabase-Compose-Datei **nicht** im Repo sie wird beim Bootstrap frisch geklont.
- Legt **`/opt/livekit`** an und schreibt Platzhalter `docker-compose.yml` + `livekit.yaml` + `coturn.conf`.
- Installiert **Caddy** und legt eine Platzhalter-`/etc/caddy/Caddyfile` an.
- Legt **`/var/www/updates/windows`** an (Artefakt-Verzeichnis; Caddy-Docroot ist das Eltern-Verzeichnis `/var/www/updates`, siehe §7).
- Setzt in `/opt/supabase/.env` die sicherheitskritischen Secrets (`JWT_SECRET`, `ANON_KEY`, `SERVICE_ROLE_KEY`, `POSTGRES_PASSWORD`, …) auf den Sentinel `__COPY_FROM_OLD_SERVER__`, damit ein vergessener Wert **laut scheitert** statt still die öffentlich bekannten Upstream-Defaults zu benutzen.
- Druckt am Ende die NEXT-STEPS: `/opt/supabase/.env` befüllen (Abschnitt 3), Prod-Compose + `livekit.yaml` + `coturn.conf` einsetzen (Abschnitt 6), `Caddyfile` einsetzen (Abschnitt 7).
> **Das Bootstrap-Skript erfindet KEINE Secrets.** Die sicherheitskritischen Keys stehen danach auf dem Sentinel `__COPY_FROM_OLD_SERVER__` (fail-loud); Custom-Secrets wie `VAPID_*` / `PUSH_FANOUT_SHARED_SECRET` sind im Upstream-`.env` gar nicht vorhanden und müssen ergänzt werden. Alle echten Werte kommen in Abschnitt 3 vom alten Server.
---
## 3. Secrets 1:1 übernehmen
Alle Server-Secrets leben auf dem Server in **`/opt/supabase/.env`**. Hole zuerst die ALTE Datei:
```bash
# ALTE .env lokal sichern (nur lesend, nichts ändern):
ssh -o StrictHostKeyChecking=accept-new prox@46.225.156.249 \
'cat /opt/supabase/.env' > old.env.backup
chmod 600 old.env.backup
```
### 3.1 Entscheidungstabelle: identisch kopieren vs. auf neuen Host umstellen
**Spalte „Aktion": `IDENTISCH` = byte-für-byte aus `old.env.backup` übernehmen; `NEU` = auf den neuen Host/Wert setzen.**
| Variable | Aktion | Woher / Neuer Wert | Begründung |
|---|---|---|---|
| `POSTGRES_PASSWORD` | **IDENTISCH** | old.env | Dump trägt Rollen-Passwort-Hashes; muss vor Restore passen, sonst können interne Dienste sich nicht an Postgres anmelden. |
| `JWT_SECRET` | **🔴 IDENTISCH** | old.env | **Signiert die eingebackenen anon/service-role-JWTs. Abweichung = alle Sessions tot.** |
| `ANON_KEY` | **🔴 IDENTISCH** | old.env | Eingebackener anon-JWT der Clients. |
| `SERVICE_ROLE_KEY` | **IDENTISCH** | old.env | service-role-JWT für Edge-Functions/Admin-Skripte; muss zu `JWT_SECRET` passen. |
| `SECRET_KEY_BASE` | **IDENTISCH** | old.env | Realtime (Phoenix) + Vault: signiert Channel-Tokens/Cookies. |
| `VAULT_ENC_KEY` | **IDENTISCH** | old.env | Entschlüsselt vault/pgsodium-verschlüsselte Zeilen aus dem Dump. |
| `PG_META_CRYPTO_KEY` | **IDENTISCH** | old.env | postgres-meta-Crypto-Key; stabil halten. |
| `SMTP_HOST` | **IDENTISCH** | old.env | Magic-Link-Mailversand erhalten (externes Relay / Mailpit). |
| `SMTP_PORT` | **IDENTISCH** | old.env | s.o. |
| `SMTP_USER` | **IDENTISCH** | old.env | s.o. |
| `SMTP_PASS` | **IDENTISCH** | old.env | s.o. |
| `SMTP_ADMIN_EMAIL` | **IDENTISCH** | old.env | Absender/SPF-Konsistenz. |
| `SMTP_SENDER_NAME` | **IDENTISCH** | old.env | Anzeigename konsistent. |
| `FUNCTIONS_VERIFY_JWT` | **IDENTISCH** | old.env (`false`) | `notify-push` nutzt Shared-Secret-Header statt User-JWT; bleibt `false`. |
| `LIVEKIT_API_KEY` | **IDENTISCH** | old.env | Muss = `keys:`-Block in `livekit.prod.yaml`, sonst SFU-Reject (403). |
| `LIVEKIT_API_SECRET` | **IDENTISCH** | old.env | s.o. |
| `VAPID_PUBLIC_KEY` | **🔴 IDENTISCH** | old.env | **Bindet bestehende Push-Abos. Abweichung = alle Push-Subscriptions tot.** |
| `VAPID_PRIVATE_KEY` | **🔴 IDENTISCH** | old.env | Muss mit unverändertem Public-Key paaren. |
| `VAPID_SUBJECT` | **IDENTISCH** | old.env | Konsistenz (mailto/URL). |
| `PUSH_FANOUT_SHARED_SECRET` | **IDENTISCH** | old.env | `x-shared-secret`-Header zwischen DB-Trigger und `notify-push`. |
| `SUPABASE_SERVICE_ROLE_KEY` | **IDENTISCH** | = `SERVICE_ROLE_KEY` | Edge-Function-Alias. |
| `SUPABASE_ANON_KEY` | **IDENTISCH** | = `ANON_KEY` | Edge-Function-Alias (mint-livekit-token RLS-Client). |
| `SITE_URL` | **NEU** | `https://supabase.netralax.de` | GoTrue-Basis-URL für Magic-Link-Redirects. |
| `API_EXTERNAL_URL` | **NEU** | `https://supabase.netralax.de` | Öffentliche Kong-URL, die GoTrue/Studio bewerben. |
| `SUPABASE_PUBLIC_URL` | **NEU** | `https://supabase.netralax.de` | Studio/Kong-Asset-/Link-Generierung. |
| `ADDITIONAL_REDIRECT_URLS` | **NEU** (Superset) | siehe 3.2 | GoTrue-Redirect-Allow-List inkl. Deep-Link-Schemata. |
| `SUPABASE_URL` (Edge-Function) | **NEU** | `https://supabase.netralax.de` (oder internes Kong) | Funktionen müssen es nur erreichen. |
| `LIVEKIT_URL` (Edge-Function) | **NEU** | `wss://livekit.netralax.de` | wss-URL für neue Builds; alte Clients nutzen `.cloud` (vom neuen Caddy mitbedient). |
| `DASHBOARD_USERNAME` | **NEU** | frei wählbar | Studio-Basic-Auth; nicht client-kritisch. |
| `DASHBOARD_PASSWORD` | **NEU** | starkes neues Passwort | s.o. |
| `POSTGRES_HOST` | Default | `db` | nicht host-spezifisch. |
| `POSTGRES_DB` | Default | `postgres` | s.o. |
| `POSTGRES_PORT` | Default | `5432` (nur an localhost gebunden) | s.o. |
| `KONG_HTTP_PORT` | Default | `8000` | muss zum Caddyfile passen. |
| `KONG_HTTPS_PORT` | Default | `8443` (ungenutzt) | Caddy terminiert TLS. |
### 3.2 `ADDITIONAL_REDIRECT_URLS` (exakt, ohne Leerzeichen)
```
ADDITIONAL_REDIRECT_URLS=chatapp://auth/callback,netralax://auth/callback,https://supabase.netralax.de,https://supabase.netralax.cloud
```
> **⚠️ GoTrue lehnt jeden Magic-Link-Redirect ab, der nicht exakt auf der Allow-List steht.** Beide Deep-Link-Schemata (`chatapp://auth/callback` **und** `netralax://auth/callback`) müssen drin sein, sonst scheitert der Native-App-Login.
### 3.3 Werte übertragen
Bearbeite `/opt/supabase/.env` auf dem neuen Server und setze die `IDENTISCH`-Werte aus `old.env.backup`, die `NEU`-Werte aus der Tabelle:
```bash
ssh debian@141.95.34.204 'sudo nano /opt/supabase/.env'
```
> **Reihenfolge-Falle:** `JWT_SECRET`, `ANON_KEY`, `SERVICE_ROLE_KEY`, `POSTGRES_PASSWORD` und das VAPID-Paar müssen in der `.env` stehen, **bevor** in Abschnitt 4 der Stack hochfährt und **bevor** in Abschnitt 5 der Restore läuft. Setze sie jetzt vollständig.
### 3.4 Verifikation (Hashes vergleichen, nicht Klartext loggen)
```bash
# Stelle sicher, dass die kritischen Secrets identisch sind:
for v in JWT_SECRET ANON_KEY SERVICE_ROLE_KEY POSTGRES_PASSWORD VAPID_PUBLIC_KEY VAPID_PRIVATE_KEY; do
old=$(ssh prox@46.225.156.249 "grep -E \"^${v}=\" /opt/supabase/.env | cut -d= -f2-" | sha256sum)
new=$(ssh debian@141.95.34.204 "grep -E \"^${v}=\" /opt/supabase/.env | cut -d= -f2-" | sha256sum)
[ "$old" = "$new" ] && echo "OK $v" || echo "DIFF $v <-- FIX BEFORE RESTORE"
done
```
Jede Zeile muss `OK` sein.
---
## 4. Stacks leer hochfahren
Bevor Daten restauriert werden, muss der frische Supabase-Stack **einmal** hochfahren, damit die Init-Skripte die Rollen anlegen (`supabase_admin`, `authenticator`, `anon`, `authenticated`, `service_role`, `supabase_auth_admin`, `supabase_storage_admin`, …), Extensions und Grants. Voraussetzung: `POSTGRES_PASSWORD` und `JWT_SECRET` sind bereits identisch gesetzt (Abschnitt 3).
```bash
# DB-Container zuerst hochfahren (legt Rollen + Extensions an):
ssh debian@141.95.34.204 \
'cd /opt/supabase && docker compose up -d db && sleep 20'
# Health-Check:
ssh debian@141.95.34.204 \
'cd /opt/supabase && docker compose exec -T db pg_isready -U postgres'
```
> Den **vollständigen** Stack (`docker compose up -d`) fahren wir erst **nach** dem Daten-Restore hoch (Abschnitt 5, Schritt 3), damit alle Dienste gegen die wiederbefüllte DB neu verbinden.
---
## 5. Datenmigration: DB + Storage
Genutzt wird **`scripts/migrate/02-migrate-data.sh`** (läuft vom Laptop, sourct `config.sh`, `set -euo pipefail`, jeder destruktive Schritt ist abgesichert).
### 5.1 🔴 Wartungsfenster: Schreibstopp ZUERST
> **Friere Schreibvorgänge ein, bevor du dumpst und bevor du Storage rsyncst.** Sonst werden DB-Zeilen und Storage-Volume inkonsistent (Objekte auf der Platte ohne Metadaten-Zeile oder umgekehrt). Setze die App in Wartungsmodus / stoppe neue Uploads/Nachrichten auf dem ALTEN System.
Pre-Flight (beide Stacks gesund):
```bash
ssh prox@46.225.156.249 'cd /opt/supabase && docker compose exec -T db pg_isready -U postgres'
ssh debian@141.95.34.204 'cd /opt/supabase && docker compose exec -T db pg_isready -U postgres'
```
### 5.2 Postgres (Major-Version 17) `pg_dumpall`, gestreamt ALT → NEU
Faithful Full-Cluster-Dump (Rollen **inkl. Passwort-Hashes** + alle DBs + auth/storage/realtime/public-Schemata), direkt vom alten in den neuen Container gestreamt:
```bash
old_remote 'cd /opt/supabase && docker compose exec -T db pg_dumpall -U postgres --clean --if-exists' \
| new_remote 'cd /opt/supabase && docker compose exec -T db psql -U postgres -d postgres -v ON_ERROR_STOP=0'
```
Wichtige Hinweise zu diesem Befehl:
- **`pg_dumpall` (nicht `pg_dump`)** ist nötig, weil es die ROLLEN-Definitionen samt Passwort-Hashes (md5/scram) mitnimmt. Da `POSTGRES_PASSWORD` auf beiden Hosts identisch ist, passen die restaurierten Rollen-Passwörter zu dem, was die Dienste benutzen.
- **`--clean --if-exists`** macht den Dump gegen den bereits initialisierten Cluster wiederholbar (droppt/erzeugt Objekte neu).
- **`ON_ERROR_STOP=0` (nicht `=1`):** `pg_dumpall` versucht, bereits existierende Rollen wie `supabase_admin`/`postgres` per `CREATE ROLE` anzulegen → harmlose „already exists"-Fehler. Mit `ON_ERROR_STOP=1` würde der erste davon einen guten Restore abbrechen. `=0` schluckt aber **auch echte Fehler** (FK/Constraint/Ownership) und hinterlässt eine teil-restaurierte DB, die „erfolgreich" aussieht. **Deshalb scannt `02-migrate-data.sh` den Restore automatisch:** es teet die Ausgabe in ein Log, grept nach `ERROR/FATAL/PANIC` abzüglich der harmlosen Muster und **bricht VOR dem Storage-rsync ab**, falls echte Fehler übrig bleiben (bewusster Override: `FORCE_RESTORE_OK=1`).
- Erfasst in einem Rutsch **alle** Schemata: `auth` (User/Identities/Sessions), `storage` (Buckets + Objekt-Metadaten), `realtime` (Tenants/Subscriptions), `public` (App-Tabellen), ggf. `_realtime`/`_analytics`.
> **🔴 Migrationen NICHT erneut anwenden.** Alle Migrationen stecken bereits im Dump. **`scripts/prod/push-migrations.sh` nach dem Restore NICHT ausführen** das riskiert Drift/Duplicate-Object-Fehler.
**Alternative (nur falls Cluster-Level scheitert):** Single-DB `pg_dump -Fc` + `pg_restore --clean --if-exists --no-owner`, plus separat `pg_dumpall --roles-only`. Der `pg_dumpall`-Pfad oben ist für self-hosted→self-hosted vorzuziehen.
### 5.3 Vollständigen Stack neu hochfahren
```bash
new_remote 'cd /opt/supabase && docker compose down && docker compose up -d'
```
### 5.4 Storage-Objekte `rsync` (ALT → NEU)
Die Objekt-Bytes liegen unter `/opt/supabase/volumes/storage` (Bind-Mount → Container `/var/lib/storage`); die Metadaten-Zeilen kamen bereits mit dem Dump. **Schreibstopp muss noch aktiv sein.** Trailing-Slashes beachten:
```bash
# Direkt ALT -> NEU (Daten fließen Server-zu-Server, wenn alt den neuen erreicht):
old_remote "sudo rsync -aHAX --numeric-ids --delete \
-e 'ssh -o StrictHostKeyChecking=accept-new' \
/opt/supabase/volumes/storage/ ${NEW_USER}@${NEW_HOST}:/opt/supabase/volumes/storage/"
```
Falls die Server sich gegenseitig **nicht** per SSH erreichen, zwei-stufig über den Laptop:
```bash
rsync -aHAX --numeric-ids -e "ssh ${SSH_OPTS}" ${OLD_USER}@${OLD_HOST}:/opt/supabase/volumes/storage/ ./_storage_stage/
rsync -aHAX --numeric-ids --delete -e "ssh ${SSH_OPTS}" ./_storage_stage/ ${NEW_USER}@${NEW_HOST}:/opt/supabase/volumes/storage/
```
- `-aHAX` erhält Hardlinks/ACLs/xattrs; `--delete` macht das Ziel zum exakten Spiegel (**nur sicher bei eingefrorenen Schreibvorgängen**).
Danach Storage-Service neu starten, damit die UID-/Ownership-Erwartung passt:
```bash
new_remote 'cd /opt/supabase && docker compose restart storage imgproxy'
```
### 5.5 Daten-Verifikation
```bash
# Tabellen-/User-Counts vergleichen (Beispiel):
new_remote 'cd /opt/supabase && docker compose exec -T db psql -U postgres -d postgres \
-c "select count(*) as users from auth.users;" \
-c "select count(*) as objects from storage.objects;"'
```
`02-migrate-data.sh` macht zusätzlich eine **Zeilen-Paritätsprüfung OLD vs NEU** über tragende Tabellen (`auth.users`, `auth.identities`, `public.profiles`, `public.messages`, `public.conversation_members`, `storage.objects`) und meldet jede Abweichung — eine reine User-/Objekt-Zählung würde Teilverluste in `messages`/`members` übersehen. Ein bekanntes Objekt sollte zudem über das neue Gateway ladbar sein (Test nach Caddy-Setup, Abschnitt 11).
> **Cold-Volume-Copy-Alternative:** Nur falls Image-Tags byte-identisch sind, kann man statt Logical-Dump **beide** DBs stoppen und `volumes/db/data` (PGDATA) **plus** das `db-config`-Named-Volume (enthält den pgsodium-Key) rsyncen. Nur mit gestoppten DBs und identischen Postgres-Image-Tags; ansonsten den Logical-Dump oben bevorzugen.
---
## 6. LiveKit/coturn Prod-Config + Firewall-Ports + TURNS-Zertifikat
> **Die Prod-Config unterscheidet sich von der Dev-`infra/livekit/livekit.yaml` im Repo.** Prod setzt `rtc.use_external_ip: true` und enthält **KEIN** `node_ip: 127.0.0.1` (das ist Dev-only).
> **🟢 Sicherster Weg — die ALTE, funktionierende Config übernehmen.** Die `.example`-Templates sind eine Referenz; produktiv erprobt ist aber die Config, die auf dem alten Server **bereits läuft**. Hol dir die echten Dateien vom alten VPS und ändere nur das Nötigste — so bleibt insbesondere erhalten, **wie** den Clients die TURN-Server/ICE-Credentials angekündigt werden (das macht der alte `livekit.yaml`-`turn:`/`rtc:`-Block bzw. die coturn-`user=`-Zeile; `mint-livekit-token` liefert nur LiveKit-URL+Token, nicht die TURN-Creds):
> ```bash
> # vom Laptop:
> scp prox@46.225.156.249:/opt/livekit/livekit.yaml ./_livekit_old.yaml
> scp prox@46.225.156.249:/opt/livekit/coturn.conf ./_coturn_old.conf
> # dann NUR anpassen: external-ip (neue IP), cert/pkey-Pfade (turn.netralax.de),
> # und — falls vorhanden — eine externe IP/Domain im livekit.yaml turn-Block.
> # Danach als /opt/livekit/{livekit.yaml,coturn.conf} auf den neuen Server.
> ```
> Wenn die alten Dateien nicht greifbar sind, nutze die Templates unten und stelle sicher, dass die coturn-`user=`-Credentials zu dem passen, was deine Clients heute für TURN verwenden.
### 6.1 Prod-Compose + `livekit.yaml` + `coturn.conf` einsetzen
Auf dem alten Server lief LiveKit/coturn über ein Compose in `/opt/livekit`. Das Repo liefert dafür **`infra/livekit/docker-compose.prod.yml.example`** (die Dev-`infra/livekit/docker-compose.yml` ist **nicht** prod-tauglich: coturn läuft dort mit `--no-tls`, ohne `5349`, ohne Zertifikat). Drei Dateien auf den Server kopieren — die **on-server-Namen** sind bewusst `livekit.yaml` / `coturn.conf` (genau die, die auch `scripts/prod/rotate-livekit-keys.sh` editiert):
| Repo-Template | → on-server |
|---|---|
| `infra/livekit/docker-compose.prod.yml.example` | `/opt/livekit/docker-compose.yml` |
| `infra/livekit/livekit.prod.yaml.example` | `/opt/livekit/livekit.yaml` |
| `infra/livekit/coturn.prod.conf.example` | `/opt/livekit/coturn.conf` |
`keys:`-Block in **`/opt/livekit/livekit.yaml`** mit den Werten aus Abschnitt 3 (`LIVEKIT_API_KEY` / `LIVEKIT_API_SECRET`) füllen:
```yaml
port: 7880
log_level: info
rtc:
tcp_port: 7881
port_range_start: 50000
port_range_end: 50100
use_external_ip: true
# KEIN node_ip: 127.0.0.1 — das ist dev-only und würde alle Remote-Clients
# ihre Medien an den eigenen Loopback schicken lassen (Call ohne Audio/Video).
keys:
__LIVEKIT_API_KEY__: __LIVEKIT_API_SECRET__
turn:
enabled: false # coturn läuft separat
```
> **🔴 `node_ip: 127.0.0.1` aus der Dev-Config NICHT übernehmen.** Sonst verbinden Calls zwar, haben aber **keinen Ton und kein Bild**, weil jeder Remote-Client Medien an seinen eigenen Loopback sendet.
>
> **🔴 `LIVEKIT_API_KEY`/`SECRET` im `keys:`-Block MÜSSEN exakt den Edge-Function-Werten in `/opt/supabase/.env` entsprechen.** Sonst signiert `mint-livekit-token` Tokens, die der SFU mit 403 ablehnt.
### 6.2 coturn Prod-Config einsetzen
Template: **`infra/livekit/coturn.prod.conf.example`** → **`/opt/livekit/coturn.conf`**. Die Zertifikatspfade zeigen auf `/etc/letsencrypt/...` — genau das Verzeichnis, das das Prod-Compose read-only in den coturn-Container einhängt:
```conf
realm=netralax.de
listening-port=3478
tls-listening-port=5349
external-ip=141.95.34.204
min-port=50200
max-port=50300
cert=/etc/letsencrypt/live/turn.netralax.de/fullchain.pem
pkey=/etc/letsencrypt/live/turn.netralax.de/privkey.pem
lt-cred-mech
user=__TURN_USER__:__TURN_PASSWORD__
fingerprint
no-multicast-peers
```
### 6.3 TURNS-Zertifikat für `turn.netralax.de` (NICHT über Caddy)
> **TURNS auf `5349` geht NICHT durch Caddy** coturn braucht ein eigenes TLS-Cert+Key auf der Platte (`cert`/`pkey`-Pfade oben). Ein reines Caddy-Cert deckt das nicht ab.
Zwei Wege, das Zertifikat bereitzustellen:
**A) certbot standalone (empfohlen, einfachster Pfad).** Schreibt direkt nach `/etc/letsencrypt/live/turn.netralax.de/` — also genau die Pfade, die `coturn.conf` referenziert und die das Prod-Compose in den Container einhängt. Kein Kopieren nötig:
```bash
# Port 80 muss kurz frei sein (Caddy ggf. stoppen oder DNS-01 nutzen):
sudo certbot certonly --standalone -d turn.netralax.de
# Renewal-Hook, damit coturn das erneuerte Cert lädt:
sudo certbot renew --deploy-hook 'docker compose -f /opt/livekit/docker-compose.yml restart turn'
```
**B) Caddy-Cert wiederverwenden.** Caddy hat ohnehin ein gültiges Cert für `turn.netralax.de`, sobald der DNS-Record steht und der Host in der Caddy-Config ist. PEM/Key aus Caddys Storage (`/var/lib/caddy/.local/share/caddy/certificates/...`) an die `/etc/letsencrypt/live/turn.netralax.de/`-Pfade symlinken/kopieren und coturn nach Renewals neu starten. Umständlicher als (A) — nur, wenn certbot nicht in Frage kommt.
> coturn liest das Cert **beim Start**; nach jeder Erneuerung den `turn`-Container neu starten (Hook oben). Das `external-ip` muss die **neue** öffentliche IP sein.
### 6.4 Firewall-Ports (ufw) ALLE öffnen, sonst kein A/V
> Diese Ports **umgehen Caddy** und müssen direkt in ufw offen sein. Fehlt einer, haben Calls **keinen Ton/kein Bild**.
```bash
ssh debian@141.95.34.204 'sudo bash -s' <<'EOF'
ufw allow 22/tcp
ufw allow 80/tcp
ufw allow 443/tcp
ufw allow 7880/tcp # LiveKit Signaling (hinter Caddy)
ufw allow 7881/tcp # RTC TCP-Fallback
ufw allow 50000:50100/udp # RTC Media
ufw allow 3478/udp # coturn STUN/TURN
ufw allow 3478/tcp # coturn STUN/TURN
ufw allow 5349/tcp # coturn TURNS (TLS)
ufw allow 50200:50300/udp # coturn TURN-Relay
ufw --force enable
ufw status verbose
EOF
```
> **Postgres NICHT öffentlich öffnen.** `5432` bleibt nur an `localhost` gebunden (wie auf dem alten Server). Für Remote-`psql` das bestehende Tunnel-Muster nutzen: `./scripts/prod/tunnel-db.sh` (SSH-Tunnel `localhost:5433 → server:5432`).
### 6.5 LiveKit-Stack starten
Voraussetzung: `/opt/livekit/docker-compose.yml` ist das **Prod**-Compose aus §6.1 (host-networking, mountet `livekit.yaml` + `coturn.conf` + `/etc/letsencrypt`), nicht das Dev-Compose.
```bash
ssh debian@141.95.34.204 'cd /opt/livekit && docker compose up -d && docker compose ps'
# coturn lauscht jetzt auf 5349/TLS? prüfen:
ssh debian@141.95.34.204 'ss -tlnp | grep -E "5349|3478" ; docker compose -f /opt/livekit/docker-compose.yml logs turn --tail=20'
```
---
## 7. Caddy mit BEIDEN Domain-Sätzen (.de + .cloud Legacy)
Template: **`infra/caddy/Caddyfile`** → auf dem Server `/etc/caddy/Caddyfile`. Caddy terminiert TLS (automatisches Let's Encrypt) und reverse-proxyt Klartext-HTTP an die lokalen Backends. **Pro Vhost genau EIN `reverse_proxy`** Kong multiplext bereits alle Supabase-Routen; keine Pfad-Splits in Caddy.
```caddyfile
# Caddyfile — Dual-Domain-Übergang .cloud -> .de
#
# Während der Migration bedient dieser Caddy BEIDE Domain-Sätze aus denselben
# lokalen Backends:
# - *.netralax.de = neue, primäre Hostnamen (neue Client-Builds)
# - *.netralax.cloud = Legacy-Hostnamen, die in bereits installierten
# Desktop-/Mobile-Bundles fest einkompiliert sind.
# Die .cloud-DNS-A-Records zeigen (nach dem Cutover) auf DIESELBE neue IP, damit
# alte Installationen weiterlaufen, bis sie sich selbst auf .de aktualisieren.
# NICHT entfernen, solange noch alte Clients .cloud ansprechen (siehe Abschnitt 14).
# AKTIV ab jetzt: nur die .de-Hosts. Die .cloud-Blöcke stehen auskommentiert
# darunter und werden ERST beim Cutover (§10) aktiviert — sonst läuft Caddy ins
# Let's-Encrypt-Rate-Limit, weil .cloud-DNS noch auf den alten Server zeigt.
# --- Supabase (Kong-Gateway :8000 multiplext auth/rest/realtime/storage/functions/Studio) ---
# Realtime-WS (/realtime/v1/websocket) wird von reverse_proxy transparent upgegradet.
supabase.netralax.de {
reverse_proxy localhost:8000
}
# --- LiveKit Signaling-WS (:7880). Caddy reicht Upgrade/Connection-Header durch. ---
livekit.netralax.de {
reverse_proxy localhost:7880
}
# --- Update-Host (electron-updater: latest.yml + .exe + changelog.json) ---
# 🔴 docroot ist /var/www/updates, NICHT .../windows: release.mjs lädt nach
# /var/www/updates/windows/ hoch, Clients holen unter URL-Pfad /windows/…
# Mit root=.../windows entstünde /windows/windows/ → 404 für JEDES Update.
update.netralax.de {
root * /var/www/updates
file_server
}
# --- CUTOVER (§10): erst NACH .cloud-DNS-Repoint einkommentieren + caddy reload ---
# supabase.netralax.cloud { reverse_proxy localhost:8000 }
# livekit.netralax.cloud { reverse_proxy localhost:7880 }
# update.netralax.cloud { root * /var/www/updates
# file_server }
```
> **🔴 Pfad-Matcher, die WS-Endpunkte ausschließen, sind tabu.** Caddy v2 reicht WebSocket-Upgrades transparent durch aber nur, wenn der **ganze** Host reverse-proxyt wird (kein Sub-Path-Matching). Das gilt für Realtime (`/realtime/v1/websocket`) **und** LiveKit (`/rtc`). Es gibt kein „websocket"-Flag und es wird keins gebraucht.
Aktivieren:
```bash
ssh debian@141.95.34.204 'sudo caddy validate --config /etc/caddy/Caddyfile && sudo systemctl reload caddy'
```
> Let's Encrypt stellt für die `.cloud`-Namen erst gültige Zertifikate aus, **nachdem** die `.cloud`-A-Records auf die neue IP zeigen (Cutover, Abschnitt 10). Bis dahin schlägt die Cert-Ausstellung für `.cloud` fehl das ist erwartbar und löst sich mit dem DNS-Repoint.
---
## 8. Edge-Functions deployen + Secrets
Edge-Functions liegen im Repo unter `supabase/functions/`: **`mint-livekit-token`**, **`notify-push`**, **`og-preview`**. Deploy via bestehendem Skript (kopiert `supabase/functions/<name>/` nach `/opt/supabase/volumes/functions/<name>/` und startet `functions`-Container neu).
> **Achtung Host-Pinning des Deploy-Skripts:** `scripts/prod/push-edge-function.sh` sourct `scripts/prod/config.sh`, das auf `PROD_SERVER="141.95.34.204"` (neuer `.de`-VPS, User `debian`) zeigt. Diese Befehle pushen also auf den NEUEN Server — erst ausführen, nachdem Bootstrap + Secrets dort stehen:
```bash
./scripts/prod/push-edge-function.sh mint-livekit-token
./scripts/prod/push-edge-function.sh notify-push
./scripts/prod/push-edge-function.sh og-preview
```
### 8.1 Erwartete Edge-Function-Secrets in `/opt/supabase/.env`
Aus dem Code verifiziert; alle in `/opt/supabase/.env` (in Abschnitt 3 bereits gesetzt):
`LIVEKIT_API_KEY`, `LIVEKIT_API_SECRET`, `LIVEKIT_URL`, `VAPID_PUBLIC_KEY`, `VAPID_PRIVATE_KEY`, `VAPID_SUBJECT`, `PUSH_FANOUT_SHARED_SECRET`, `SUPABASE_URL`, `SUPABASE_SERVICE_ROLE_KEY`, `SUPABASE_ANON_KEY`.
Erinnerung: `FUNCTIONS_VERIFY_JWT=false` lassen (notify-push gatet über `x-shared-secret`-Header, nicht über User-JWT).
> **🔴 Custom-Secrets müssen den `functions`-Container auch erreichen.** Im **frisch geklonten** Supabase-Compose bekommt der `functions`-Service nur die env-Variablen, die in seinem `environment:`/`env_file:`-Block stehen. `LIVEKIT_API_KEY/SECRET`, `VAPID_*`, `PUSH_FANOUT_SHARED_SECRET` und `SUPABASE_ANON_KEY` sind **Custom-Variablen** und stehen dort per Default **nicht** drin. Auf dem alten Server ist das verdrahtet (es läuft ja) — auf dem neuen muss es nachgezogen werden: entweder `env_file: .env` am `functions`-Service ergänzen oder die Variablen explizit in dessen `environment:` listen. Sonst sieht `mint-livekit-token` leere Strings → `livekit-not-configured` (500) und `notify-push` lehnt mangels `SHARED_SECRET` jede Anfrage ab.
### 8.2 Verifikation
```bash
# 1) Erreichen die Secrets den Container wirklich? (vor dem Funktionstest!)
ssh debian@141.95.34.204 'cd /opt/supabase && docker compose exec -T functions \
env | grep -E "LIVEKIT_API_KEY|LIVEKIT_API_SECRET|VAPID_PUBLIC_KEY|PUSH_FANOUT_SHARED_SECRET|SUPABASE_ANON_KEY"'
# -> Es müssen NICHT-leere Werte erscheinen. Fehlt einer: env_file/environment im
# functions-Service nachziehen und 'docker compose up -d functions'.
# 2) Logs:
./scripts/prod/logs.sh # bzw. docker compose logs functions --tail=20
# 403 bei mint-livekit-token? -> LIVEKIT_API_KEY/SECRET stimmen nicht mit /opt/livekit/livekit.yaml überein.
```
---
## 9. Update-Host migrieren + Dual-Publish (.de UND .cloud)
Der Update-Host ist ein statisches Verzeichnis `/var/www/updates/windows` mit `latest.yml`, `.exe`-Installern und `changelog.json`, ausgeliefert per `file_server` (Abschnitt 7). SSH-Deploy-User: `chatapp-deploy`.
### 9.1 Bestehende Artefakte ALT → NEU spiegeln
```bash
rsync -aHAX --numeric-ids -e "ssh ${SSH_OPTS}" \
chatapp-deploy@46.225.156.249:/var/www/updates/windows/ ./_updates_stage/
rsync -aHAX --numeric-ids -e "ssh ${SSH_OPTS}" \
./_updates_stage/ chatapp-deploy@141.95.34.204:/var/www/updates/windows/
```
### 9.2 Dual-Publish-Garantie
Beide Hosts (`update.netralax.de` und ab Cutover `update.netralax.cloud`) haben im Caddyfile denselben docroot **`/var/www/updates`** (nicht `…/windows`). Die Artefakte liegen physisch in `/var/www/updates/windows/` und werden so unter dem URL-Pfad `/windows/latest.yml` usw. ausgeliefert unter **beiden** Hosts aus **einem** Verzeichnis. (Den Docroot-Fallstrick `/windows/windows/` → 404 siehe §7.)
> **🔴 Alte Clients prüfen `update.netralax.cloud`.** Liegt die Switch-over-Release nicht (auch) unter `.cloud`, können alte Installationen sich **niemals** auf `.de` aktualisieren. Der `changelog.ts` der neuen Builds zeigt zwar auf `https://update.netralax.de/windows/changelog.json`, aber die im Bundle der **alten** Clients eingebackene URL ist `.cloud` beide müssen funktionieren.
### 9.3 Deploy-Konfiguration
`.env.release` ist bereits gesetzt (`UPDATE_HOST=update.netralax.de`, `UPDATE_SSH_USER=chatapp-deploy`, `UPDATE_REMOTE_PATH=/var/www/updates/windows`). Stelle sicher, dass der Deploy-User `chatapp-deploy` auf dem neuen VPS existiert und Schreibrechte auf `/var/www/updates/windows` hat.
---
## 10. Cutover & DNS scharf schalten
> **Erst hier wird DNS umgebogen.** Voraussetzung: Abschnitte 29 abgeschlossen, neuer VPS steht, Stacks laufen, Caddy lädt (für `.de` bereits mit gültigem Cert), Storage + DB migriert, Wartungsfenster ggf. noch aktiv.
### 10.1 Reihenfolge
1. **`.de`-A-Records anlegen** (Abschnitt 1.2, neue Records) → Caddy holt sofort Let's-Encrypt-Certs für `.de`.
2. Interner Smoke-Test über `.de` (Abschnitt 11) **bevor** alte Clients umgeschwenkt werden.
3. **`.cloud`-A-Records repointen** auf `141.95.34.204` (Abschnitt 1.2, Legacy-Records) → Caddy stellt jetzt auch für `.cloud` Certs aus; alte Clients landen ab jetzt auf dem neuen VPS.
4. Propagation prüfen (Abschnitt 1.3).
5. **Wartungsmodus aufheben**, Schreibvorgänge auf dem **neuen** System freigeben.
### 10.2 Verifikation der TLS-Ausstellung
```bash
for h in supabase.netralax.de supabase.netralax.cloud livekit.netralax.de livekit.netralax.cloud update.netralax.de update.netralax.cloud; do
echo "== $h =="
echo | openssl s_client -connect "$h:443" -servername "$h" 2>/dev/null | openssl x509 -noout -subject -dates
done
```
Jeder Host muss ein gültiges, nicht abgelaufenes Cert liefern.
---
## 11. Smoke-Test-Checkliste
Nach dem Cutover, in dieser Reihenfolge:
### 11.1 Supabase / Auth / Magic-Link
- [ ] `https://supabase.netralax.de/auth/v1/health` und `https://supabase.netralax.cloud/auth/v1/health` liefern `200`.
- [ ] **Login per Magic-Link, pro Plattform mit dem JEWEILS registrierten Schema** testen: **Desktop** über `chatapp://auth/callback`, **Mobile** über `netralax://auth/callback` (das in `apps/mobile/app.json` registrierte Schema). `ADDITIONAL_REDIRECT_URLS` enthält beide, daher akzeptiert GoTrue beides — aber das OS routet nur das tatsächlich registrierte Schema zurück in die App.
> ⚠️ Vorbestehend (nicht durch den Umzug verursacht): `apps/mobile/.env.local` setzt aktuell `EXPO_PUBLIC_AUTH_REDIRECT_URL=chatapp://auth/callback`, `app.json` registriert aber nur `netralax://`. Für funktionierende Mobile-Magic-Links sollte das App-Team den Mobile-Wert auf `netralax://auth/callback` setzen (Desktop bleibt `chatapp://`). Außerhalb des Server-Umzugs — hier nur als Flag.
- [ ] PostgREST-Zugriff mit dem **eingebackenen** anon-Key wird akzeptiert (kein 401 wegen falschem `JWT_SECRET`):
```bash
curl -s -H "apikey: <ANON_KEY>" "https://supabase.netralax.de/rest/v1/" | head
```
### 11.2 Nachricht senden / Realtime
- [ ] Zwei eingeloggte Clients: Nachricht von A erscheint bei B in Echtzeit (Realtime-WS `/realtime/v1/websocket` über Caddy).
- [ ] Storage: Upload + Re-Download eines Bildes (`/storage/v1/object/...`) funktioniert (DB-Metadaten + Volume-Bytes konsistent).
### 11.3 Voice-Call mit echtem Ton (über TURN)
- [ ] **Call zwischen zwei Geräten in unterschiedlichen Netzen** (mind. eins hinter NAT/CGNAT, das TURN erzwingt): Verbindung steht **und es ist echter Ton/Bild hörbar/sichtbar**.
- [ ] Bestätigt indirekt: `rtc.use_external_ip: true`, **kein** `node_ip: 127.0.0.1`, alle Media-Ports offen, TURNS-Cert für `turn.netralax.de` gültig.
- [ ] `mint-livekit-token` liefert ein Token, das der SFU akzeptiert (kein 403 → Keys stimmen mit `/opt/livekit/livekit.yaml` überein).
### 11.4 Web-Push
- [ ] Ein **bestehender** (vor der Migration angelegter) Push-Abonnent erhält weiterhin Benachrichtigungen → bestätigt identisches VAPID-Paar.
- [ ] Neue Subscription + Test-Push über `notify-push` (mit korrektem `x-shared-secret` / `PUSH_FANOUT_SHARED_SECRET`) kommt an.
### 11.5 Auto-Update-Check von einem ALTEN `.cloud`-Client
- [ ] `latest.yml` ist unter **beiden** Hosts mit echtem `200` abrufbar (nicht nur „erreichbar" — der Docroot-Bug aus §7 würde hier 404 liefern):
```bash
curl -sI https://update.netralax.de/windows/latest.yml | head -1 # HTTP/2 200
curl -sI https://update.netralax.cloud/windows/latest.yml | head -1 # HTTP/2 200
curl -sI https://update.netralax.cloud/windows/changelog.json | head -1
```
- [ ] Eine **bestehende, alte** Desktop-Installation (Hostnamen `.cloud` eingebacken) prüft auf Updates: electron-updater findet die Switch-over-Release, lädt sie und installiert.
- [ ] Nach dem Update zeigt der Client auf `.de` (neue Bundle-Werte) und funktioniert vollständig (Login, Nachricht, Call, Push).
> **Dieser letzte Test ist der wichtigste.** Er beweist den gesamten Übergangspfad: alter Client → `.cloud` (neue IP) → lädt Update → wird zu `.de`-Client.
---
## 12. Repo-Änderungen + neues Release bauen/ausliefern
### 12.1 Bereits gemachte Edits (verifiziert im Repo)
| Datei | Änderung | Status |
|---|---|---|
| `scripts/prod/config.sh` | `PROD_SERVER="141.95.34.204"`, `PROD_DOMAIN_SUPABASE=supabase.netralax.de`, `PROD_DOMAIN_LIVEKIT=livekit.netralax.de` | ✅ erledigt (End-Zustand) |
| `apps/desktop/.env` | `SUPABASE_URL` + `VITE_SUPABASE_URL` = `https://supabase.netralax.de`; `VITE_LIVEKIT_URL=wss://livekit.netralax.de`; anon-Key + `VITE_VAPID_PUBLIC_KEY` (unverändert übernommen) | ✅ erledigt |
| `apps/mobile/.env.local` | `EXPO_PUBLIC_SUPABASE_URL=https://supabase.netralax.de` (anon-Key, redirect-Schema unverändert) | ✅ erledigt |
| `.env.release` | `UPDATE_HOST=update.netralax.de`, `UPDATE_SSH_USER=chatapp-deploy`, `UPDATE_REMOTE_PATH=/var/www/updates/windows` | ✅ erledigt |
| `package.json` | `release`-Script + `prod:*`-Scripts vorhanden (unverändert; nutzen `scripts/prod/config.sh`) | ✅ vorhanden |
| `apps/desktop/src/lib/changelog.ts` | `CHANGELOG_URL='https://update.netralax.de/windows/changelog.json'` (mit Kommentar, dass alte Clients weiter `.cloud` abfragen) | ✅ erledigt |
> **✅ Erledigt:** Die neue IP `141.95.34.204` ist in `scripts/prod/config.sh` (`PROD_SERVER`) und `scripts/migrate/config.sh` (`NEW_HOST`) eingetragen; Login-User dort ist `debian`.
### 12.2 Neues Desktop-Release bauen + dual publizieren
```bash
# Vom Laptop, mit korrektem .env.release:
pnpm install
pnpm --filter @chat-app/desktop build
pnpm release # = node scripts/release.mjs
```
`scripts/release.mjs` lädt `latest.yml` + `.exe` + aktualisiertes `changelog.json` nach `UPDATE_HOST` (`update.netralax.de`). Da Caddy `update.netralax.de` **und** `update.netralax.cloud` aus demselben Verzeichnis bedient, ist diese eine Veröffentlichung **automatisch** unter beiden Hosts verfügbar (Dual-Publish, Abschnitt 9).
> **🔴 Diese Release MUSS unter `.cloud` erreichbar sein**, denn nur sie schaltet alte Installationen auf `.de` um. Nach dem Upload mit Abschnitt 11.5 verifizieren.
### 12.3 Neues Mobile-Release
```bash
pnpm --filter @chat-app/mobile typecheck
# Expo-Build/Submit nach eurem üblichen EAS-/Store-Prozess.
# .env.local trägt bereits EXPO_PUBLIC_SUPABASE_URL=https://supabase.netralax.de.
```
> Mobile-Clients aktualisieren über die App-Stores, nicht über den Update-Host. Bis ein User die neue Store-Version installiert, hält ihn der `.cloud`-Vhost am Leben.
---
## 13. Rollback-Plan
Der alte VPS bleibt **vollständig intakt und laufend**, bis der neue verifiziert ist. Rollback heißt im Kern: **DNS zurückbiegen**.
1. **Schnell-Rollback (DNS):** Alle `.cloud`-A-Records zurück auf `46.225.156.249` (alte IP), `.de`-Records entfernen oder ebenfalls auf alt zeigen lassen. Dank niedriger TTL (Abschnitt 1.1) greift das in Minuten. Alte Clients landen wieder auf dem alten, intakten Server.
2. **Voraussetzung dafür:** Während der Migration **keine destruktiven Änderungen am alten Server** (alter Stack nicht löschen, alte Volumes nicht anfassen). Der Schreibstopp (Abschnitt 5.1) bedeutet nur Wartungsmodus, kein Datenverlust.
3. **Daten-Divergenz beachten:** Wurden nach dem Cutover bereits Schreibvorgänge auf dem **neuen** Server akzeptiert, gehen diese bei einem reinen DNS-Rollback verloren. Deshalb: Cutover (Abschnitt 10.5, Schreibfreigabe) erst nach den Smoke-Tests; bis dahin ist der Rollback verlustfrei.
4. **Update-Host-Rollback:** `.exe`/`latest.yml` auf dem alten Host wurden nicht verändert; alte Clients, die noch nicht aktualisiert haben, finden dort weiterhin den alten Stand.
5. Wenn nur **eine** Komponente klemmt (z. B. nur TURN ohne Ton), kann punktuell zurückgerollt werden, indem nur der betroffene `.cloud`-Record zurückzeigt die übrigen können auf neu bleiben.
---
## 14. Aufräumen / `.cloud` später abschalten
Die `.cloud`-Hosts dürfen **erst** verschwinden, wenn praktisch keine alten Clients mehr darauf zugreifen.
### 14.1 Reihenfolge der Abschaltung (frühestens → spätestens)
1. **Alten VPS dekommissionieren:** Erst nachdem `.cloud`-DNS auf den **neuen** VPS repointet ist und über die neue IP läuft. (Der alte Server liefert dann ohnehin keinen Traffic mehr.) Vorher als Rollback-Sicherheit behalten (Abschnitt 13).
2. **Supabase-/LiveKit-`.cloud`-Vhosts in Caddy** entfernen, sobald Telemetrie/Logs zeigen, dass praktisch alle aktiven Sessions auf `.de` laufen (d. h. die meisten Desktop-Clients haben die Switch-over-Release gezogen und Mobile-Clients die neue Store-Version).
3. **Update-`.cloud`-Vhost als LETZTES abschalten.**
### 14.2 Warum der Update-Host am längsten bleiben muss
> Eine Desktop-Installation, die **noch nie** die Switch-over-Release gezogen hat, kennt **nur** `update.netralax.cloud` (eingebacken). Sie erreicht `.de` ausschließlich, indem sie die neue Version über **`.cloud`** herunterlädt. Schaltest du `update.netralax.cloud` zu früh ab, **stranden** alle noch nicht aktualisierten Clients dauerhaft auf der alten Version sie können sich nie mehr selbst auf `.de` updaten und müssten manuell neu installiert werden.
>
> Faustregel: `update.netralax.cloud` so lange behalten, bis die Update-Metriken zeigen, dass der Long-Tail alter Installationen vernachlässigbar ist (eher Monate als Wochen). Supabase-/LiveKit-`.cloud` können früher fallen als Update-`.cloud`, aber niemals umgekehrt.
### 14.3 Endzustand
- DNS: nur noch `*.netralax.de` aktiv; `*.netralax.cloud` entfernt (zuletzt `update.netralax.cloud`).
- Caddyfile: nur noch die `.de`-Vhosts (Legacy-Block + Kommentar entfernt).
- `scripts/prod/config.sh` ist die alleinige Live-Konfiguration; `scripts/migrate/` wird nicht mehr gebraucht (kann archiviert bleiben).
- Lokale Sicherungen (`old.env.backup`, `_storage_stage/`, `_updates_stage/`) sicher löschen (`shred`/Secure-Delete), da sie Secrets enthalten.
---
**Grounding-Hinweise (Repo-Fakten):** Postgres-Major-Version aus `supabase/config.toml` = `17`. Edge-Functions im Repo: `supabase/functions/{mint-livekit-token,notify-push,og-preview}`. Dev-`infra/livekit/livekit.yaml` enthält absichtlich `use_external_ip: false` + `node_ip: 127.0.0.1` (Dev-only — in Prod invertiert/entfernt). `apps/desktop/.env`, `apps/mobile/.env.local`, `.env.release`, `scripts/prod/config.sh` und `apps/desktop/src/lib/changelog.ts` sind bereits auf `.de` umgestellt (verifiziert).
@@ -0,0 +1,673 @@
# Message-List / Scroll Rewrite — Implementation Plan
> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
**Goal:** Replace the `react-virtuoso` message list with a TanStack-Virtual list that opens/switches chats flicker-free (Discord-like), preserving every existing behavior.
**Architecture:** Pure scroll-decision logic (`scrollController.ts`, unit-tested) + an isolated virtualization component (`MessageList.tsx`, TanStack Virtual, deferred reveal) + `ConversationPage` wiring. The flicker is killed by keeping the list hidden until messages+reactions+divider are stable, then anchoring before paint.
**Tech Stack:** React 18, TypeScript, `@tanstack/react-virtual` (new), vitest, electron-vite.
Spec: `docs/superpowers/specs/2026-06-02-message-list-scroll-rewrite-design.md`
---
## File Structure
- Create: `apps/desktop/src/lib/scrollController.ts` — pure scroll math (no DOM/React).
- Create: `apps/desktop/src/lib/scrollController.test.ts` — vitest unit tests.
- Create: `apps/desktop/src/components/MessageList.tsx` — TanStack virtual list + reveal/stick/load-older. Exports `MessageList`, `MessageListHandle`, `VirtuosoRow` is imported from ConversationPage's shared type (moved in Task 5).
- Modify: `apps/desktop/src/lib/useMessageReactions.ts` — add `ready` flag for the reveal gate.
- Modify: `apps/desktop/src/pages/ConversationPage.tsx` — export the row type, swap `<Virtuoso>` for `<MessageList>`, drive the handle, pass `ready`.
- Modify: `apps/desktop/package.json` — add `@tanstack/react-virtual`; remove `react-virtuoso` (Task 8).
---
## Task 1: Add the TanStack Virtual dependency
**Files:**
- Modify: `apps/desktop/package.json`
- [ ] **Step 1: Install**
Run (from repo root `chat-app/`):
```bash
pnpm --filter @chat-app/desktop add @tanstack/react-virtual@^3.10.0
```
Expected: adds `@tanstack/react-virtual` to `apps/desktop/package.json` dependencies; lockfile updated.
- [ ] **Step 2: Verify it resolves**
Run: `pnpm --filter @chat-app/desktop exec node -e "require.resolve('@tanstack/react-virtual'); console.log('ok')"`
Expected: `ok`
- [ ] **Step 3: Commit**
```bash
git add apps/desktop/package.json pnpm-lock.yaml
git commit -m "build(desktop): add @tanstack/react-virtual"
```
---
## Task 2: Pure scroll-decision logic (TDD)
**Files:**
- Create: `apps/desktop/src/lib/scrollController.ts`
- Test: `apps/desktop/src/lib/scrollController.test.ts`
- [ ] **Step 1: Write the failing tests**
```ts
// apps/desktop/src/lib/scrollController.test.ts
import { describe, expect, it } from 'vitest';
import { isNearBottom, isNearTop, resolveInitialAnchor } from './scrollController';
const m = (scrollTop: number, scrollHeight: number, clientHeight: number) => ({
scrollTop,
scrollHeight,
clientHeight,
});
describe('isNearBottom', () => {
it('true exactly at the bottom', () => {
expect(isNearBottom(m(900, 1000, 100), 64)).toBe(true);
});
it('true within threshold', () => {
expect(isNearBottom(m(860, 1000, 100), 64)).toBe(true);
});
it('false beyond threshold', () => {
expect(isNearBottom(m(800, 1000, 100), 64)).toBe(false);
});
});
describe('isNearTop', () => {
it('true at top', () => {
expect(isNearTop(m(0, 1000, 100), 64)).toBe(true);
});
it('false past threshold', () => {
expect(isNearTop(m(200, 1000, 100), 64)).toBe(false);
});
});
describe('resolveInitialAnchor', () => {
it('anchors to last row at end by default (no saved position)', () => {
expect(resolveInitialAnchor(null, 50)).toEqual({ index: 49, align: 'end' });
});
it('anchors to bottom when saved position stuck to bottom', () => {
expect(resolveInitialAnchor({ topmostIndex: 10, stickToBottom: true }, 50)).toEqual({
index: 49,
align: 'end',
});
});
it('restores the saved row at the top when scrolled up', () => {
expect(resolveInitialAnchor({ topmostIndex: 12, stickToBottom: false }, 50)).toEqual({
index: 12,
align: 'start',
});
});
it('clamps a stale saved index to the current row count', () => {
expect(resolveInitialAnchor({ topmostIndex: 999, stickToBottom: false }, 50)).toEqual({
index: 49,
align: 'start',
});
});
it('handles an empty list', () => {
expect(resolveInitialAnchor(null, 0)).toEqual({ index: 0, align: 'end' });
});
});
```
- [ ] **Step 2: Run, verify FAIL**
Run: `pnpm --filter @chat-app/desktop exec vitest run src/lib/scrollController.test.ts`
Expected: FAIL — "Failed to resolve import './scrollController'".
- [ ] **Step 3: Implement**
```ts
// apps/desktop/src/lib/scrollController.ts
// Pure, DOM-free scroll-decision logic for MessageList. Unit-tested so the
// tricky math is verified without a browser (jsdom has no layout).
export interface ScrollMetrics {
scrollTop: number;
scrollHeight: number;
clientHeight: number;
}
/** Distance from the bottom edge is within `threshold` px. */
export function isNearBottom(m: ScrollMetrics, threshold: number): boolean {
return m.scrollHeight - (m.scrollTop + m.clientHeight) <= threshold;
}
/** Scroll offset is within `threshold` px of the top. */
export function isNearTop(m: ScrollMetrics, threshold: number): boolean {
return m.scrollTop <= threshold;
}
export interface SavedPosition {
topmostIndex: number;
stickToBottom: boolean;
}
export interface Anchor {
index: number;
align: 'start' | 'end';
}
/**
* Where a freshly-opened chat should start.
* - default / "left at bottom" → last row, aligned to the viewport bottom.
* - "left scrolled up" → the saved top-most row, aligned to the viewport top
* (clamped in case the cached row count shrank).
*/
export function resolveInitialAnchor(saved: SavedPosition | null, rowCount: number): Anchor {
if (rowCount <= 0) return { index: 0, align: 'end' };
if (saved && !saved.stickToBottom) {
const index = Math.max(0, Math.min(saved.topmostIndex, rowCount - 1));
return { index, align: 'start' };
}
return { index: rowCount - 1, align: 'end' };
}
```
- [ ] **Step 4: Run, verify PASS**
Run: `pnpm --filter @chat-app/desktop exec vitest run src/lib/scrollController.test.ts`
Expected: PASS (11 tests).
- [ ] **Step 5: Commit**
```bash
git add apps/desktop/src/lib/scrollController.ts apps/desktop/src/lib/scrollController.test.ts
git commit -m "feat(desktop): pure scroll-decision logic for new message list"
```
---
## Task 3: Reveal-gate flag on `useMessageReactions`
**Files:**
- Modify: `apps/desktop/src/lib/useMessageReactions.ts`
Reactions are the main post-paint height changer. The list reveal waits on their first
fetch, so add a `ready` flag that is true once reactions for the current message-id set
have been fetched (or there are no messages).
- [ ] **Step 1: Add `ready` to the result type + state**
In `UseMessageReactionsResult` add:
```ts
ready: boolean;
```
After `const [rows, setRows] = useState<MessageReaction[]>([]);` add:
```ts
const [readyKey, setReadyKey] = useState<string | null>(null);
```
- [ ] **Step 2: Set the key after each fetch**
Replace the `refresh` callback body so both branches stamp `readyKey`:
```ts
const refresh = useCallback(async () => {
if (messageIds.length === 0) {
setRows([]);
setReadyKey(idsKey);
return;
}
try {
const data = await listReactionsForMessages(supabase, messageIds);
setRows(data);
} catch (err: unknown) {
console.error('listReactionsForMessages failed', err);
} finally {
setReadyKey(idsKey);
}
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [idsKey]);
```
- [ ] **Step 3: Derive + return `ready`**
Before the `return`:
```ts
const ready = readyKey === idsKey;
```
And add `ready` to the returned object:
```ts
return { byMessage, toggle, voteExclusive, ready };
```
- [ ] **Step 4: Typecheck**
Run: `pnpm --filter @chat-app/desktop typecheck`
Expected: PASS (no output).
- [ ] **Step 5: Commit**
```bash
git add apps/desktop/src/lib/useMessageReactions.ts
git commit -m "feat(desktop): expose reactions reveal-gate flag (ready)"
```
---
## Task 4: Export the shared row type from ConversationPage
**Files:**
- Modify: `apps/desktop/src/pages/ConversationPage.tsx`
`MessageList` needs the row union. Export it from ConversationPage (smallest change;
the type already lives there).
- [ ] **Step 1: Export the type**
Change the `type VirtuosoRow = …` declaration (near the top of the file) to:
```ts
export type VirtuosoRow =
| { kind: 'loader'; key: string }
| { kind: 'message'; key: string; message: DecryptedMessage; idx: number }
| { kind: 'pending'; key: string; item: OutboxItem };
```
- [ ] **Step 2: Typecheck**
Run: `pnpm --filter @chat-app/desktop typecheck`
Expected: PASS.
- [ ] **Step 3: Commit**
```bash
git add apps/desktop/src/pages/ConversationPage.tsx
git commit -m "refactor(desktop): export VirtuosoRow type for MessageList"
```
---
## Task 5: The `MessageList` component
**Files:**
- Create: `apps/desktop/src/components/MessageList.tsx`
This is the integration unit. It is verified by typecheck here and **visually in dev**
in Task 7 (jsdom can't layout-test it). The TanStack specifics (scrollToIndex timing,
prepend offset) are the parts to refine during dev iteration.
- [ ] **Step 1: Implement**
```tsx
// apps/desktop/src/components/MessageList.tsx
import { useVirtualizer } from '@tanstack/react-virtual';
import {
forwardRef,
useCallback as _unused, // placeholder removed below
} from 'react';
```
> NOTE for the implementer: write the file with the imports below (the line above is
> illustrative only — do not keep it). Full file:
```tsx
import { useVirtualizer } from '@tanstack/react-virtual';
import {
forwardRef,
useCallback,
useImperativeHandle,
useLayoutEffect,
useRef,
useState,
type ReactNode,
} from 'react';
import { isNearBottom, isNearTop, resolveInitialAnchor, type Anchor } from '../lib/scrollController';
import type { VirtuosoRow } from '../pages/ConversationPage';
export interface MessageListHandle {
scrollToBottom(behavior?: ScrollBehavior): void;
scrollToRow(index: number, align?: 'center' | 'end', behavior?: ScrollBehavior): void;
}
export interface MessageListProps {
rows: VirtuosoRow[];
renderRow: (index: number, row: VirtuosoRow) => ReactNode;
computeKey: (row: VirtuosoRow) => string;
initialAnchor: { type: 'bottom' } | { type: 'row'; index: number };
/** Reveal gate — list stays hidden behind a spinner until true (no flicker). */
ready: boolean;
estimateRowHeight?: number;
atBottomThreshold?: number;
onReachTop?: () => void;
onAtBottomChange?: (atBottom: boolean) => void;
onTopRowChange?: (topIndex: number) => void;
}
export const MessageList = forwardRef<MessageListHandle, MessageListProps>(function MessageList(
{
rows,
renderRow,
computeKey,
initialAnchor,
ready,
estimateRowHeight = 64,
atBottomThreshold = 64,
onReachTop,
onAtBottomChange,
onTopRowChange,
},
ref,
) {
const scrollElRef = useRef<HTMLDivElement>(null);
const [revealed, setRevealed] = useState(false);
const atBottomRef = useRef(true);
// Load-older preservation: remember scrollHeight + first key across renders.
const prevFirstKeyRef = useRef<string | null>(null);
const prevScrollHeightRef = useRef(0);
const virtualizer = useVirtualizer({
count: rows.length,
getScrollElement: () => scrollElRef.current,
estimateSize: () => estimateRowHeight,
overscan: 8,
getItemKey: (index) => computeKey(rows[index]!),
});
const metrics = () => {
const el = scrollElRef.current;
return el
? { scrollTop: el.scrollTop, scrollHeight: el.scrollHeight, clientHeight: el.clientHeight }
: { scrollTop: 0, scrollHeight: 0, clientHeight: 0 };
};
const applyAnchor = useCallback(
(anchor: Anchor) => {
virtualizer.scrollToIndex(anchor.index, { align: anchor.align });
// Re-apply on the next frame: dynamic measurement settles after the first
// paint, so a single scrollToIndex can land a few px off. The list is still
// hidden here, so this correction is never visible.
requestAnimationFrame(() => virtualizer.scrollToIndex(anchor.index, { align: anchor.align }));
},
[virtualizer],
);
// Deferred reveal: when ready, anchor (before paint) then reveal.
useLayoutEffect(() => {
if (!ready || revealed || rows.length === 0) return;
const anchor: Anchor =
initialAnchor.type === 'bottom'
? { index: rows.length - 1, align: 'end' }
: { index: Math.max(0, Math.min(initialAnchor.index, rows.length - 1)), align: 'start' };
applyAnchor(anchor);
atBottomRef.current = initialAnchor.type === 'bottom';
onAtBottomChange?.(atBottomRef.current);
requestAnimationFrame(() => setRevealed(true));
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [ready, rows.length]);
// Stick-to-bottom: when content grows and we were at the bottom, re-pin.
useLayoutEffect(() => {
if (!revealed) return;
if (atBottomRef.current) {
virtualizer.scrollToIndex(rows.length - 1, { align: 'end' });
}
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [rows.length, virtualizer.getTotalSize()]);
// Load-older preservation: if rows were prepended (first key changed and count
// grew), restore scrollTop by the height delta so the viewport stays put.
useLayoutEffect(() => {
const firstKey = rows.length > 0 ? computeKey(rows[0]!) : null;
const el = scrollElRef.current;
if (el && revealed && prevFirstKeyRef.current && firstKey !== prevFirstKeyRef.current) {
const delta = el.scrollHeight - prevScrollHeightRef.current;
if (delta > 0 && el.scrollTop < atBottomThreshold) {
el.scrollTop += delta;
}
}
prevFirstKeyRef.current = firstKey;
prevScrollHeightRef.current = el?.scrollHeight ?? 0;
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [rows]);
const handleScroll = useCallback(() => {
const m = metrics();
const atBottom = isNearBottom(m, atBottomThreshold);
if (atBottom !== atBottomRef.current) {
atBottomRef.current = atBottom;
onAtBottomChange?.(atBottom);
}
if (isNearTop(m, atBottomThreshold * 4)) onReachTop?.();
const first = virtualizer.getVirtualItems()[0];
if (first) onTopRowChange?.(first.index);
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [atBottomThreshold, onAtBottomChange, onReachTop, onTopRowChange, virtualizer]);
useImperativeHandle(
ref,
() => ({
scrollToBottom: () => {
atBottomRef.current = true;
virtualizer.scrollToIndex(rows.length - 1, { align: 'end' });
},
scrollToRow: (index, align = 'center') => {
virtualizer.scrollToIndex(index, { align });
},
}),
// eslint-disable-next-line react-hooks/exhaustive-deps
[virtualizer, rows.length],
);
const items = virtualizer.getVirtualItems();
return (
<div
ref={scrollElRef}
onScroll={handleScroll}
className="h-full overflow-y-auto"
style={{ opacity: revealed ? 1 : 0, position: 'relative' }}
>
<div style={{ height: virtualizer.getTotalSize(), position: 'relative', width: '100%' }}>
{items.map((vi) => (
<div
key={vi.key}
data-index={vi.index}
ref={virtualizer.measureElement}
style={{ position: 'absolute', top: 0, left: 0, width: '100%', transform: `translateY(${vi.start}px)` }}
>
{renderRow(vi.index, rows[vi.index]!)}
</div>
))}
</div>
{/* 12px bottom breathing space (matches the old Footer). */}
<div style={{ height: 12 }} />
</div>
);
});
```
> Implementer note: delete the illustrative first `import` snippet; keep only the full
> file. The `requestAnimationFrame` timing in `applyAnchor`/reveal is the most likely
> spot to refine during dev (Task 7).
- [ ] **Step 2: Typecheck**
Run: `pnpm --filter @chat-app/desktop typecheck`
Expected: PASS.
- [ ] **Step 3: Commit**
```bash
git add apps/desktop/src/components/MessageList.tsx
git commit -m "feat(desktop): TanStack-Virtual MessageList with deferred reveal"
```
---
## Task 6: Wire `MessageList` into `ConversationPage`
**Files:**
- Modify: `apps/desktop/src/pages/ConversationPage.tsx`
- [ ] **Step 1: Imports + reveal gate**
Replace the `react-virtuoso` import with:
```ts
import { MessageList, type MessageListHandle } from '../components/MessageList';
```
Capture the reactions `ready` flag — change the `useMessageReactions` destructure to also pull `ready`:
```ts
const {
byMessage: reactionsByMessage,
toggle: toggleReaction,
voteExclusive: votePoll,
ready: reactionsReady,
} = useMessageReactions(messageIds, session?.user.id);
```
Add a reveal gate with a 300ms max-timeout fallback (so empty/slow reactions never hang):
```ts
const [revealTimedOut, setRevealTimedOut] = useState(false);
useEffect(() => {
if (!id || loading || messages.length === 0) return;
const t = window.setTimeout(() => setRevealTimedOut(true), 300);
return () => window.clearTimeout(t);
}, [id, loading, messages.length]);
const listReady = !loading && messages.length > 0 && (reactionsReady || revealTimedOut);
```
- [ ] **Step 2: Replace the `virtuosoRef` type + handle**
Change:
```ts
const virtuosoRef = useRef<VirtuosoHandle>(null);
```
to:
```ts
const listRef = useRef<MessageListHandle>(null);
```
Replace every `virtuosoRef.current?.scrollToIndex({ index: 'LAST', align: 'end', behavior })`
call (in `jumpToBottom`, the pending-snap effect, `snapToBottom`) with:
```ts
listRef.current?.scrollToBottom('auto');
```
Replace the `jumpToMessage` scroll (`virtuosoRef.current?.scrollToIndex({ index: rowIndex, align: 'center', behavior: 'smooth' })`) with:
```ts
listRef.current?.scrollToRow(rowIndex, 'center', 'smooth');
```
- [ ] **Step 3: Compute `initialAnchor`**
Replace the `initialTopMostIndex` `useMemo` (the `IndexLocationWithAlign` one from the
earlier hotfix) with:
```ts
const initialAnchor = useMemo<{ type: 'bottom' } | { type: 'row'; index: number }>(() => {
const saved = savedPositionRef.current;
if (saved && !saved.stickToBottom) return { type: 'row', index: saved.topmostIndex };
return { type: 'bottom' };
}, []);
```
Remove the now-unused `IndexLocationWithAlign` import.
- [ ] **Step 4: Swap the JSX**
Replace the entire `<Virtuoso … />` element with:
```tsx
<MessageList
ref={listRef}
rows={virtuosoRows}
ready={listReady}
computeKey={(row) => row.key}
initialAnchor={initialAnchor}
atBottomThreshold={250}
onReachTop={handleStartReached}
onAtBottomChange={handleAtBottomStateChange}
onTopRowChange={(topIndex) => handleRangeChanged({ startIndex: topIndex, endIndex: topIndex })}
renderRow={(_index, row) => {
// ...exact same body the old `itemContent` had (loader / pending /
// message branches) — move it verbatim from the deleted <Virtuoso>.
return renderConversationRow(row);
}}
/>
```
Move the old `itemContent` body into a local `renderConversationRow(row)` helper (or inline it) so the message/loader/pending branches are unchanged. `handleRangeChanged` already accepts `{ startIndex, endIndex }`.
- [ ] **Step 5: Typecheck + unit tests**
Run: `pnpm --filter @chat-app/desktop typecheck && pnpm --filter @chat-app/desktop test`
Expected: PASS.
- [ ] **Step 6: Commit**
```bash
git add apps/desktop/src/pages/ConversationPage.tsx
git commit -m "feat(desktop): use MessageList in ConversationPage (replace react-virtuoso)"
```
---
## Task 7: Dev verification (with the user) — iterate until smooth
**Files:** none (runtime verification)
- [ ] **Step 1: Run the dev build**
User runs (in `chat-app/`): `pnpm desktop:dev`
- [ ] **Step 2: Verify behaviors live**
Switch between several chats repeatedly and confirm, using the `SCROLL_DEBUG` console
output where helpful:
- No jump and no multi-flicker on chat switch (opens cleanly at the bottom / saved row).
- New message while at bottom auto-scrolls; while scrolled up shows the pill.
- Unread divider present without a later shift.
- Scroll to top loads older without the viewport jumping.
- Jump-to-message (reply tap / pinned / search) scrolls to the target.
- Sent/pending message snaps to bottom.
- [ ] **Step 3: Refine**
If any behavior is off, adjust `MessageList.tsx` (most likely the `applyAnchor`/reveal
`requestAnimationFrame` timing or the stick-to-bottom effect) and re-verify. Commit each
refinement:
```bash
git commit -am "fix(desktop): refine MessageList <specific behavior>"
```
---
## Task 8: Cleanup + release 0.21.6
**Files:**
- Modify: `apps/desktop/src/pages/ConversationPage.tsx` (remove instrumentation)
- Modify: `apps/desktop/package.json` (remove `react-virtuoso`)
- [ ] **Step 1: Remove the `SCROLL_DEBUG` instrumentation**
Delete the `SCROLL_DEBUG`/`dbgNow`/`dbgLog` block, the render-logger `useEffect`, and the
`dbgLog(...)` calls inside `handleRangeChanged` / `handleAtBottomStateChange`.
- [ ] **Step 2: Remove the old dependency**
Run: `pnpm --filter @chat-app/desktop remove react-virtuoso`
Then confirm no references remain:
Run: `grep -rn "react-virtuoso\|Virtuoso\b" apps/desktop/src || echo "clean"`
Expected: `clean`.
- [ ] **Step 3: Typecheck + tests + commit**
Run: `pnpm --filter @chat-app/desktop typecheck && pnpm --filter @chat-app/desktop test`
Expected: PASS.
```bash
git add -A
git commit -m "chore(desktop): drop react-virtuoso + scroll debug instrumentation"
```
- [ ] **Step 4: Release**
Run (from `chat-app/`, tree clean): `node scripts/release.mjs 0.21.6 "- Nachrichtenliste komplett überarbeitet: Chat-Wechsel öffnet jetzt ruckel- und flackerfrei direkt unten\n- Älteren Verlauf laden springt nicht mehr"`
Then verify `latest.yml` shows 0.21.6 on `update.netralax.de` **and** `update.netralax.cloud`.
---
## Self-Review
- **Spec coverage:** deferred reveal (Tasks 3,5,6) ✓; TanStack virtualization (Tasks 1,5) ✓; isolation into MessageList + scrollController (Tasks 2,5) ✓; stick-to-bottom (Task 5) ✓; load-older preservation (Task 5) ✓; preserved behaviors incl. jump-to-message/pill/divider/pending (Task 6) ✓; pure-logic unit tests (Task 2) ✓; dev verification (Task 7) ✓; cleanup + release (Task 8) ✓.
- **Placeholders:** the only prose-only steps are the deliberately runtime Task 7 (no code possible) and the "move itemContent verbatim" in Task 6 Step 4 (the body is large and unchanged — copying it verbatim, not rewriting). The illustrative throwaway import in Task 5 Step 1 is explicitly flagged for deletion.
- **Type consistency:** `MessageListHandle.scrollToBottom/scrollToRow`, `VirtuosoRow`, `Anchor`, `ScrollMetrics`, `resolveInitialAnchor` signatures are consistent across Tasks 2/5/6. `ready` flag added in Task 3 is consumed in Task 6.
@@ -0,0 +1,184 @@
# Message-List / Scroll Rewrite — Design
Date: 2026-06-02
Status: Approved (brainstorming) — pending spec review → implementation plan
Scope: Desktop app only (`apps/desktop`). Mobile is out of scope.
## 1. Problem & Root Cause
Switching conversations causes a visible jump and then a multi-flicker. Root cause
(established via systematic debugging, not guessing):
- The message list is **virtualized** (`react-virtuoso`). Virtualization paints rows
with *estimated* heights, then measures real heights and corrects `scrollTop`.
- On chat open, several async sources change **row heights after the first paint**:
message **reactions** (`useMessageReactions`), the **unread divider**
(`firstUnreadId`), delivery/read **receipts**, and the **two-phase message load**
(in-memory cache render → server `refresh()` replaces the array).
- Each post-paint height change makes the virtualizer re-measure and re-anchor →
the viewport visibly moves several times = "flickert paar mal".
A first targeted fix (`initialTopMostItemIndex: { index: 'LAST', align: 'end' }`)
addressed only the *initial* anchor, not the post-paint cascade — so the flicker
remained/worsened. Conclusion: re-architect the scroll system.
## 2. Goals / Success Criteria
1. Opening or switching a chat lands cleanly at the bottom (or the saved scrolled-up
row) with **no visible jump or flicker**.
2. Discord-like live behavior: auto-scroll on new message when at bottom; "X new
messages" pill when scrolled up; unread divider; load-older without the viewport
jumping; jump-to-message / search / pin scroll.
3. Scales to **large conversations** (thousands of messages, deep back-scroll) —
virtualization stays.
4. No regression of the existing features that live in `ConversationPage`.
## 3. Decision
Build on **`@tanstack/react-virtual`** (MIT, free) as the virtualization primitive,
and kill the flicker at its root with a **deferred-reveal** strategy: never show the
list while its row heights are still settling.
Rejected alternatives: keeping `react-virtuoso` (we are fighting it); the commercial
`@virtuoso.dev/message-list` (license cost); dropping virtualization entirely
(large chats would render thousands of DOM nodes).
## 4. Architecture (isolation)
The scroll/virtualization logic moves out of the ~2000-line `ConversationPage` into
two focused, independently-testable units:
- **`apps/desktop/src/components/MessageList.tsx`** — owns the scroll container,
the TanStack virtualizer, dynamic measurement, deferred reveal, stick-to-bottom,
and load-older position preservation. Receives rows + a render function; emits
scroll events + exposes an imperative handle. Knows nothing about messages,
reactions, drafts, calls, etc.
- **`apps/desktop/src/lib/scrollController.ts`** — the **pure**, DOM-free decision
logic (anchor computation, "should auto-scroll to bottom?", load-older index/offset
math, at-bottom threshold). Unit-tested with vitest.
- **`ConversationPage`** keeps all feature state and rendering; it builds the same
`VirtuosoRow[]` discriminated union (`loader | message | pending`), passes them +
the existing per-row render (`itemContent`) into `<MessageList>`, and drives the
imperative handle for jump-to-message/search.
Boundary contract: *in* = rows + renderRow; *out* = scroll events + an imperative
handle. The internals of `MessageList` can change without touching `ConversationPage`.
## 5. No-Flicker Core
### 5.1 Deferred reveal
`MessageList` is always mounted (so TanStack can measure the initial window), but
rendered **visually hidden** (`opacity: 0`, pointer-events none) behind a spinner
until `ready` is true. When `ready` flips true, in a `useLayoutEffect` (before the
browser paints) it scrolls to `initialAnchor` (bottom, or the saved row), then
reveals (`opacity: 1`) and removes the spinner. The user sees: brief spinner →
final, correctly-anchored list. The height-changing cascade happens **while hidden**.
`ready` (owned by `ConversationPage`, passed in) is defined as:
- messages loaded (`!loading && rows.length > 0`), **AND**
- the initial **reactions** fetch for the current message-id set has completed
(requires adding a `ready`/`loaded` flag to `useMessageReactions`), **AND**
- a hard **max-timeout of ~300 ms** fallback so a slow/empty reactions fetch never
hangs the reveal.
The unread divider is computed synchronously in an effect right after messages load,
i.e. before reactions resolve — so it is present before reveal. Delivery/read receipts
render as inline ticks (no meaningful height change) and are intentionally **not**
gated.
### 5.2 Stick-to-bottom
A `ResizeObserver` on the inner content element: while the user is at the bottom
(within `atBottomThreshold`, default 64px), any content-size growth re-pins the view
to the bottom in a layout effect (before paint) — so a live incoming message/reaction
never leaves the newest message half-scrolled.
### 5.3 Dynamic measurement
TanStack `measureElement` (ResizeObserver per rendered row) handles variable bubble
heights. Only the virtual window (visible + overscan ~8 rows) is rendered/measured.
### 5.4 Load-older without jump
On `onReachTop`, `ConversationPage` grows `displayCount` (prepending older rows).
Because prepending shifts indices, `MessageList` preserves position: capture
`scrollHeight` before the row growth, then after re-render set
`scrollTop += (newScrollHeight oldScrollHeight)` in a layout effect keyed on
"rows grew at the top". Items keep stable keys via `computeKey` (message id). This
also fixes the second audit gap (older-load jump).
## 6. Preserved Behaviors
| Behavior | New mechanism |
|---|---|
| Open → bottom / saved row | `initialAnchor` applied in `useLayoutEffect` before reveal |
| New message while at bottom → follow | stick-to-bottom controller |
| New message while scrolled up → pill | `onAtBottomChange` drives the existing pill |
| Unread divider | computed before reveal → no later height change |
| Load older (scroll to top) | `onReachTop` + scrollHeight-delta preservation |
| Jump-to-message / search / pin | imperative `scrollToRow(index, align)` |
| Pending (outbox) bubbles | stay as a row kind in the same list |
| Per-conversation scroll memory | unchanged module-scoped `scrollPositions` map, fed by `onAtBottomChange` + `onTopRowChange` |
## 7. Interface
```ts
export interface MessageListHandle {
scrollToBottom(behavior?: 'auto' | 'smooth'): void;
scrollToRow(index: number, align?: 'center' | 'end', behavior?: 'auto' | 'smooth'): void;
}
export interface MessageListProps {
rows: VirtuosoRow[]; // loader | message | pending
renderRow: (index: number, row: VirtuosoRow) => React.ReactNode;
computeKey: (row: VirtuosoRow) => string; // message id / pending id / '__loader__'
initialAnchor: { type: 'bottom' } | { type: 'row'; index: number };
ready: boolean; // reveal gate (§5.1)
estimateRowHeight?: number; // default ~64
atBottomThreshold?: number; // default 64px
onReachTop(): void; // load older
onAtBottomChange(atBottom: boolean): void;
onTopRowChange(topIndex: number): void; // scroll memory
}
```
## 8. Error Handling / Edge Cases
- Empty conversation: `rows.length === 0``MessageList` renders nothing; `ready`
short-circuits to the existing empty state in `ConversationPage`.
- Single / very short conversation (content < viewport): bottom-anchor is a no-op;
reveal immediately.
- Rapid chat switching: each switch remounts `ConversationPage` (per-id key) → a
fresh `MessageList` instance with fresh measurements (no stale sizes carried over).
- Very large `displayCount` after deep back-scroll: only the virtual window renders;
memory bounded by overscan.
- Reactions fetch error/empty: max-timeout reveals the list anyway.
## 9. Testing
- **Unit (vitest):** `scrollController.ts` pure functions — anchor resolution,
should-auto-scroll decision, load-older offset math, at-bottom threshold.
- **Manual (dev):** iterate in `pnpm desktop:dev` with the temporary `SCROLL_DEBUG`
logging until chat-switch is flicker-free and all §6 behaviors verified live.
- No automated DOM/layout test (jsdom has no layout); the running app is the test.
## 10. Rollout
1. Add `@tanstack/react-virtual`.
2. Build `scrollController.ts` (+ tests) and `MessageList.tsx`.
3. Swap the `<Virtuoso>` block in `ConversationPage` for `<MessageList>`; add the
`ready` flag to `useMessageReactions`.
4. Verify in dev with the user (flicker-free + all behaviors).
5. Remove the `SCROLL_DEBUG` instrumentation and the `react-virtuoso` dependency.
6. Release **0.21.6** to `update.netralax.de` (served on `.de` + `.cloud`).
## 11. Out of Scope
Composer, header, dialogs, calls, search UI, message rendering (`MessageBubble`),
encryption/data layer, mobile app. Reaction/receipt *data* loading is touched only to
add the `ready` flag for the reveal gate.
## 12. Open Risks
- TanStack prepend position-preservation needs careful layout-effect timing; mitigated
by dev iteration before release.
- The `ready` reveal adds a brief (≤300 ms) spinner on chat open even for cached
chats; acceptable trade-off vs flicker. A future in-memory reaction cache could make
revisits instant (not in this scope).
+91
View File
@@ -0,0 +1,91 @@
# ─────────────────────────────────────────────────────────────────────────────
# Caddyfile — Produktion (NEW VPS, netralax.de)
#
# Dual-Domain-Übergang (.de + .cloud):
# Bereits installierte Desktop- (Vite) und Mobile- (Expo) Clients haben die
# ALTEN Hostnamen fest in ihre Bundles eingebacken
# (supabase.netralax.cloud, livekit.netralax.cloud, update.netralax.cloud).
# Deshalb bedient dieser NEUE Server BEIDE Domains aus denselben Backends:
# - die neuen *.netralax.de Hosts für aktuelle/neue Releases
# - die legacy *.netralax.cloud Hosts NUR damit Alt-Installationen weiter
# funktionieren, bis sie sich per Auto-Update auf .de umgestellt haben.
# Voraussetzung: die .cloud-DNS-A-Records müssen auf die NEUE VPS-IP zeigen.
# Die .cloud-Blöcke dürfen NICHT entfernt werden, solange noch Alt-Clients
# im Umlauf sind — sonst brechen alle bestehenden Installationen.
#
# TLS: Automatisches HTTPS via Let's Encrypt für alle Hosts.
# WebSockets: Caddy v2 reicht Upgrade/Connection-Header bei reverse_proxy
# transparent durch — sowohl für Supabase Realtime (/realtime/v1/websocket)
# als auch für LiveKit (/rtc). KEINE websocket-Direktive nötig/vorhanden.
#
# WICHTIG: Nur der Signaling-WS (7880) und das Supabase-Gateway (Kong 8000)
# laufen über Caddy. RTC-Medien (7881/tcp, 50000-50100/udp) und coturn
# (3478, 5349/TLS, 50200-50300/udp) gehen NICHT über Caddy und müssen direkt
# in der ufw geöffnet werden. TURNS auf 5349 braucht ein EIGENES Zertifikat
# für turn.netralax.de (siehe coturn.prod.conf.example).
# ─────────────────────────────────────────────────────────────────────────────
# ─────────────────────────────────────────────────────────────────────────────
# AKTIV ab Bootstrap: die NEUEN .de-Hosts.
# Die .cloud-Legacy-Blöcke stehen weiter unten und werden ERST beim Cutover
# (Runbook §10) einkommentiert — nämlich NACHDEM die .cloud-A-Records auf die
# neue VPS-IP zeigen. Grund: stehen die .cloud-Namen schon vorher in der aktiven
# Config, scheitert Caddy wiederholt an der Let's-Encrypt-Ausstellung (DNS zeigt
# noch auf den alten Server) und läuft ins ACME-Rate-Limit (5 Fehler/Host/Stunde).
# ─────────────────────────────────────────────────────────────────────────────
# Supabase API-Gateway (Kong multiplext auth/rest/realtime/storage/functions
# + Studio). EIN reverse_proxy genügt — KEINE Routen in Caddy aufsplitten.
supabase.netralax.de {
reverse_proxy localhost:8000
}
# LiveKit Signaling-WebSocket. Caddy übernimmt den WS-Upgrade automatisch.
# CORS-Header + OPTIONS-Preflight wie auf dem alten Server (Browser/Electron-
# Clients erwarten sie beim Token-/Connect-Handshake).
livekit.netralax.de {
header Access-Control-Allow-Origin "*"
header Access-Control-Allow-Methods "GET, POST, OPTIONS"
header Access-Control-Allow-Headers "Authorization, Content-Type"
header Access-Control-Expose-Headers "*"
@options method OPTIONS
handle @options {
respond 204
}
reverse_proxy localhost:7880
}
# electron-updater Artefakte (latest.yml + .exe + changelog.json).
# WICHTIG: docroot ist /var/www/updates (NICHT .../windows). release.mjs lädt
# nach /var/www/updates/windows/ hoch und die Clients holen unter dem URL-Pfad
# /windows/latest.yml — der Pfad-Präfix /windows/ muss also auf das Unterverzeichnis
# mappen. Mit root=/var/www/updates/windows entstünde .../windows/windows → 404.
update.netralax.de {
root * /var/www/updates
file_server
}
# ─────────────────────────────────────────────────────────────────────────────
# LEGACY .cloud-Hosts — AKTIV seit dem Cutover (DNS .cloud → neue VPS-IP).
# Liefern aus denselben Backends wie die .de-Hosts, damit bereits installierte
# Clients weiterlaufen, bis sie sich per Auto-Update auf .de umgestellt haben.
# NICHT entfernen, solange Alt-Clients im Umlauf sind.
# ─────────────────────────────────────────────────────────────────────────────
supabase.netralax.cloud {
reverse_proxy localhost:8000
}
livekit.netralax.cloud {
header Access-Control-Allow-Origin "*"
header Access-Control-Allow-Methods "GET, POST, OPTIONS"
header Access-Control-Allow-Headers "Authorization, Content-Type"
header Access-Control-Expose-Headers "*"
@options method OPTIONS
handle @options {
respond 204
}
reverse_proxy localhost:7880
}
update.netralax.cloud {
root * /var/www/updates
file_server
}
+50
View File
@@ -0,0 +1,50 @@
# ─────────────────────────────────────────────────────────────────────────────
# coturn — Produktionskonfiguration (turnserver.conf) für turn.netralax.de
#
# coturn läuft EIGENSTÄNDIG (LiveKit-internes TURN ist deaktiviert).
# TURNS (5349/TLS) läuft NICHT über Caddy und braucht daher ein EIGENES
# TLS-Zertifikat für turn.netralax.de auf der Platte (cert/pkey unten).
#
# Zertifikat besorgen — zwei Wege:
# (a) certbot standalone (Port 80 muss frei sein, nicht von Caddy belegt):
# certbot certonly --standalone -d turn.netralax.de
# -> liefert /etc/letsencrypt/live/turn.netralax.de/{fullchain,privkey}.pem
# coturn nach Renewals neu laden (z. B. certbot --deploy-hook 'systemctl reload coturn').
# (b) Caddy-Zertifikat wiederverwenden: lasse Caddy zusätzlich turn.netralax.de
# ausstellen und kopiere/symlinke das Zert aus Caddys data-Verzeichnis
# (~/.local/share/caddy/certificates/...) an die Pfade unten. Achtung:
# coturn braucht Leserechte auf cert+pkey.
#
# ufw muss offen sein: 3478/udp+tcp, 5349/tcp (TURNS), 50200-50300/udp (Relay).
# Diese Ports gehen NICHT über Caddy.
#
# external-ip auf die ÖFFENTLICHE IP der NEUEN VPS setzen.
# lt-cred-mech-User muss zu dem passen, den mint-livekit-token / die Clients
# erwarten (Platzhalter unten ersetzen).
# ─────────────────────────────────────────────────────────────────────────────
listening-port=3478
tls-listening-port=5349
# Öffentliche IP der neuen VPS.
external-ip=141.95.34.204
# Relay-Port-Range (muss in ufw offen sein).
min-port=50200
max-port=50300
realm=netralax.de
# Long-Term-Credential-Mechanismus. User-Platzhalter ersetzen
# (Format: user=NAME:PASSWORT). Passwort z. B. via `openssl rand -hex 16`.
lt-cred-mech
user=turnuser:<REPLACE_WITH_TURN_PASSWORD>
# TLS-Material für TURNS (turn.netralax.de) — siehe Kopf-Kommentar.
cert=/etc/letsencrypt/live/turn.netralax.de/fullchain.pem
pkey=/etc/letsencrypt/live/turn.netralax.de/privkey.pem
# Härtung / Korrektheit.
fingerprint
no-multicast-peers
no-cli
@@ -0,0 +1,41 @@
# ─────────────────────────────────────────────────────────────────────────────
# LiveKit + coturn — Produktions-Compose (NEW VPS, netralax.de)
#
# Dies ist die PROD-Variante von infra/livekit/docker-compose.yml (das ist nur
# Dev: coturn läuft dort mit --no-tls/--no-dtls, ohne 5349, ohne Zertifikat).
#
# Auf den Server kopieren als /opt/livekit/docker-compose.yml und daneben:
# /opt/livekit/livekit.yaml <- infra/livekit/livekit.prod.yaml.example (Keys eintragen)
# /opt/livekit/coturn.conf <- infra/livekit/coturn.prod.conf.example (external-ip + Cert)
# Start: cd /opt/livekit && docker compose up -d && docker compose ps
#
# network_mode: host — auf einem Linux-Server ist das für WebRTC der robusteste
# Weg: die RTC-UDP-Range (50000-50100) und die TURN-Relay-Range (50200-50300)
# müssen NICHT einzeln gemappt werden, und coturn/LiveKit sehen die echten
# Quell-IPs. Welche Ports tatsächlich erreichbar sind, regelt ufw (siehe
# Runbook §6.4). Auf macOS/Docker-Desktop wird host-networking NICHT unterstützt
# — dort gilt weiterhin die Dev-Compose mit explizitem Port-Mapping.
# ─────────────────────────────────────────────────────────────────────────────
services:
livekit:
image: livekit/livekit-server:latest
restart: unless-stopped
network_mode: host
command: ["--config", "/etc/livekit.yaml"]
volumes:
- ./livekit.yaml:/etc/livekit.yaml:ro
turn:
image: coturn/coturn:4.6
restart: unless-stopped
network_mode: host
# Prod: vollständige turnserver.conf statt der Dev-CLI-Flags. Diese Datei
# aktiviert TURNS auf 5349 mit dem Zertifikat für turn.netralax.de.
command: ["-c", "/etc/coturn/turnserver.conf"]
volumes:
- ./coturn.conf:/etc/coturn/turnserver.conf:ro
# TLS-Material für turn.netralax.de. coturn.conf verweist mit
# cert=/etc/letsencrypt/live/turn.netralax.de/fullchain.pem (und privkey)
# auf genau diese Pfade — daher /etc/letsencrypt read-only einhängen.
- /etc/letsencrypt:/etc/letsencrypt:ro
+41
View File
@@ -0,0 +1,41 @@
# ─────────────────────────────────────────────────────────────────────────────
# LiveKit — Produktionskonfiguration (NEW VPS)
#
# Diese Datei ERSETZT die Dev-Werte aus infra/livekit/livekit.yaml.
# Unterschiede zur Dev-Config (WICHTIG):
# - rtc.use_external_ip: true (Dev: false)
# - KEIN rtc.node_ip: 127.0.0.1 (Dev-only — würde im Prod jeden Client
# veranlassen, Medien an seinen eigenen Loopback zu senden: Call verbindet,
# aber KEIN Audio/Video).
# - echte keys: (Platzhalter unten) statt der öffentlich bekannten devkey.
#
# Die keys: müssen EXAKT zu LIVEKIT_API_KEY / LIVEKIT_API_SECRET in
# /opt/supabase/.env passen (mint-livekit-token signiert damit). Wird nur eine
# Seite rotiert, lehnt die SFU die Tokens beim Join ab (403).
#
# ufw muss offen sein: 7880/tcp (Signaling, hinter Caddy), 7881/tcp (RTC TCP),
# 50000-50100/udp (RTC). Diese Ports außer 7880 gehen NICHT über Caddy.
#
# Kopiere diese Datei als /opt/livekit/livekit.yaml und trage echte Keys ein.
# ─────────────────────────────────────────────────────────────────────────────
port: 7880
log_level: info
rtc:
tcp_port: 7881
port_range_start: 50000
port_range_end: 50100
# Prod: öffentliche IP des Servers ankündigen (NICHT Loopback wie im Dev).
use_external_ip: true
# KEIN node_ip hier — das war dev-only (127.0.0.1) und bricht im Prod die Medien.
# Produktionsschlüssel — Platzhalter. Muss zu /opt/supabase/.env passen
# (LIVEKIT_API_KEY = der key, LIVEKIT_API_SECRET = das secret).
# Erzeugen z. B. mit: openssl rand -hex 32
keys:
APIxxxxxxxxxxxx: <REPLACE_WITH_LIVEKIT_API_SECRET>
# coturn läuft separat (siehe coturn.prod.conf.example) — eingebauter TURN aus.
turn:
enabled: false
+22 -1
View File
@@ -28,7 +28,28 @@ export interface ConvKeyHandle {
const cache = new Map<string, ConvKeyHandle>(); const cache = new Map<string, ConvKeyHandle>();
const cacheKey = (convId: string, v: number) => convId + '@' + v; const cacheKey = (convId: string, v: number) => convId + '@' + v;
export function clearConvKeyCache(): void { cache.clear(); } // Clear the in-memory conv-key cache. Three modes:
// * no args → clear everything (e.g. on logout)
// * convId only → clear all key-version entries for this conversation
// * convId + v → clear just the specific (conv, version) entry
//
// Callers that observe a peer rotation or a server-side conv-keys mutation
// MUST invalidate the affected entries so subsequent `getOrCreateConvKey` /
// `tryGetConvKey` calls re-fetch the canonical bundle from the server
// instead of returning a now-stale cached key.
export function clearConvKeyCache(conversationId?: string, keyVersion?: number): void {
if (conversationId === undefined) {
cache.clear();
return;
}
if (keyVersion !== undefined) {
cache.delete(cacheKey(conversationId, keyVersion));
return;
}
for (const key of Array.from(cache.keys())) {
if (key.startsWith(conversationId + '@')) cache.delete(key);
}
}
async function listMemberPublicKeys( async function listMemberPublicKeys(
client: AppSupabaseClient, client: AppSupabaseClient,
+20 -14
View File
@@ -65,6 +65,9 @@ importers:
'@supabase/supabase-js': '@supabase/supabase-js':
specifier: ^2.46.0 specifier: ^2.46.0
version: 2.103.3 version: 2.103.3
'@tanstack/react-virtual':
specifier: ^3.10.0
version: 3.14.2(react-dom@18.3.1(react@18.3.1))(react@18.3.1)
better-sqlite3: better-sqlite3:
specifier: ^11.3.0 specifier: ^11.3.0
version: 11.10.0 version: 11.10.0
@@ -98,9 +101,6 @@ importers:
react-router-dom: react-router-dom:
specifier: ^6.28.0 specifier: ^6.28.0
version: 6.30.3(react-dom@18.3.1(react@18.3.1))(react@18.3.1) version: 6.30.3(react-dom@18.3.1(react@18.3.1))(react@18.3.1)
react-virtuoso:
specifier: ^4.18.7
version: 4.18.7(react-dom@18.3.1(react@18.3.1))(react@18.3.1)
zustand: zustand:
specifier: ^5.0.1 specifier: ^5.0.1
version: 5.0.12(@types/react@18.3.28)(react@18.3.1)(use-sync-external-store@1.6.0(react@18.3.1)) version: 5.0.12(@types/react@18.3.28)(react@18.3.1)(use-sync-external-store@1.6.0(react@18.3.1))
@@ -1915,6 +1915,15 @@ packages:
resolution: {integrity: sha512-4BAffykYOgO+5nzBWYwE3W90sBgLJoUPRWWcL8wlyiM8IB8ipJz3UMJ9KXQd1RKQXpKp8Tutn80HZtWsu2u76w==} resolution: {integrity: sha512-4BAffykYOgO+5nzBWYwE3W90sBgLJoUPRWWcL8wlyiM8IB8ipJz3UMJ9KXQd1RKQXpKp8Tutn80HZtWsu2u76w==}
engines: {node: '>=10'} engines: {node: '>=10'}
'@tanstack/react-virtual@3.14.2':
resolution: {integrity: sha512-IpWnmCLvuymRfeeLNVXIzNEYBFLpd3drVIS91sqV78VTZFyldlChkOocZRCPp1B+Wnk09bcLNme8WaMU/9/9bQ==}
peerDependencies:
react: ^16.8.0 || ^17.0.0 || ^18.0.0 || ^19.0.0
react-dom: ^16.8.0 || ^17.0.0 || ^18.0.0 || ^19.0.0
'@tanstack/virtual-core@3.17.0':
resolution: {integrity: sha512-gOxY/hFkPh/XQYhnThBHzkbkX3Ed+z/iushyz+R+JAr213aXxUDgQoTgTdrDpBSRsjFM73P/KfUyWmaF9WHMkQ==}
'@testing-library/react-native@12.9.0': '@testing-library/react-native@12.9.0':
resolution: {integrity: sha512-wIn/lB1FjV2N4Q7i9PWVRck3Ehwq5pkhAef5X5/bmQ78J/NoOsGbVY2/DG5Y9Lxw+RfE+GvSEh/fe5Tz6sKSvw==} resolution: {integrity: sha512-wIn/lB1FjV2N4Q7i9PWVRck3Ehwq5pkhAef5X5/bmQ78J/NoOsGbVY2/DG5Y9Lxw+RfE+GvSEh/fe5Tz6sKSvw==}
deprecated: React Native Testing Library v12 is no longer maintained. Please upgrade to v13 or v14. deprecated: React Native Testing Library v12 is no longer maintained. Please upgrade to v13 or v14.
@@ -5479,12 +5488,6 @@ packages:
peerDependencies: peerDependencies:
react: ^18.3.1 react: ^18.3.1
react-virtuoso@4.18.7:
resolution: {integrity: sha512-xNF5zDGEEIMB7cKwcen/pLig0YDf6OnfFrVgKFa7sHPf9fRem0CaLshyObbBcP88jzn0enavL39EgplgdyT21g==}
peerDependencies:
react: '>=16 || >=17 || >= 18 || >= 19'
react-dom: '>=16 || >=17 || >= 18 || >=19'
react@18.3.1: react@18.3.1:
resolution: {integrity: sha512-wS+hAgJShR0KhEvPJArfuPVN1+Hz1t0Y6n5jLrGQbkb4urgPE/0Rve+1kMB1v/oWgHgm4WIcV+i7F2pTVj+2iQ==} resolution: {integrity: sha512-wS+hAgJShR0KhEvPJArfuPVN1+Hz1t0Y6n5jLrGQbkb4urgPE/0Rve+1kMB1v/oWgHgm4WIcV+i7F2pTVj+2iQ==}
engines: {node: '>=0.10.0'} engines: {node: '>=0.10.0'}
@@ -8760,6 +8763,14 @@ snapshots:
dependencies: dependencies:
defer-to-connect: 2.0.1 defer-to-connect: 2.0.1
'@tanstack/react-virtual@3.14.2(react-dom@18.3.1(react@18.3.1))(react@18.3.1)':
dependencies:
'@tanstack/virtual-core': 3.17.0
react: 18.3.1
react-dom: 18.3.1(react@18.3.1)
'@tanstack/virtual-core@3.17.0': {}
'@testing-library/react-native@12.9.0(react-native@0.76.9(@babel/core@7.29.0)(@babel/preset-env@7.29.2(@babel/core@7.29.0))(@types/react@18.3.28)(encoding@0.1.13)(react@18.3.1))(react-test-renderer@18.3.1(react@18.3.1))(react@18.3.1)': '@testing-library/react-native@12.9.0(react-native@0.76.9(@babel/core@7.29.0)(@babel/preset-env@7.29.2(@babel/core@7.29.0))(@types/react@18.3.28)(encoding@0.1.13)(react@18.3.1))(react-test-renderer@18.3.1(react@18.3.1))(react@18.3.1)':
dependencies: dependencies:
jest-matcher-utils: 29.7.0 jest-matcher-utils: 29.7.0
@@ -12972,11 +12983,6 @@ snapshots:
react-shallow-renderer: 16.15.0(react@18.3.1) react-shallow-renderer: 16.15.0(react@18.3.1)
scheduler: 0.23.2 scheduler: 0.23.2
react-virtuoso@4.18.7(react-dom@18.3.1(react@18.3.1))(react@18.3.1):
dependencies:
react: 18.3.1
react-dom: 18.3.1(react@18.3.1)
react@18.3.1: react@18.3.1:
dependencies: dependencies:
loose-envify: 1.4.0 loose-envify: 1.4.0
+244
View File
@@ -0,0 +1,244 @@
#!/usr/bin/env bash
#
# Bootstrap a fresh netralax.de VPS so it can host the Supabase + LiveKit stack.
#
# COPY THIS SCRIPT TO THE NEW SERVER AND RUN IT THERE as root (or via sudo):
# scp scripts/migrate/01-bootstrap-new-server.sh debian@141.95.34.204:/tmp/
# ssh debian@141.95.34.204 'sudo bash /tmp/01-bootstrap-new-server.sh'
#
# It is idempotent: re-running it only fills in what is missing. It installs
# Docker CE + the compose plugin, opens the firewall, clones supabase/supabase,
# prepares /opt/livekit, installs Caddy, creates the update host + deploy user,
# and writes placeholder config. It NEVER fabricates secret values — those you
# copy from the old server (see the NEXT STEPS block it prints at the end).
set -euo pipefail
# --- must run as root ------------------------------------------------------
if [[ "${EUID}" -ne 0 ]]; then
echo "this script must run as root (use: sudo bash $0)" >&2
exit 1
fi
SUPABASE_DIR="/opt/supabase"
LIVEKIT_DIR="/opt/livekit"
UPDATES_DIR="/var/www/updates/windows"
DEPLOY_USER="chatapp-deploy"
log() { echo "==> $*"; }
# --- base packages ---------------------------------------------------------
log "updating apt and installing base packages"
export DEBIAN_FRONTEND=noninteractive
apt-get update -y
apt-get install -y \
ca-certificates curl gnupg lsb-release git ufw rsync apt-transport-https
# --- Docker CE + compose plugin -------------------------------------------
if command -v docker >/dev/null 2>&1 && docker compose version >/dev/null 2>&1; then
log "docker + compose plugin already installed — skipping"
else
log "installing Docker CE + compose plugin (official repo)"
install -m 0755 -d /etc/apt/keyrings
if [[ ! -f /etc/apt/keyrings/docker.gpg ]]; then
curl -fsSL https://download.docker.com/linux/debian/gpg \
| gpg --dearmor -o /etc/apt/keyrings/docker.gpg
chmod a+r /etc/apt/keyrings/docker.gpg
fi
. /etc/os-release
echo \
"deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.gpg] \
https://download.docker.com/linux/${ID} ${VERSION_CODENAME} stable" \
> /etc/apt/sources.list.d/docker.list
apt-get update -y
apt-get install -y \
docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin
systemctl enable --now docker
fi
# Let the login user run docker/compose without sudo (effective on next login).
usermod -aG docker "${SUDO_USER:-debian}" || true
# --- firewall (ufw) --------------------------------------------------------
# Media + TURN ports bypass Caddy entirely and MUST be open or calls have no A/V.
log "configuring ufw"
ufw allow 22/tcp comment 'ssh'
ufw allow 80/tcp comment 'http (caddy / lets encrypt)'
ufw allow 443/tcp comment 'https (caddy)'
ufw allow 7880/tcp comment 'livekit signaling ws (behind caddy)'
ufw allow 7881/tcp comment 'livekit rtc tcp fallback'
ufw allow 50000:50100/udp comment 'livekit rtc udp'
ufw allow 3478/tcp comment 'coturn'
ufw allow 3478/udp comment 'coturn'
ufw allow 5349/tcp comment 'coturn turns (tls)'
ufw allow 50200:50300/udp comment 'coturn turn relay'
# Enable non-interactively (idempotent — re-enabling is a no-op).
ufw --force enable
ufw status verbose || true
# --- Supabase (clone upstream, prepare .env) -------------------------------
if [[ -d "${SUPABASE_DIR}/.git" || -f "${SUPABASE_DIR}/docker-compose.yml" ]]; then
log "${SUPABASE_DIR} already populated — skipping clone"
else
log "cloning supabase/supabase into a temp dir and laying out ${SUPABASE_DIR}"
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/supabase/supabase "${tmp}/supabase"
mkdir -p "${SUPABASE_DIR}"
# The runnable self-hosted stack lives in supabase/docker.
cp -r "${tmp}/supabase/docker/." "${SUPABASE_DIR}/"
rm -rf "${tmp}"
fi
# Prepare .env from the example WITHOUT inventing secrets.
#
# IMPORTANT: Supabase's upstream .env.example does NOT ship blank secrets — it
# ships well-known PUBLIC default values (JWT_SECRET=your-super-secret..., the
# matching default ANON_KEY/SERVICE_ROLE_KEY, POSTGRES_PASSWORD, etc.). Booting
# with those is both a security hole AND wrong: the baked anon key in installed
# clients is signed with the OLD server's JWT_SECRET, so a default secret makes
# the gateway reject every token and drop all sessions — silently. So we
# OVERWRITE the security-critical keys with a loud sentinel that fails fast if
# someone forgets to fill them from the old server.
SENTINEL="__COPY_FROM_OLD_SERVER__"
CRIT_KEYS=(POSTGRES_PASSWORD JWT_SECRET ANON_KEY SERVICE_ROLE_KEY \
SECRET_KEY_BASE VAULT_ENC_KEY DASHBOARD_PASSWORD)
if [[ -f "${SUPABASE_DIR}/.env" ]]; then
log "${SUPABASE_DIR}/.env already exists — leaving it untouched"
elif [[ -f "${SUPABASE_DIR}/.env.example" ]]; then
cp "${SUPABASE_DIR}/.env.example" "${SUPABASE_DIR}/.env"
for k in "${CRIT_KEYS[@]}"; do
sed -i "s|^${k}=.*|${k}=${SENTINEL}|" "${SUPABASE_DIR}/.env" || true
done
log "wrote ${SUPABASE_DIR}/.env — critical secrets set to ${SENTINEL}."
log "These are NOT blank by default upstream; you MUST copy the real values"
log "1:1 from the OLD server's /opt/supabase/.env (esp. JWT_SECRET + VAPID)."
else
log "WARNING: no .env.example found in ${SUPABASE_DIR}; create .env by hand"
fi
# --- LiveKit dir -----------------------------------------------------------
log "preparing ${LIVEKIT_DIR}"
mkdir -p "${LIVEKIT_DIR}"
if [[ ! -f "${LIVEKIT_DIR}/livekit.yaml" ]]; then
cat > "${LIVEKIT_DIR}/livekit.yaml" <<'YAML'
# PLACEHOLDER — replace with infra/livekit/livekit.prod.yaml.example contents.
# Prod config MUST set rtc.use_external_ip: true and must NOT hardcode
# node_ip: 127.0.0.1 (that is dev-only). Fill the keys: block with the SAME
# API key/secret as LIVEKIT_API_KEY / LIVEKIT_API_SECRET in /opt/supabase/.env.
YAML
log "wrote placeholder ${LIVEKIT_DIR}/livekit.yaml"
fi
if [[ ! -f "${LIVEKIT_DIR}/coturn.conf" ]]; then
cat > "${LIVEKIT_DIR}/coturn.conf" <<'CONF'
# PLACEHOLDER — replace with infra/livekit/coturn.prod.conf.example contents.
# Set external-ip to this VPS's public IP, point cert/pkey at the TLS cert for
# turn.netralax.de, and set a real lt-cred-mech user/password.
CONF
log "wrote placeholder ${LIVEKIT_DIR}/coturn.conf"
fi
if [[ ! -f "${LIVEKIT_DIR}/docker-compose.yml" ]]; then
cat > "${LIVEKIT_DIR}/docker-compose.yml" <<'YAML'
# PLACEHOLDER — replace with infra/livekit/docker-compose.prod.yml.example.
# The dev infra/livekit/docker-compose.yml is NOT suitable for prod (coturn runs
# with --no-tls, no 5349, no cert). The prod compose uses network_mode: host,
# mounts ./livekit.yaml + ./coturn.conf, runs coturn with -c turnserver.conf,
# and mounts /etc/letsencrypt for the turn.netralax.de TURNS cert.
YAML
log "wrote placeholder ${LIVEKIT_DIR}/docker-compose.yml"
fi
# --- Caddy (official apt repo) --------------------------------------------
if command -v caddy >/dev/null 2>&1; then
log "caddy already installed — skipping"
else
log "installing Caddy (official repo)"
curl -1sLf 'https://dl.cloudsmith.io/public/caddy/stable/gpg.key' \
| gpg --dearmor -o /usr/share/keyrings/caddy-stable-archive-keyring.gpg
curl -1sLf 'https://dl.cloudsmith.io/public/caddy/stable/debian.deb.txt' \
> /etc/apt/sources.list.d/caddy-stable.list
apt-get update -y
apt-get install -y caddy
systemctl enable caddy
fi
# Write a placeholder Caddyfile if none exists (do not clobber a real one).
if [[ ! -s /etc/caddy/Caddyfile ]] || grep -q 'PLACEHOLDER' /etc/caddy/Caddyfile 2>/dev/null; then
cat > /etc/caddy/Caddyfile <<'CADDY'
# PLACEHOLDER Caddyfile — replace with infra/caddy/Caddyfile from the repo.
# Serve the .de vhosts now; add the legacy .cloud vhosts only at cutover (after
# the .cloud DNS is repointed) so they keep already-installed clients working:
# supabase.netralax.de { reverse_proxy localhost:8000 }
# livekit.netralax.de { reverse_proxy localhost:7880 }
# update.netralax.de { root * /var/www/updates # NOT .../windows — see Caddyfile
# file_server }
CADDY
log "wrote placeholder /etc/caddy/Caddyfile"
fi
# --- update host + deploy user --------------------------------------------
log "preparing update host at ${UPDATES_DIR}"
mkdir -p "${UPDATES_DIR}"
if id "${DEPLOY_USER}" >/dev/null 2>&1; then
log "user ${DEPLOY_USER} already exists — skipping"
else
log "creating deploy user ${DEPLOY_USER}"
useradd --create-home --shell /bin/bash "${DEPLOY_USER}"
mkdir -p "/home/${DEPLOY_USER}/.ssh"
chmod 700 "/home/${DEPLOY_USER}/.ssh"
touch "/home/${DEPLOY_USER}/.ssh/authorized_keys"
chmod 600 "/home/${DEPLOY_USER}/.ssh/authorized_keys"
chown -R "${DEPLOY_USER}:${DEPLOY_USER}" "/home/${DEPLOY_USER}/.ssh"
fi
# Let the deploy user write release artifacts.
chown -R "${DEPLOY_USER}:${DEPLOY_USER}" "${UPDATES_DIR}"
# --- next steps ------------------------------------------------------------
cat <<EOF
============================================================================
BOOTSTRAP DONE — manual NEXT STEPS (this script invents NO secrets):
============================================================================
1. Fill ${SUPABASE_DIR}/.env. Copy these 1:1 from the OLD server's
/opt/supabase/.env so baked-in client tokens + push keep working:
POSTGRES_PASSWORD, JWT_SECRET, ANON_KEY, SERVICE_ROLE_KEY,
SECRET_KEY_BASE, VAULT_ENC_KEY, PG_META_CRYPTO_KEY,
SMTP_*, VAPID_PUBLIC_KEY, VAPID_PRIVATE_KEY, VAPID_SUBJECT,
PUSH_FANOUT_SHARED_SECRET, LIVEKIT_API_KEY, LIVEKIT_API_SECRET.
Set these to the NEW host:
SITE_URL / API_EXTERNAL_URL / SUPABASE_PUBLIC_URL = https://supabase.netralax.de
SUPABASE_URL = https://supabase.netralax.de
LIVEKIT_URL = wss://livekit.netralax.de
ADDITIONAL_REDIRECT_URLS must include (comma-separated, no spaces):
chatapp://auth/callback,netralax://auth/callback,
https://supabase.netralax.de,https://supabase.netralax.cloud
2. Drop the real LiveKit + coturn config in ${LIVEKIT_DIR}:
docker-compose.yml <- infra/livekit/docker-compose.prod.yml.example
livekit.yaml <- infra/livekit/livekit.prod.yaml.example
coturn.conf <- infra/livekit/coturn.prod.conf.example
Set rtc.use_external_ip: true, NO node_ip: 127.0.0.1, coturn external-ip
= this VPS's public IP, and a TLS cert for turn.netralax.de.
The LiveKit keys: block MUST match LIVEKIT_API_KEY/SECRET in .env.
(The on-server filenames are livekit.yaml / coturn.conf — same names
rotate-livekit-keys.sh expects.)
3. Place the real Caddyfile:
cp infra/caddy/Caddyfile /etc/caddy/Caddyfile && systemctl reload caddy
(serves both .de and .cloud vhosts).
4. Bring the Supabase DB up ONCE so init scripts create the roles, then
restore data from the laptop:
cd ${SUPABASE_DIR} && docker compose up -d db && sleep 20
# then on the laptop: ./scripts/migrate/02-migrate-data.sh
5. Repoint DNS A-records to THIS VPS's IP for BOTH domains:
supabase.netralax.de / .cloud, livekit.netralax.de / .cloud,
turn.netralax.de, update.netralax.de / .cloud.
6. Add the chatapp-deploy public key to
/home/${DEPLOY_USER}/.ssh/authorized_keys
and mirror electron-updater artifacts (latest.yml, *.exe, changelog.json)
under ${UPDATES_DIR} so BOTH update.netralax.de and .cloud serve them.
============================================================================
EOF
+207
View File
@@ -0,0 +1,207 @@
#!/usr/bin/env bash
#
# One-time data move: OLD (.cloud) -> NEW (.de). Run this FROM THE DEV LAPTOP
# (Linux / macOS / WSL), not on a server. It:
# 1. pre-flight checks both stacks are reachable and the DB containers are up,
# 2. streams a full-cluster pg_dumpall from OLD straight into NEW (psql),
# 3. rsyncs ${SUPABASE_DIR}/volumes/storage from OLD to NEW.
#
# Usage:
# ./scripts/migrate/02-migrate-data.sh # interactive, asks to confirm
# ./scripts/migrate/02-migrate-data.sh --check # pre-flight only, no changes
# FORCE=1 ./scripts/migrate/02-migrate-data.sh # skip the confirm prompt
#
# BEFORE running: put the OLD app into maintenance / freeze writes, and make
# sure ${SUPABASE_DIR}/.env on NEW already has the SAME POSTGRES_PASSWORD and
# JWT_SECRET as OLD, and that NEW's db container has been started once so the
# Supabase init scripts created the roles (see bootstrap NEXT STEPS step 4).
set -euo pipefail
here="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
source "${here}/config.sh"
require_new_host
mode="${1:-}"
log() { echo "==> $*"; }
# --- pre-flight ------------------------------------------------------------
log "pre-flight: checking SSH reachability"
old_remote 'echo ok' >/dev/null || { echo "cannot ssh to OLD (${OLD_SSH})" >&2; exit 1; }
new_remote 'echo ok' >/dev/null || { echo "cannot ssh to NEW (${NEW_SSH})" >&2; exit 1; }
log "pre-flight: checking OLD Supabase db is up"
old_remote "cd ${SUPABASE_DIR} && docker compose exec -T db pg_isready -U postgres" \
|| { echo "OLD db not ready — start the stack first" >&2; exit 1; }
log "pre-flight: checking NEW Supabase db is up (must be initialized once)"
new_remote "cd ${SUPABASE_DIR} && docker compose exec -T db pg_isready -U postgres" \
|| { echo "NEW db not ready — run 'docker compose up -d db' on NEW first" >&2; exit 1; }
log "pre-flight: checking NEW storage volume dir exists"
new_remote "test -d ${SUPABASE_DIR}/volumes/storage || mkdir -p ${SUPABASE_DIR}/volumes/storage"
if [[ "${mode}" == "--check" ]]; then
log "pre-flight OK — --check requested, stopping before any changes."
exit 0
fi
# --- loud confirm ----------------------------------------------------------
cat <<EOF
----------------------------------------------------------------------------
ABOUT TO MIGRATE DATA: OLD ${OLD_SSH} -> NEW ${NEW_SSH}
----------------------------------------------------------------------------
This will:
* pg_dumpall the WHOLE OLD cluster and restore it into the NEW db
(DROP/CREATE objects on NEW via --clean --if-exists),
* rsync ${SUPABASE_DIR}/volumes/storage OLD -> NEW with --delete
(the NEW storage dir becomes an EXACT mirror of OLD).
MAKE SURE FIRST:
* the OLD app is in MAINTENANCE / writes are FROZEN (no new uploads,
no new rows) so DB + storage stay consistent,
* NEW /opt/supabase/.env already has the OLD POSTGRES_PASSWORD + JWT_SECRET,
* you have a backup / you can roll DNS back to the OLD VPS.
----------------------------------------------------------------------------
EOF
if [[ "${FORCE:-0}" != "1" ]]; then
read -rp "Type 'migrate' to proceed: " confirm
if [[ "${confirm}" != "migrate" ]]; then
echo "aborted — nothing changed."
exit 1
fi
fi
# --- 1) Postgres: full-cluster dump OLD -> restore NEW ---------------------
# pg_dumpall (not pg_dump) carries the ROLE definitions + password hashes, so
# with an identical POSTGRES_PASSWORD on both hosts the restored roles line up
# with what the services use. ON_ERROR_STOP=0 because pg_dumpall will try to
# CREATE ROLE supabase_admin/postgres etc. that already exist on the freshly
# initialized NEW cluster — those 'already exists' errors are harmless.
#
# ALTERNATIVE (highest fidelity): if both servers run the SAME Postgres image
# tag, a cold volume copy avoids logical-restore-over-initialized-cluster
# fragility entirely: stop both DB containers, rsync ${SUPABASE_DIR}/volumes/db
# OLD -> NEW, start both again. Use that if the scan below keeps flagging errors.
log "dumping OLD cluster and restoring into NEW (streamed over SSH)"
log "this can take a while; harmless 'already exists' errors are expected."
restore_log="$(mktemp)"
set +e
old_remote "cd ${SUPABASE_DIR} && docker compose exec -T db pg_dumpall -U postgres --clean --if-exists" \
| new_remote "cd ${SUPABASE_DIR} && docker compose exec -T db psql -U postgres -d postgres -v ON_ERROR_STOP=0" \
2>&1 | tee "${restore_log}"
set -e
# ON_ERROR_STOP=0 keeps the restore going past harmless 'already exists', but it
# ALSO swallows genuine failures (FK/constraint/ownership/extension errors)
# that would leave a partially-restored DB looking successful. Surface any
# non-benign ERROR/FATAL/PANIC and refuse to continue to the storage rsync.
real_errors="$(grep -E 'ERROR:|FATAL:|PANIC:' "${restore_log}" 2>/dev/null \
| grep -Eiv 'already exists|cannot drop the currently open database|is being accessed by other users|must be member of role|role .* cannot be dropped|current transaction is aborted' \
|| true)"
if [[ -n "${real_errors}" ]]; then
echo >&2
echo "!!! Non-benign errors during restore (full log: ${restore_log}):" >&2
echo "${real_errors}" | head -n 50 >&2
if [[ "${FORCE_RESTORE_OK:-0}" != "1" ]]; then
echo "Aborting BEFORE the storage rsync. Inspect/fix and re-run, or consider" >&2
echo "the cold volume-copy path. Override with FORCE_RESTORE_OK=1 only if you" >&2
echo "are certain these are harmless." >&2
exit 1
fi
log "FORCE_RESTORE_OK=1 — continuing despite the errors above."
else
log "restore output scanned: no non-benign errors found."
fi
log "DO NOT run push-migrations.sh: all migrations are already in the dump."
# --- 2) Storage objects: rsync OLD -> NEW ----------------------------------
# Object bytes live on the bind-mounted volume; their metadata rows came with
# the dump above. -aHAX keeps perms/hardlinks/ACLs/xattrs; --delete makes NEW an
# exact mirror (safe only because writes are frozen). Trailing slashes matter.
log "rsyncing storage volume OLD -> NEW (server-to-server via SSH)"
if old_remote "command -v rsync >/dev/null 2>&1"; then
# Direct server-to-server: the OLD host pushes to NEW. Needs the OLD host to
# be able to ssh to NEW (key in OLD ~/.ssh, NEW in known_hosts).
old_remote "sudo rsync -aHAX --numeric-ids --delete \
-e 'ssh -o StrictHostKeyChecking=accept-new' \
${SUPABASE_DIR}/volumes/storage/ ${NEW_SSH}:${SUPABASE_DIR}/volumes/storage/" \
|| {
log "direct server-to-server rsync failed — falling back to two-hop via laptop"
stage="$(mktemp -d)"
log "staging into ${stage}"
# shellcheck disable=SC2086
rsync -aHAX --numeric-ids -e "ssh ${SSH_OPTS}" \
"${OLD_SSH}:${SUPABASE_DIR}/volumes/storage/" "${stage}/"
# shellcheck disable=SC2086
rsync -aHAX --numeric-ids --delete -e "ssh ${SSH_OPTS}" \
"${stage}/" "${NEW_SSH}:${SUPABASE_DIR}/volumes/storage/"
rm -rf "${stage}"
}
else
log "rsync missing on OLD — using two-hop via laptop"
stage="$(mktemp -d)"
log "staging into ${stage}"
# shellcheck disable=SC2086
rsync -aHAX --numeric-ids -e "ssh ${SSH_OPTS}" \
"${OLD_SSH}:${SUPABASE_DIR}/volumes/storage/" "${stage}/"
# shellcheck disable=SC2086
rsync -aHAX --numeric-ids --delete -e "ssh ${SSH_OPTS}" \
"${stage}/" "${NEW_SSH}:${SUPABASE_DIR}/volumes/storage/"
rm -rf "${stage}"
fi
# --- 3) Restart NEW stack so every service reconnects to the new data ------
log "restarting the NEW Supabase stack (down + up -d)"
new_remote "cd ${SUPABASE_DIR} && docker compose down && docker compose up -d"
# --- 4) Row-count parity check OLD vs NEW (load-bearing tables) -------------
# A users/objects-only check can miss partial loss in messages/members/etc.,
# so compare the tables the app actually depends on. Non-fatal (table names can
# legitimately vary), but a mismatch on auth.users / public.messages is a red
# flag — do NOT cut over until it is understood.
log "waiting for NEW db to accept connections, then checking row-count parity"
for _ in $(seq 1 30); do
new_remote "cd ${SUPABASE_DIR} && docker compose exec -T db pg_isready -U postgres" >/dev/null 2>&1 && break
sleep 2
done
count_on() { # $1=old|new $2=table
local q="select count(*) from $2;"
local runner=old_remote
[[ "$1" == "new" ]] && runner=new_remote
"${runner}" "cd ${SUPABASE_DIR} && docker compose exec -T db psql -U postgres -d postgres -tAc \"${q}\"" 2>/dev/null | tr -d '[:space:]'
}
parity_fail=0
for t in auth.users auth.identities public.profiles public.messages \
public.conversation_members storage.objects; do
o="$(count_on old "$t" 2>/dev/null || echo '?')"
n="$(count_on new "$t" 2>/dev/null || echo '?')"
if [[ -n "$o" && "$o" == "$n" ]]; then
log " OK ${t}: ${o}"
else
log " MISMATCH ${t}: OLD=${o:-?} NEW=${n:-?}"
parity_fail=1
fi
done
[[ "${parity_fail}" == "1" ]] && log "⚠ row-count mismatch — investigate BEFORE cutover."
cat <<EOF
============================================================================
DATA MIGRATION DONE.
============================================================================
Verify on NEW:
* users can log in (existing baked anon JWT must be accepted),
* a known storage object downloads via
https://supabase.netralax.de/storage/v1/object/...,
* realtime + push still work.
If storage objects 403 due to ownership, on NEW run:
cd ${SUPABASE_DIR} && docker compose restart storage imgproxy
Do NOT decommission the OLD VPS until the .cloud DNS A-records point at NEW
and old clients have had a chance to auto-update.
============================================================================
EOF
+189
View File
@@ -0,0 +1,189 @@
#!/usr/bin/env node
// 03-copy-secrets.mjs — merge the OLD server's /opt/supabase/.env into the NEW
// server's .env, keeping every secret byte-identical, then force the public-URL
// / redirect vars to the netralax.de host. Run from the dev laptop.
//
// WHY Node (not bash/sed): secret values (JWT tokens, base64 keys, SMTP
// passwords) contain characters that wreck shell/sed escaping. Here the values
// only ever travel over SSH stdin/stdout and are handled as plain JS strings —
// never interpolated into a shell command. Nothing is written to the laptop disk.
//
// MERGE SEMANTICS (loss-free):
// - base = the NEW .env (fresh upstream structure + comments + new-only keys)
// - for every key that exists on BOTH sides -> take the OLD value
// - for every key that exists ONLY on OLD -> append it (this is how the
// custom edge secrets VAPID_*/PUSH_FANOUT_SHARED_SECRET/LIVEKIT_API_*
// survive — they are not in the fresh upstream .env)
// - keys ONLY on NEW -> keep their fresh default
// - finally, the OVERRIDES below are upserted (public host = .de)
//
// Usage:
// node scripts/migrate/03-copy-secrets.mjs --check # show plan, change nothing
// node scripts/migrate/03-copy-secrets.mjs # back up + apply on NEW
//
// Pre-req: SSH works to BOTH hosts (prox@OLD, debian@NEW) and NEW's .env exists
// (bootstrap step done). OLD/NEW are read from scripts/migrate/config.sh.
import { execFileSync } from 'node:child_process';
import { readFileSync } from 'node:fs';
import { fileURLToPath } from 'node:url';
import { dirname, join } from 'node:path';
import { createHash } from 'node:crypto';
const here = dirname(fileURLToPath(import.meta.url));
// --- read hosts from config.sh (single source of truth) --------------------
const cfg = readFileSync(join(here, 'config.sh'), 'utf8');
const cfgVal = (name) => {
const m = cfg.match(new RegExp(`^export ${name}="([^"]*)"`, 'm'));
if (!m) throw new Error(`could not find ${name} in config.sh`);
return m[1];
};
const OLD_USER = cfgVal('OLD_USER');
const OLD_HOST = cfgVal('OLD_HOST');
const NEW_USER = cfgVal('NEW_USER');
const NEW_HOST = cfgVal('NEW_HOST');
const SUPABASE_DIR = cfgVal('SUPABASE_DIR');
const ENV_PATH = `${SUPABASE_DIR}/.env`;
const OLD_SSH = `${OLD_USER}@${OLD_HOST}`;
const NEW_SSH = `${NEW_USER}@${NEW_HOST}`;
const SSH_OPTS = ['-o', 'StrictHostKeyChecking=accept-new'];
if (NEW_HOST === '__NETRALAX_DE_SERVER_IP__' || !NEW_HOST) {
console.error('NEW_HOST is still the placeholder — edit scripts/migrate/config.sh first.');
process.exit(1);
}
// --- public-host overrides (forced to .de AFTER the merge) -----------------
// NOTE: SUPABASE_URL is deliberately NOT overridden — for the edge-runtime it
// is the INTERNAL gateway URL and is handled by the compose env in §8, not here.
const NEW_SITE = 'https://supabase.netralax.de';
const NEW_LIVEKIT = 'wss://livekit.netralax.de';
const OVERRIDES = {
SITE_URL: NEW_SITE,
API_EXTERNAL_URL: NEW_SITE,
SUPABASE_PUBLIC_URL: NEW_SITE,
LIVEKIT_URL: NEW_LIVEKIT,
ADDITIONAL_REDIRECT_URLS:
'chatapp://auth/callback,netralax://auth/callback,' +
'https://supabase.netralax.de,https://supabase.netralax.cloud',
};
// Continuity-critical keys: their value MUST end up identical to OLD.
const CRITICAL = [
'JWT_SECRET', 'ANON_KEY', 'SERVICE_ROLE_KEY', 'POSTGRES_PASSWORD',
'VAPID_PUBLIC_KEY', 'VAPID_PRIVATE_KEY', 'PUSH_FANOUT_SHARED_SECRET',
'LIVEKIT_API_KEY', 'LIVEKIT_API_SECRET',
];
const check = process.argv.includes('--check');
// --- ssh helpers (values flow via stdio, never via argv) -------------------
function ssh(target, remoteCmd, input) {
return execFileSync('ssh', [...SSH_OPTS, target, remoteCmd], {
encoding: 'utf8',
input: input ?? undefined,
maxBuffer: 16 * 1024 * 1024,
});
}
const readOldEnv = () => ssh(OLD_SSH, `sudo cat ${ENV_PATH} 2>/dev/null || cat ${ENV_PATH}`);
const readNewEnv = () => ssh(NEW_SSH, `sudo cat ${ENV_PATH}`);
// --- env parsing (split on FIRST '='; keep comments/blank lines as raw) -----
const KEY_RE = /^([A-Za-z_][A-Za-z0-9_]*)=(.*)$/;
function parse(text) {
const map = new Map();
for (const line of text.split('\n')) {
const m = line.match(KEY_RE);
if (m) map.set(m[1], m[2]);
}
return map;
}
const sha = (s) => createHash('sha256').update(s ?? '').digest('hex').slice(0, 12);
// --- main ------------------------------------------------------------------
console.log(`OLD: ${OLD_SSH} NEW: ${NEW_SSH} file: ${ENV_PATH}\n`);
let oldText, newText;
try { oldText = readOldEnv(); } catch (e) {
console.error(`Failed to read OLD .env via ssh ${OLD_SSH}.\n${e.message}`);
process.exit(1);
}
try { newText = readNewEnv(); } catch (e) {
console.error(`Failed to read NEW .env via ssh ${NEW_SSH} (bootstrap done?).\n${e.message}`);
process.exit(1);
}
const oldMap = parse(oldText);
const newMap = parse(newText);
const onlyOld = [...oldMap.keys()].filter((k) => !newMap.has(k)).sort();
const onlyNew = [...newMap.keys()].filter((k) => !oldMap.has(k)).sort();
const shared = [...oldMap.keys()].filter((k) => newMap.has(k)).sort();
console.log(`shared keys (value taken from OLD): ${shared.length}`);
console.log(`OLD-only keys (appended — incl. custom edge secrets): ${onlyOld.length}`);
onlyOld.forEach((k) => console.log(` + ${k}`));
console.log(`NEW-only keys (kept at fresh default): ${onlyNew.length}`);
onlyNew.forEach((k) => console.log(` . ${k}`));
console.log(`\noverrides forced to the .de host:`);
for (const [k, v] of Object.entries(OVERRIDES)) console.log(` ${k}=${v}`);
// sanity: warn if a continuity-critical key is missing on OLD
const missingCrit = CRITICAL.filter((k) => !oldMap.has(k));
if (missingCrit.length) {
console.log(`\n⚠ NOTE: these critical keys are absent on OLD (verify they aren't named differently): ${missingCrit.join(', ')}`);
}
// --- build merged content (preserve NEW order/comments) --------------------
const used = new Set();
let lines = newText.split('\n').map((line) => {
const m = line.match(KEY_RE);
if (m && oldMap.has(m[1])) { used.add(m[1]); return `${m[1]}=${oldMap.get(m[1])}`; }
return line;
});
// append OLD-only keys
if (onlyOld.length) {
if (lines.length && lines[lines.length - 1] !== '') lines.push('');
lines.push('# --- merged from OLD server (keys not present in fresh upstream .env) ---');
for (const k of onlyOld) { lines.push(`${k}=${oldMap.get(k)}`); used.add(k); }
}
// upsert overrides
for (const [k, v] of Object.entries(OVERRIDES)) {
let hit = false;
lines = lines.map((line) => {
const m = line.match(KEY_RE);
if (m && m[1] === k) { hit = true; return `${k}=${v}`; }
return line;
});
if (!hit) lines.push(`${k}=${v}`);
}
const merged = lines.join('\n');
if (check) {
console.log('\n--check: nothing written. Re-run without --check to apply.');
process.exit(0);
}
// --- apply on NEW: backup, then write via `sudo tee` (content via stdin) ----
console.log('\nbacking up NEW .env and writing merged result...');
ssh(NEW_SSH, `sudo cp ${ENV_PATH} ${ENV_PATH}.bak.$(date +%s)`);
ssh(NEW_SSH, `sudo tee ${ENV_PATH} > /dev/null`, merged.endsWith('\n') ? merged : merged + '\n');
// --- verify continuity: critical values identical OLD vs NEW ---------------
const newAfter = parse(readNewEnv());
console.log('\nverifying continuity (OLD value == NEW value):');
let fail = 0;
for (const k of CRITICAL) {
if (!oldMap.has(k)) { console.log(` skip ${k} (not on OLD)`); continue; }
const ok = oldMap.get(k) === newAfter.get(k);
console.log(` ${ok ? 'OK ' : 'FAIL'} ${k} (sha ${sha(oldMap.get(k))} vs ${sha(newAfter.get(k))})`);
if (!ok) fail++;
}
console.log('\noverrides now on NEW:');
for (const k of Object.keys(OVERRIDES)) console.log(` ${k}=${newAfter.get(k)}`);
if (fail) {
console.error(`\n${fail} critical key(s) did not match — DO NOT proceed. Restore from the .bak.* backup and investigate.`);
process.exit(1);
}
console.log('\n✓ secrets merged; JWT_SECRET + VAPID + LiveKit keys are identical to OLD. Continue with runbook §6 (LiveKit/coturn) and §5 (data).');
@@ -0,0 +1,127 @@
#!/usr/bin/env python3
# Incident fix (2026-06-02): after the netralax.cloud -> netralax.de move, every
# storage object GET returned HTTP 500 with:
# { "code": "ENODATA", "errno": 61, "message": "The extended attribute does not exist." }
#
# Root cause: the OBJECT BYTES were copied to the new server, but Supabase
# Storage (supabase/storage-api:v1.48.26, file backend) keeps each object's
# response metadata in Linux extended attributes (xattrs) on the version file:
# user.supabase.content-type
# user.supabase.cache-control
# user.supabase.etag
# The migration copy did not preserve xattrs, so storage's getObject() throws
# ENODATA when it reads them. The OLD server is gone, so we cannot re-copy —
# but every value we need is still in the database column storage.objects.metadata
# (mimetype / cacheControl / eTag). This script reconstructs the missing xattrs
# from that column. It is idempotent and only ADDS metadata xattrs; it never
# touches object bytes.
#
# RUN ON THE NEW SERVER (needs /opt/supabase, docker, and root for setxattr):
# sudo python3 04-restore-storage-xattrs.py # all buckets
# sudo python3 04-restore-storage-xattrs.py --dry-run # show, change nothing
# sudo python3 04-restore-storage-xattrs.py --bucket profile-avatars
# After it finishes, no storage restart is needed (xattrs are read per request).
import argparse
import json
import os
import subprocess
import sys
SUPABASE_DIR = "/opt/supabase"
# Single-tenant self-hosted layout: <volume>/stub/stub/<bucket>/<name>/<version-file>
STORAGE_ROOT = os.path.join(SUPABASE_DIR, "volumes/storage/stub/stub")
# DB metadata field -> (xattr name, default when the field is absent)
XATTRS = [
("mimetype", "user.supabase.content-type", "application/octet-stream"),
("cacheControl", "user.supabase.cache-control", "no-cache"),
("eTag", "user.supabase.etag", None), # None default => skip if missing
]
def fetch_objects():
"""Return [(bucket_id, name, metadata_dict), ...] from storage.objects."""
# Tab-separate so object names containing '|' can't break parsing.
query = (
"select bucket_id||chr(9)||name||chr(9)||coalesce(metadata::text,'{}') "
"from storage.objects"
)
raw = subprocess.check_output(
[
"docker", "compose", "exec", "-T", "db",
"psql", "-U", "postgres", "-d", "postgres", "-tAc", query,
],
cwd=SUPABASE_DIR,
).decode()
rows = []
for line in raw.splitlines():
line = line.rstrip("\r")
if not line.strip():
continue
parts = line.split("\t", 2)
if len(parts) < 3:
continue
bucket, name, meta = parts
try:
md = json.loads(meta) if meta else {}
except json.JSONDecodeError:
md = {}
rows.append((bucket, name, md))
return rows
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--bucket", help="only this bucket (e.g. profile-avatars)")
ap.add_argument("--dry-run", action="store_true", help="print, change nothing")
args = ap.parse_args()
objects = fetch_objects()
fixed = missing_dir = missing_file = 0
for bucket, name, md in objects:
if args.bucket and bucket != args.bucket:
continue
objdir = os.path.join(STORAGE_ROOT, bucket, name)
if not os.path.isdir(objdir):
print("NO_DIR ", bucket, name)
missing_dir += 1
continue
version_files = [
os.path.join(objdir, f)
for f in os.listdir(objdir)
if os.path.isfile(os.path.join(objdir, f))
]
if not version_files:
print("NO_FILE ", bucket, name)
missing_file += 1
continue
for path in version_files:
for field, xattr, default in XATTRS:
value = md.get(field, default)
if value is None:
continue
if args.dry_run:
print(f" would set {xattr}={value!r} on {path}")
else:
os.setxattr(path, xattr, str(value).encode())
fixed += 1
print("OK ", bucket, name)
print(
f"\n{'DRY-RUN: would fix' if args.dry_run else 'fixed'} {fixed} file(s); "
f"{missing_dir} missing dir(s), {missing_file} empty object dir(s)."
)
if missing_dir or missing_file:
print(
"NOTE: objects with a missing dir/file have lost their bytes and "
"cannot be recovered from xattrs — those are genuinely gone."
)
if __name__ == "__main__":
if os.geteuid() != 0 and "--dry-run" not in sys.argv:
print("Re-run with sudo (setxattr needs root).", file=sys.stderr)
sys.exit(1)
main()
+132
View File
@@ -0,0 +1,132 @@
# Server-Umzug: netralax.cloud -> netralax.de
Einmalige Migration des selbst gehosteten Chat-Backends vom **alten VPS**
(`46.225.156.249`, `*.netralax.cloud`) auf einen **neuen, leeren VPS**
(`*.netralax.de`). Der neue Server bedient anschliessend **beide** Domains,
damit bereits installierte Desktop-/Mobile-Clients (die alte Hostnamen und den
alten anon-JWT fest eingebaut haben) weiterlaufen, bis sie sich selbst
aktualisieren.
> Diese Skripte sind bewusst getrennt von `scripts/prod/`. `scripts/prod/config.sh`
> kennt nur den jeweils **aktiven** Server; der Umzug braucht **beide** Hosts und
> hat deshalb seine eigene `scripts/migrate/config.sh`.
## Dateien
| Datei | Wo ausführen | Zweck |
|-------|--------------|-------|
| `config.sh` | | Gemeinsame Konfiguration (alter + neuer Host, SSH-Helfer). Wird von den anderen Skripten eingebunden. |
| `01-bootstrap-new-server.sh` | **auf dem neuen VPS** (als root / sudo) | Richtet den leeren Server ein: Docker, ufw, Supabase-Clone, LiveKit-Verzeichnis, Caddy, Update-Host, Deploy-User. |
| `02-migrate-data.sh` | **auf dem Entwickler-Laptop** | Überträgt Postgres-Daten (pg_dumpall) und die Storage-Objekte (rsync) von alt nach neu. |
## Voraussetzungen / Einrichtung (einmalig)
1. **Neue Server-IP — bereits eingetragen.** `scripts/migrate/config.sh` hat
`NEW_HOST="141.95.34.204"` und `NEW_USER="debian"`. (Der `require_new_host`-
Guard greift nur, falls der Platzhalter wieder drinsteht.)
2. **SSH-Zugriff.** Vom Laptop muss `ssh prox@46.225.156.249` (alt) **und**
`ssh debian@141.95.34.204` (neu) ohne Passwort funktionieren:
```
ssh-copy-id prox@46.225.156.249
ssh-copy-id debian@141.95.34.204
```
Für den direkten Storage-Transfer (Server-zu-Server) muss zusätzlich der
**alte** Server per SSH auf den **neuen** zugreifen können. Klappt das nicht,
fällt `02-migrate-data.sh` automatisch auf den Umweg über den Laptop zurück.
3. **Skripte ausführbar machen:**
```
chmod +x scripts/migrate/*.sh
```
## Ablauf (Reihenfolge unbedingt einhalten)
1. **Bootstrap auf dem neuen Server.** Skript hochladen und als root ausführen:
```
scp scripts/migrate/01-bootstrap-new-server.sh debian@141.95.34.204:/tmp/
ssh debian@141.95.34.204 'sudo bash /tmp/01-bootstrap-new-server.sh'
```
Das Skript ist idempotent (mehrfaches Ausführen schadet nicht) und gibt am
Ende einen **NEXT STEPS**-Block aus.
2. **Secrets eintragen.** `/opt/supabase/.env` auf dem neuen Server befüllen.
Diese Werte **1:1 vom alten Server kopieren** (sonst brechen eingebaute
Tokens, Sessions und Web-Push):
`POSTGRES_PASSWORD`, `JWT_SECRET`, `ANON_KEY`, `SERVICE_ROLE_KEY`,
`SECRET_KEY_BASE`, `VAULT_ENC_KEY`, `PG_META_CRYPTO_KEY`, alle `SMTP_*`,
`VAPID_PUBLIC_KEY`, `VAPID_PRIVATE_KEY`, `VAPID_SUBJECT`,
`PUSH_FANOUT_SHARED_SECRET`, `LIVEKIT_API_KEY`, `LIVEKIT_API_SECRET`.
Auf die **neue** Domain zeigen:
`SITE_URL`, `API_EXTERNAL_URL`, `SUPABASE_PUBLIC_URL`, `SUPABASE_URL`
= `https://supabase.netralax.de`, `LIVEKIT_URL` = `wss://livekit.netralax.de`.
`ADDITIONAL_REDIRECT_URLS` (komma-getrennt, **ohne Leerzeichen**) muss
enthalten: `chatapp://auth/callback`, `netralax://auth/callback` sowie
`https://supabase.netralax.de` und `https://supabase.netralax.cloud`.
3. **Server-Konfig platzieren.**
- `infra/livekit/docker-compose.prod.yml.example` -> `/opt/livekit/docker-compose.yml`
(Prod-Compose: host-networking, mountet `livekit.yaml` + `coturn.conf` +
`/etc/letsencrypt`; das Dev-Compose taugt **nicht** für Prod).
- `infra/livekit/livekit.prod.yaml.example` -> `/opt/livekit/livekit.yaml`
(`rtc.use_external_ip: true`, **kein** `node_ip: 127.0.0.1`, `keys:`-Block
identisch zu `LIVEKIT_API_KEY/SECRET` aus der `.env`).
- `infra/livekit/coturn.prod.conf.example` -> `/opt/livekit/coturn.conf`
(`external-ip` = öffentliche IP des neuen VPS, TLS-Cert für
`turn.netralax.de`).
- `infra/caddy/Caddyfile` -> `/etc/caddy/Caddyfile`, danach
`systemctl reload caddy`. Caddy bedient **beide** Domains (.de und .cloud).
4. **Stacks starten DB zuerst einmal hochfahren**, damit die Supabase-Init-
Skripte die Rollen anlegen (vor dem Restore):
```
ssh debian@141.95.34.204 'cd /opt/supabase && docker compose up -d db && sleep 20'
```
5. **Schreibzugriffe auf dem ALTEN System einfrieren** (Wartungsmodus). Sonst
landen während des Umzugs neue Uploads/Zeilen nur auf einer Seite und
DB + Storage werden inkonsistent.
6. **Daten migrieren** (vom Laptop). Erst der Trockenlauf, dann die Migration:
```
./scripts/migrate/02-migrate-data.sh --check # nur Pre-Flight, keine Änderung
./scripts/migrate/02-migrate-data.sh # fragt nach Bestätigung
```
Das Skript dumpt den **gesamten** Cluster per `pg_dumpall` und spielt ihn auf
dem neuen Server ein, danach rsync der Storage-Objekte. `push-migrations.sh`
**nicht** erneut ausführen die Migrationen sind bereits im Dump enthalten.
7. **DNS umstellen.** A-Records für **beide** Domains auf die neue IP zeigen
lassen: `supabase.netralax.de` / `.cloud`, `livekit.netralax.de` / `.cloud`,
`turn.netralax.de`, `update.netralax.de` / `.cloud`.
8. **Update-Artefakte spiegeln.** electron-updater-Dateien (`latest.yml`,
`*.exe`, `changelog.json`) unter `/var/www/updates/windows` ablegen, sodass
**sowohl** `update.netralax.de` **als auch** `update.netralax.cloud` sie
ausliefern. Nur so können alte (.cloud-)Clients die Umstiegs-Version ziehen.
## Sicherheitshinweise
- **`JWT_SECRET`, `ANON_KEY`, `SERVICE_ROLE_KEY`** müssen byteweise identisch
vom alten Server stammen, **bevor** der erste Client den neuen Server trifft
sonst werden alle eingebauten Tokens abgelehnt und alle Sessions fliegen raus.
- **`POSTGRES_PASSWORD`** muss vor dem Restore identisch gesetzt sein, weil der
Dump die Rollen-Passwort-Hashes mitbringt. Sonst können sich die internen
Dienste (auth/rest/storage) nach dem Restore nicht mehr an Postgres anmelden.
- **VAPID-Schlüsselpaar** identisch übernehmen, sonst sind alle bestehenden
Web-Push-Abos ungültig.
- **Medien-/TURN-Ports** müssen in ufw offen sein (7880/7881 tcp, 50000-50100
udp, coturn 3478 tcp+udp, 5349 tcp, 50200-50300 udp) sonst haben Anrufe kein
Audio/Video. Diese Ports laufen **nicht** über Caddy.
- **TURNS auf 5349** braucht ein eigenes TLS-Zertifikat für `turn.netralax.de`
auf der Platte (Pfade in `coturn.conf`) ein reines Caddy-Zertifikat reicht
nicht.
- **Alten VPS nicht abschalten**, bevor die `.cloud`-DNS-Einträge auf den neuen
Server zeigen und alte Clients Zeit zum Auto-Update hatten.
- Beim Restore werden harmlose `already exists`-Fehler für vorhandene Rollen
(`supabase_admin`, `postgres` …) ausgegeben das ist gewollt
(`ON_ERROR_STOP=0`). `02-migrate-data.sh` scannt die Restore-Ausgabe
**automatisch** auf echte `ERROR/FATAL/PANIC` und **bricht vor dem Storage-
rsync ab**, wenn welche übrig bleiben (Override: `FORCE_RESTORE_OK=1`).
Danach macht es eine Zeilen-Paritätsprüfung (alt vs. neu) über die tragenden
Tabellen.
+51
View File
@@ -0,0 +1,51 @@
#!/usr/bin/env bash
# Shared config for the one-time netralax.cloud -> netralax.de server move.
#
# This is SEPARATE from scripts/prod/config.sh on purpose: the migration knows
# BOTH the old (.cloud) and the new (.de) host, whereas scripts/prod/config.sh
# only ever points at the live server. Source this in each migrate script:
# source "$(dirname "$0")/config.sh"
#
# Fill NEW_HOST once the netralax.de VPS exists. OLD_HOST is the .cloud VPS.
# Old, currently-live VPS (Supabase + LiveKit on *.netralax.cloud).
export OLD_HOST="46.225.156.249"
export OLD_USER="prox"
# New, empty VPS that will serve *.netralax.de (and keep serving *.netralax.cloud
# for already-installed clients). Fill in the IP before running 02-migrate-data.sh.
# NOTE: the login user on the new .de VPS is "debian" (the old .cloud VPS uses "prox").
export NEW_HOST="141.95.34.204"
export NEW_USER="debian"
# Paths on BOTH servers (same layout on old and new).
export SUPABASE_DIR="/opt/supabase"
export LIVEKIT_DIR="/opt/livekit"
# SSH helper opts: accept new host keys on first connect without prompting.
# Override SSH_OPTS from the environment if you need a jumphost etc.
export SSH_OPTS="${SSH_OPTS:--o StrictHostKeyChecking=accept-new}"
# Convenience SSH targets.
export OLD_SSH="${OLD_USER}@${OLD_HOST}"
export NEW_SSH="${NEW_USER}@${NEW_HOST}"
# Run a command on the OLD server.
old_remote() {
# shellcheck disable=SC2086
ssh ${SSH_OPTS} "${OLD_SSH}" "$@"
}
# Run a command on the NEW server.
new_remote() {
# shellcheck disable=SC2086
ssh ${SSH_OPTS} "${NEW_SSH}" "$@"
}
# Guard: refuse to run anything against the unfilled new-host placeholder.
require_new_host() {
if [[ "${NEW_HOST}" == "__NETRALAX_DE_SERVER_IP__" || -z "${NEW_HOST}" ]]; then
echo "NEW_HOST is still the placeholder — edit scripts/migrate/config.sh first." >&2
exit 1
fi
}
+3 -2
View File
@@ -8,8 +8,9 @@ All commands read shared config from `config.sh`.
1. Copy your SSH key to the server so scripts don't prompt for a password: 1. Copy your SSH key to the server so scripts don't prompt for a password:
``` ```
ssh-keygen -t ed25519 # only if you don't already have one ssh-keygen -t ed25519 # only if you don't already have one
ssh-copy-id prox@46.225.156.249 # PROD now points at the netralax.de VPS (user "debian"; see config.sh).
ssh prox@46.225.156.249 'echo ok' ssh-copy-id debian@141.95.34.204
ssh debian@141.95.34.204 'echo ok'
``` ```
2. Make the scripts executable: 2. Make the scripts executable:
``` ```
+13 -5
View File
@@ -5,16 +5,24 @@
# Customize here when the server IP / domains change — all other scripts pick # Customize here when the server IP / domains change — all other scripts pick
# the values up automatically. # the values up automatically.
export PROD_SERVER="46.225.156.249" # End state after the netralax.de migration. The old .cloud VPS was
export PROD_USER="prox" # 46.225.156.249 — for the one-time move (data dump/restore, storage rsync)
# use scripts/migrate/, which knows the old host explicitly. Fill in the new
# server IP once the netralax.de VPS exists, then this becomes the live config
# for all push-migrations / push-edge-function / create-invite / logs scripts.
export PROD_SERVER="141.95.34.204"
# Login user on the new .de VPS is "debian" (the old .cloud VPS used "prox").
export PROD_USER="debian"
# Paths on the remote server. # Paths on the remote server.
export PROD_SUPABASE_DIR="/opt/supabase" export PROD_SUPABASE_DIR="/opt/supabase"
export PROD_LIVEKIT_DIR="/opt/livekit" export PROD_LIVEKIT_DIR="/opt/livekit"
# Public domains (served via Caddy on the same VPS). # Public domains (served via Caddy on the new VPS). Caddy also keeps serving
export PROD_DOMAIN_SUPABASE="supabase.netralax.cloud" # the legacy supabase.netralax.cloud / livekit.netralax.cloud vhosts (same
export PROD_DOMAIN_LIVEKIT="livekit.netralax.cloud" # backends) so already-installed clients keep working until they auto-update.
export PROD_DOMAIN_SUPABASE="supabase.netralax.de"
export PROD_DOMAIN_LIVEKIT="livekit.netralax.de"
# SSH helper: forwards the standard `-o StrictHostKeyChecking=accept-new` so # SSH helper: forwards the standard `-o StrictHostKeyChecking=accept-new` so
# first connections don't prompt. Override SSH_OPTS from the environment if # first connections don't prompt. Override SSH_OPTS from the environment if