Prevent stale 404s for files written by another realm-server instance - #5923
Prevent stale 404s for files written by another realm-server instance#5923jurgenwerk wants to merge 2 commits into
Conversation
On a multi-instance realm-server sharing one EFS volume, a peer that had looked up a path before it existed keeps answering 404 for it until the kernel's directory attribute cache expires (~50s measured on staging). Re-listing the parent directory when the write notification arrives makes the kernel re-read the directory and drop the stale negative entry. CS-12032 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Host Test Results 1 files ±0 1 suites ±0 2h 13m 7s ⏱️ - 1m 21s Results for commit 65b7292. ± Comparison against earlier commit bdd13a8. Realm Server Test Results 1 files ±0 1 suites ±0 18m 43s ⏱️ + 3m 20s Results for commit 65b7292. ± Comparison against earlier commit bdd13a8. |
There was a problem hiding this comment.
Pull request overview
Adds cross-instance NFS directory-cache refreshes so peer realm servers can see newly written files promptly.
Changes:
- Adds
Realm.refreshDirectoryView(). - Refreshes directories after per-path notifications.
- Tests dispatch, wildcard exclusion, and failure handling.
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.
| File | Description |
|---|---|
packages/runtime-common/realm.ts |
Adds directory refresh logic. |
packages/realm-server/lib/realm-file-changes-listener.ts |
Triggers refreshes from notifications. |
packages/realm-server/tests/realm-file-changes-listener-test.ts |
Tests listener behavior. |
💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
| async refreshDirectoryView(path: LocalPath): Promise<void> { | ||
| let separator = path.lastIndexOf('/'); | ||
| let directory = separator === -1 ? '' : path.slice(0, separator); | ||
| for await (let _entry of this.#adapter.readdir(directory)) { |
There was a problem hiding this comment.
[Claude Code 🤖] (Written by Claude on Matic's behalf.) Fixed: the refresh now lists every ancestor from the realm root down to the immediate parent (ancestorDirectories), so a parent directory the peer had cached as absent is refreshed before its children are looked up. Covered by the new directory-view-refresher-test.ts.
| // Fire-and-forget: the NOTIFY handler stays synchronous, and a failed | ||
| // directory listing only means the kernel cache expires on its own. | ||
| realm.refreshDirectoryView(parsed.path).catch((err: unknown) => { |
There was a problem hiding this comment.
[Claude Code 🤖] (Written by Claude on Matic's behalf.) Fixed both parts: NodeAdapter.readdir now uses fs.promises-style readdir (the sync helper it used is removed), and the refresh logic moved into DirectoryViewRefresher, which coalesces per directory — while a listing is in flight, later requests share it and at most one follow-up is queued behind it. A batch write into one directory therefore costs a bounded number of listings instead of one per file; the test asserts listings never overlap and that five concurrent refreshes produce fewer listings than refreshes.
…ronously Address review: a peer that cached a parent directory as absent needs the ancestors re-listed too; batch writes emit one notification per file, so listings of one directory are coalesced (one in flight, one queued); NodeAdapter.readdir now uses the async fs API so the refresh and the request path do not block the event loop on a network filesystem. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
jurgenwerk
changed the title
Prevent a realm-server task from returning 404 for a file another task just wrote.
Staging runs two tasks on one shared EFS volume. A task that looked up a path before the file existed keeps trusting that "not there" answer for up to a minute (measured: 50 s), because the kernel caches it. The AI assistant then sees its own new files as missing and re-creates them.
Fix: every write already sends
NOTIFY realm_file_changesto the other task. The receiving task now also lists the directories leading to the path, which makes the kernel re-read them and drop the stale entry. On staging a directory listing made the 404s stop within 0.2 s. Listings are coalesced per directory, run asynchronously, and are covered by unit tests. Verify after deploy with the probe script on the ticket.🤖 Generated with Claude Code