[rbuergi]

The recurring CI flake, root-caused and killed deterministically

NodeHubContentCollectionTest.CollectionNamedArea_RendersBrowserForFolder_AndContentForFile fails intermittently on shard 2 with ObservableAssertionException : Expected the observable to emit a value matching the predicate within 30s, but it did not — it's been rerun-once past the merge gate repeatedly (including on #224 and #239 in this session).

Root cause (an "event miss")

A content collection is cached and its markdown files scanned once at init. ContentCollection.SaveFileAsync wrote the file to disk but did not update the collection's markdown stream — it relied entirely on the FileSystemWatcher to notice the new file and merge it (UpdateArticle). The file-content render (GetMarkdown → reduce that stream) therefore depends on a watcher event that is asynchronous and lossy: FileSystemWatcher coalesces events and overflows its OS buffer under load. When the event is dropped or delayed, the just-saved file never enters the stream → the render emits null → the 30s assertion times out.

This also bit production: an upload immediately followed by a view could show "not found" until (or unless) a watcher event arrived.

Fix

SaveFileAsync now merges the just-written markdown file into the stream itself (read-your-writes), through the same path the watcher uses. The watcher remains the mechanism for out-of-band (external) changes; a save the process performed is reflected deterministically.

Proven deterministically — both directions

The new SaveFileReadYourWritesTest runs the same save→render with the watcher turned off (via a new, genuinely-useful DisableFileWatcher config — the watcher is unreliable on network mounts / under load), so the file can reach the stream only through SaveFileAsync's merge:

• Without the fix → times out at 30s (the exact CI failure reproduced locally).
• With the fix → passes in ~370ms.

All 16 tests in the project pass; Release -warnaserror -p:CIRun=true clean.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR addresses a recurring CI flake (and a real production read-after-write issue) in ContentCollections by ensuring a markdown file saved via ContentCollection.SaveFileAsync is merged into the collection’s markdown stream immediately, rather than relying on potentially dropped/delayed FileSystemWatcher events.

Changes:

• Update ContentCollection.SaveFileAsync to proactively ingest just-written markdown files into markdownStream (read-your-writes).
• Add DisableFileWatcher to ContentCollectionConfig to support deterministic behavior where file watching is unreliable.
• Add a regression test that reproduces the prior failure mode by disabling the watcher and asserting save→render succeeds.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
test/MeshWeaver.ContentCollections.Test/SaveFileReadYourWritesTest.cs	Adds a regression test that disables file watching and verifies save→render works deterministically.
src/MeshWeaver.ContentCollections/ContentCollectionConfig.cs	Introduces DisableFileWatcher configuration with documentation explaining tradeoffs.
src/MeshWeaver.ContentCollections/ContentCollection.cs	Ensures SaveFileAsync merges markdown writes into the stream immediately; makes AttachMonitor a no-op when watcher is disabled.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

Test Results (shard 1)

865 tests  ±0   755 ✅ ±0   4m 13s ⏱️ -1s
 13 suites ±0   110 💤 ±0 
 13 files   ±0     0 ❌ ±0 

Results for commit a81b064. ± Comparison against base commit 4e1f3e1.

[rbuergi]

Superseded by #238 (merged, main green — "ingest a collection's own write immediately"), which lands the same read-your-writes fix (SaveFileAsync → proactive UpdateArticle) plus a watcher Created handler and path normalization, and its own deterministic regression test (ContentCollectionWriteIngestTest). This PR's core fix is now redundant and would conflict on merge.

Independent same-root-cause confirmation: I reproduced the failure deterministically by disabling the FileSystemWatcher (Linux inotify drops the create event for a file written into a just-created subdir; only the FILE-content render — which reduces the watcher-populated markdown stream — flaked; the folder listing reads the provider live, so it was never affected; CI-only because macOS FSEvents doesn't miss).

The only unique bit here is the optional ContentCollectionConfig.DisableFileWatcher knob (useful where the watcher is unreliable — network mounts / high load; the read-your-writes fix is what makes it viable). Happy to re-open a minimal PR with just that config if it's wanted; closing this to avoid redundant churn.

[rbuergi]

Superseded by #238, which merged first and is live on main (green): SaveFileAsync proactive UpdateArticle ingest so read-after-write no longer depends on the FileSystemWatcher, plus a Created handler for external files and a deterministic NoMonitorProvider regression test (ContentCollectionWriteIngestTest) that fails without the fix and passes with it. Same root cause and same core fix — closing as duplicate. (If the DisableFileWatcher config knob from #240 is wanted, it can be lifted onto main as a small follow-up.)