Skip to content

fix: Poll the override files when they cannot be watched - #495

Merged
kinyoklion merged 1 commit into
v7from
rlamb/overrides-go-watch-fallback
Nov 3, 2026
Merged

kinyoklion merged 1 commit into
v7from
rlamb/overrides-go-watch-fallback

Conversation

@kinyoklion

@kinyoklion kinyoklion commented Oct 28, 2026 •

Copy link
Copy Markdown
Member

Summary

With ChangeDetection(Watching), a file override source whose file watcher cannot be created, for example because the limit on inotify instances is reached, logged "Unable to watch override files" and ran with no change detection at all. A later edit, creation, or deletion of an override file never took effect, which the OVERRIDE specification's Requirement 1.3.7 does not allow ("detecting changes by watching the files or by polling them, according to platform support").

The source now falls back to polling in that case. It logs at Error level "Unable to watch override files; polling them instead every {interval}: {reason}" and polls at the configured interval raised to MinimumPollInterval, because watching mode otherwise ignores the minimum. Polling mode and a watcher that starts normally are unchanged; the watcher's own retry for a directory it cannot watch yet is unchanged.

The same gap was found by Bugbot on the Java implementation (launchdarkly/java-core#224) and is fixed there the same way.

Testing

TestFileSourceWatchingModeFallsBackToPollingWhenTheWatcherCannotBeCreated replaces the watch setup with one that fails, then checks that an edit is applied by polling and that the error is logged. With the poller creation removed, the test fails. go test -race ./... passes (31 packages) and make lint reports no issues.

Opened from termag

SDK-3419


Note

Overview
Watching mode for file override sources no longer leaves change detection disabled when the OS cannot create a file watcher (for example inotify limits). The source falls back to polling at the configured PollInterval, clamped to MinimumPollInterval, and logs an error that includes the poll interval and failure reason.

Production wiring is unchanged via an injectable watchFiles hook (default ldfilewatch.WatchOptionalFiles) so tests can simulate watcher creation failure. Polling mode and successful watchers are unaffected.

A new test verifies that edits are picked up via polling and that the fallback error message is logged.

Reviewed by Cursor Bugbot for commit ea03508. Bugbot is set up for automated code reviews on this repo. Configure here.

When watching was selected and the file watcher could not be created,
for example because the limit on inotify instances was reached, the
source logged the error and ran with no change detection, so a later
edit never took effect. It now polls the files instead, at no less than
the minimum poll interval, and the log says so.
@kinyoklion

Copy link
Copy Markdown
Member Author

bugbot review

@cursor cursor Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

Reviewed by Cursor Bugbot for commit ea03508. Configure here.

@kinyoklion
kinyoklion marked this pull request as ready for review October 30, 2026 23:54
@kinyoklion
kinyoklion requested a review from a team as a code owner October 30, 2026 23:54
@kinyoklion
kinyoklion merged commit 170ad3b into v7 Nov 3, 2026
34 checks passed
@kinyoklion
kinyoklion deleted the rlamb/overrides-go-watch-fallback branch November 3, 2026 22:04
beekld pushed a commit that referenced this pull request Nov 6, 2026
🤖 I have created a release *beep* *boop*
---


##
[7.22.0](v7.21.2...v7.22.0)
(2026-11-06)


### Features

* Broadcast Off to FDv2 status listeners when the data system stops
([#505](#505))
([5c5b8ce](5c5b8ce))
* End an FDv1 fallback after a default TTL of one hour
([#499](#499))
([1f03a5f](1f03a5f))
* Park FDv2 synchronizers instead of retiring them on unexpected
failures
([#489](#489))
([644253a](644253a))


### Bug Fixes

* Keep data from a store that is not initialized out of override
evaluations and AllFlagsState, and report a store failure
([#491](#491))
([1005a1a](1005a1a))
* Keep the override marking when a migration flag serves an unrecognized
stage ([#492](#492))
([9f9911f](9f9911f))
* Leave StateSince alone when a synchronizer starts
([#502](#502))
([823b9aa](823b9aa))
* Make TestData ValueForAll return its value when targeting is off
([#507](#507))
([7300bc1](7300bc1))
* Park a synchronizer that reports Off or closes its results
([#500](#500))
([f6e9910](f6e9910))
* Poll the override files when they cannot be watched
([#495](#495))
([170ad3b](170ad3b))
* Prevent a full status listener from hanging Close
([#498](#498))
([9efd570](9efd570))
* Say that the data system retries after an unrecoverable HTTP error
([#501](#501))
([d4066cc](d4066cc))
* Treat a repeated override file path in a different case as a repeat on
Windows
([#494](#494))
([eeb68d3](eeb68d3))
* Warn when offline mode skips the override source
([#493](#493))
([69bbe61](69bbe61))

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).

<!-- CURSOR_SUMMARY -->
---

> [!NOTE]
> **Overview**
> **Cuts the Go server SDK release `7.22.0` and records its notes.**
> 
> The root **`.release-please-manifest.json`** and
**`internal/version.go`** now declare **`7.22.0`** (replacing `7.21.2`).
**`CHANGELOG.md`** gains a new **`[7.22.0]`** section dated
**2026-11-06** documenting shipped work: FDv2 data-system behavior
(status **Off** broadcasts, synchronizer **parking** instead of retire,
**FDv1 fallback TTL**, **StateSince** handling, parking on
**Off**/closed results), plus multiple **bug fixes** around
uninitialized store data in overrides/**AllFlagsState**, migration
override marking, **TestData** when targeting is off, **override file**
polling/deduping on Windows, offline-mode warnings, status listener
**Close** hangs, and messaging for unrecoverable HTTP retries.
> 
> No application code changes appear in this PR; it’s version
bookkeeping and release notes only.
> 
> <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit
c763cc0. Bugbot is set up for automated
code reviews on this repo. Configure
[here](https://www.cursor.com/dashboard/bugbot).</sup>
<!-- /CURSOR_SUMMARY -->

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants