Updater follow-ups from the reviews of #744 #754

Open
opened 2026-11-16 06:38:48 +00:00 by charles · 0 comments
Owner

Non-blocking findings from the three reviews of PR #744 (#740), to do after it has run on real machines:

  • crates/fluxer-platform/src/update/velopack.rs (Windows): the abandoned-download receiver put back on cancel can race a second waiter within the 100 ms poll; a new download then hits Velopack's lock and fails once (error if user-started, retry at the next check otherwise). Nothing wrong is installed.
  • Windows, documented in spec §5.3 and logged, untested: after a withdrawn release the first install can fail once; a newer download finishing during the shutdown after the quit deadline means nothing is installed that run.
  • apps/fluxer/src/main.rs: the process end now gives the storage worker 5 s + 3 s; a slower settings/cache write is lost (atomic, not corrupted). Say so in the docs; consider logging it.
  • Exit code 130 is used for SIGTERM too (143 is the convention).
  • After an AppImage relaunch the old process still holds the update lock (close-on-exec fd), so the new one skips its start-up sweep that run; release the lock before relaunching.
  • A second instance started during the quit wait gets Busy, later downloads the same release again (bandwidth only).
  • The runtime listens for SIGINT/SIGTERM only during the quit wait: a second Ctrl-C between main catching the first and that wait is lost.
  • None of the cfg(windows) updater code has been run on Windows: check a real Velopack update, a quit during a download, and a quit with an update staged.
  • Flaky under load: textchat_session_end_to_end ("per-channel notification sounds").
Non-blocking findings from the three reviews of PR #744 (#740), to do after it has run on real machines: - `crates/fluxer-platform/src/update/velopack.rs` (Windows): the abandoned-download receiver put back on cancel can race a second waiter within the 100 ms poll; a new download then hits Velopack's lock and fails once (error if user-started, retry at the next check otherwise). Nothing wrong is installed. - Windows, documented in spec §5.3 and logged, untested: after a withdrawn release the first install can fail once; a newer download finishing during the shutdown after the quit deadline means nothing is installed that run. - `apps/fluxer/src/main.rs`: the process end now gives the storage worker 5 s + 3 s; a slower settings/cache write is lost (atomic, not corrupted). Say so in the docs; consider logging it. - Exit code 130 is used for SIGTERM too (143 is the convention). - After an AppImage relaunch the old process still holds the update lock (close-on-exec fd), so the new one skips its start-up sweep that run; release the lock before relaunching. - A second instance started during the quit wait gets `Busy`, later downloads the same release again (bandwidth only). - The runtime listens for SIGINT/SIGTERM only during the quit wait: a second Ctrl-C between main catching the first and that wait is lost. - None of the `cfg(windows)` updater code has been run on Windows: check a real Velopack update, a quit during a download, and a quit with an update staged. - Flaky under load: `textchat_session_end_to_end` ("per-channel notification sounds").
charles referenced this issue from a commit 2026-11-18 19:47:55 +00:00
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
charles/fluxer-native#754
No description provided.