Repository navigation
fix(gateway): replay routed profiles' pending-message spools at boot - #123585
Closed
Totoro-qaq wants to merge 1 commit into
Closed
Totoro-qaq wants to merge 1 commit into
Totoro-qaq wants to merge 1 commit into
Conversation
A routed turn on a multiplexed gateway runs inside its profile's HERMES_HOME, and _get_flush_dir follows that home, so a transcript backlog spooled while the profile's store is unwritable lands in profiles/<name>/pending_messages/. Boot recovery only scanned the launch home, and the runtime drain keys on an in-memory set that a restart empties, so after a restart nothing read those files again: the messages never reached state.db and the files stayed on disk. Recover the launch home first as before, then each served profile that has a pending_messages directory, inside that profile's HERMES_HOME so the replay also lands in its own state.db. A failure in one profile is logged and does not stop the others.
|
🤖 Hermes Agent automated review: PR reviewed. Diff analyzed (128 lines changed). Check CI status and manual review recommended. |
alt-glitch
added
type/bug
|
Mirrored this exact tip on my fork and re-ran the boot-recovery witness. Base behavior: 1 failure with the routed profile spool left stranded; patched tip: 1 pass; restoring base reproduces the failure. The useful invariant for me is that recovery has to enter the routed profile's HERMES_HOME before calling recover_pending_to_db(), otherwise the replay can target the wrong state.db. No additional gap found. |
Author
|
Re the cluster note in #123658: this PR doesn't touch It composes with #117310: that PR decides which store a cap-dropped message replays into, and this one makes boot scan the routed profiles' spool dirs at all. Today |
teknium1
added a commit
that referenced
this pull request
Sep 28, 2026
…3584, salvage #123585) Tighten the boot replay helper and prove the class in one test: two homes under set_multiplex_active(True), a message spooled under each while its store was unwritable, both replayed into their own state.db, no spool file left, and the launch scope restored afterwards (A->B->A). Import gateway.run at module level so the suite's real-home I/O guard does not trip on the lazy bootstrap import.
teknium1
added a commit
that referenced
this pull request
Sep 28, 2026
…3584, salvage #123585) Tighten the boot replay helper and prove the class in one test: two homes under set_multiplex_active(True), a message spooled under each while its store was unwritable, both replayed into their own state.db, no spool file left, and the launch scope restored afterwards (A->B->A). Import gateway.run at module level so the suite's real-home I/O guard does not trip on the lazy bootstrap import.
teknium1
added a commit
that referenced
this pull request
Sep 28, 2026
…3584, salvage #123585) Tighten the boot replay helper and prove the class in one test: two homes under set_multiplex_active(True), a message spooled under each while its store was unwritable, both replayed into their own state.db, no spool file left, and the launch scope restored afterwards (A->B->A). Import gateway.run at module level so the suite's real-home I/O guard does not trip on the lazy bootstrap import.
teknium1
added a commit
that referenced
this pull request
Sep 28, 2026
…3584, salvage #123585) Tighten the boot replay helper and prove the class in one test: two homes under set_multiplex_active(True), a message spooled under each while its store was unwritable, both replayed into their own state.db, no spool file left, and the launch scope restored afterwards (A->B->A). Import gateway.run at module level so the suite's real-home I/O guard does not trip on the lazy bootstrap import.
teknium1
added a commit
that referenced
this pull request
Sep 28, 2026
…3584, salvage #123585) Tighten the boot replay helper and prove the class in one test: two homes under set_multiplex_active(True), a message spooled under each while its store was unwritable, both replayed into their own state.db, no spool file left, and the launch scope restored afterwards (A->B->A). Import gateway.run at module level so the suite's real-home I/O guard does not trip on the lazy bootstrap import.
teknium1
added a commit
that referenced
this pull request
Sep 28, 2026
…3584, salvage #123585) Tighten the boot replay helper and prove the class in one test: two homes under set_multiplex_active(True), a message spooled under each while its store was unwritable, both replayed into their own state.db, no spool file left, and the launch scope restored afterwards (A->B->A). Import gateway.run at module level so the suite's real-home I/O guard does not trip on the lazy bootstrap import.
Author
13 of 14 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What does this PR do?
On a multiplexed gateway, a routed turn runs inside its profile's HERMES_HOME, and
_get_flush_dir()follows that home, so a transcript backlog spooled while the profile's store is unwritable lands inprofiles/<name>/pending_messages/. Boot recovery only scanned the launch home, and the runtime drain keys on the in-memory_spooled_drop_sessionsset that a restart empties. After a restart nothing read those files: the messages never reachedstate.dband the files stayed on disk.Boot recovery now replays the launch home first, as before, then each served profile that has a
pending_messagesdirectory, inside that profile's HERMES_HOME sorecover_pending_to_db()opens (and appends to) that profile's store. A failure in one profile is logged and does not stop the others. Single-profile gateways are unchanged.Related Issue
Fixes #123584
Type of Change
Changes Made
gateway/run.py: the boot_recover_pendingclosure now calls a module-level_recover_pending_flushes(runner), which adds the per-profile pass.tests/gateway/test_profile_spool_recovery.py: spools a message under a routed profile's home, runs boot recovery, and checks the message is in the profile'sstate.dband the spool file is gone.How to Test
scripts/run_tests.sh tests/gateway/test_profile_spool_recovery.py: fails with the pre-fix behaviour (assert 0 == 1, the file stays inprofiles/work/pending_messages/) and passes with this change.scripts/run_tests.sh tests/gateway/test_profile_spool_recovery.py tests/gateway/test_shutdown_flush.py tests/gateway/test_pending_queue_spool.py: 18 passed.scripts/run_tests.sh tests/gateway/: 8848 passed, 77 skipped. The known failures intest_media_download_retry.py(Slack ×3) andtest_update_streaming.py::test_gateway_flag_enables_gateway_prompt_for_stashfail the same way onmain.test_failure_writer_ownership.pyfailed once under the parallel full run and passes 3/3 when run alone, with or without this change; it does not touch recovery.ruff check(0.15.10) on the changed files: clean.Checklist
Code
Documentation & Housekeeping
cli-config.yaml.example— N/ACONTRIBUTING.md/AGENTS.md— N/A