Hotfix 0.34.1: upgrade no longer runs a whole-index influence pass; a stop during startup is not fatal (#334) #342
No reviewers
Labels
No labels
code-review
correctness
dos
performance
security
severity/high
severity/low
severity/medium
tech-debt
Kind/Breaking
Kind/Bug
Kind/Documentation
Kind/Enhancement
Kind/Feature
Kind/Security
Kind/Testing
Priority
Critical
Priority
High
Priority
Low
Priority
Medium
Reviewed
Confirmed
Reviewed
Duplicate
Reviewed
Invalid
Reviewed
Won't Fix
Status
Abandoned
Status
Blocked
Status
Need More Info
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
h-dv/code-index!342
Loading…
Reference in a new issue
No description provided.
Delete branch "hotfix-334"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Fixes #334, a field report from zammad-dev. On 0.34.0 with de.h-dv.ruby owning Ruby, the first daemon after the upgrade sat in one resolve step for 45 minutes, then exited with code 1 on
sqlite error: interrupted.Root cause
An upgrade changes the host identity, so the daemon's startup activation rebuilds the tree as a new generation and promotes it.
build::requestwrote the new generation throughactivation::write_generation, which leavesbuiltin_equivalent = 0. Onlyactivation::record, on the ordinary pass, ever set it to 1.influence::apply_generalran over every ref, about 646k on Zammad.resolution_influencebefore the fix, 1,070 ms after.Commits
68bca7b5:build::requestrecords builtin-equivalence from the livePackageHost, as the ordinary pass does.first_party_influence::a_pipeline_built_generation_keeps_the_first_party_exemption.…must be recorded builtin-equivalent: [0].7a5f4d5f: a requested stop (SIGTERM/SIGINT) during the startup activation goes through the clean-shutdown path instead ofFatalexit 1. The owed work was already durable in the m0007 pending tables.a_requested_stop_never_makes_a_startup_failure_fatal.514e37e0:version_bump.sh 0.34.1.Verification on this tree
-D warnings --document-private-items,guest_gates.sh: clean.Follow-ups, not in this PR
🤖 Generated with Claude Code
https://claude.ai/code/session_0126PDDLB4wNHxKXvWM1VNmu
SIGINT/SIGTERM during the startup activation's promotion interrupts the resolve that follows the committed pointer switch. `startup_activation` classified that SQLITE_INTERRUPT as StartupFailure::Fatal, so the daemon exited 1 with "startup generation N committed; resolution remains pending: sqlite error: interrupted" although the switch had already made the owed work durable and the next start repairs it (measured: the restarted daemon ran the owed Full pass and indexed the edits made during the dead one). A Fatal failure now ends the process only when no stop was requested; otherwise it reaches the existing clean-shutdown path. Measured on the Zammad copy, SIGTERM mid-promotion: before, exit 1 + ERROR; after, exit 0 + "shutting down during initial reconcile; pending resolve work is durable". Test: tests::a_requested_stop_never_makes_a_startup_failure_fatal. Mutations run: drop the `!stopping` guard -> RED ("a requested stop must reach the clean-shutdown path"); retry every Fatal -> RED ("a real refusal of this activation must still exit non-zero"). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0126PDDLB4wNHxKXvWM1VNmuThe local full gate run on
514e37e0passed with no failures:-D warnings --document-private-items) andguest_gates.sh: clean;test result: oklines;corpus_ratchet: 9/9;precision_gate: 7/7 with 0 phantoms; the POPULATION lines match those from the #275 run earlier today.Fast-forwarded into master at
514e37e0and released as v0.34.1 (release run 1035, 56 assets): https://git.h-dv.de/h-dv/code-index/releases/tag/v0.34.1CI passed on both the PR (runs 1029–1031) and the master push (runs 1032–1034). Forgejo does not mark a fast-forward as merged, so I'm closing this by hand.
Pull request closed