PGRAC v0.130.1 — MVP correctness maintenance
Maintenance update to v0.130.0, focused on consistent-read completion,
transaction/deadlock waits, update-receipt lifetime, visibility and shared-read
reservation cleanup. No experimental scheduling or worker-isolation performance
changes are included. Existing safety refusals remain enforced.
Qualification
- Four valid PRE samples, four Linux instances with 32 clients per instance.
- Zero unexpected client/server errors, no forced cancellation, full ordered
comparisons of 1,000,000 rows on every node after each sample, and all required
health, invariant and outstanding-work checks passed. - Retained clean data was reused. Final coordinated normal shutdown passed on
all four nodes, with all control files reporting shut down and data retained. - Median recorded TPS: 404.385; peak: 568.54. These are not a controlled
performance comparison or evidence of improvement. The four individual
results, including the lower later samples, are preserved in the release notes. - Full Fast and MVP release CI passed on the exact tagged commit
30a86aece4ada5f4ca60d3a8dc794da64b1d7a43. The
attached JSON records the accepted immutable commit and workflow/job identities.
Earlier failed attempts remain visible in GitHub Actions; passing an unchanged
rerun does not establish the root cause of an earlier intermittent failure.
Verified tag-triggered runs:
Fast CI,
MVP release CI.
All seven required jobs executed successfully. The
main integration CI
also passed all five jobs.
Scope
This is a source-only MVP maintenance release, not certification of crash
recovery, automatic failover, independent-host shared-LUN/GFS2 deployment,
production external fencing or mixed-version/rolling upgrades. Clean restart
evidence is not crash-recovery evidence. Existing v0.130.0 stays immutable.