Issues

--max-wal-size never reaches the P6 overflow ceiling: WalWriterV3::set_wal_bounds has no production caller, so the ceiling is always the 256 MiB default

#916

· TinDang97 opened

on Sep 10, 2026

Stale comment in promote_cold_outcome claims the cold tier never sweeps TTL — ColdIndex::sweep_expired does exactly that

#915

· TinDang97 opened

on Sep 10, 2026

Cold-plane cut is not seeded on tokio --shards 1 fresh boot, on legacy AOF generations, or on the WAL-authority path

#914

· TinDang97 opened

on Sep 10, 2026

CI stages are defined in three places that have silently diverged (ci-local.sh, ci.yml, the Linux bar)

#913

· TinDang97 opened

on Sep 10, 2026

tokio leg: --client-query-buffer-limit lets 5.29 MB through a 1 MiB cap

#906

· TinDang97 opened

on Sep 10, 2026

Seven of eight secondary write commands permanently flatten a small container — one HINCRBY/SREM/ZINCRBY/LPOP loses the compact encoding forever

#897

· TinDang97 opened

on Sep 10, 2026

The hosted per-PR memory gate runs --disk-offload disable, so it never measures the shipped default — which idles ~7x higher

#895

· TinDang97 opened

on Sep 9, 2026

EVAL/EVALSHA/FCALL/EVAL_RO inside MULTI are queued, then dropped at EXEC with 'unknown command' — the script's writes never apply

#894

· TinDang97 opened

on Sep 9, 2026

Warm vector segment file_id reuse makes boot recovery delete the segment directory it just attached

#893

· TinDang97 opened

on Sep 9, 2026

maxmemory footprint correction: two uncorrected consumers left over from #475 (Linux swap; db-maxmemory)

#892

· TinDang97 opened

on Sep 9, 2026

The production host costs 5.80x on reconcile — measured, same commit and same data, macOS vs Linux

#891

· TinDang97 opened

on Sep 9, 2026

Warm vector segments orphaned from their keymaps are re-encoded on every boot — 21% of segments, ~46% of the reconcile's re-encode

#888

· TinDang97 opened

on Sep 9, 2026

Navigation Menu

Uh oh!

All issues

--max-wal-size never reaches the P6 overflow ceiling: WalWriterV3::set_wal_bounds has no production caller, so the ceiling is always the 256 MiB default

Stale comment in promote_cold_outcome claims the cold tier never sweeps TTL — ColdIndex::sweep_expired does exactly that

Cold-plane cut is not seeded on tokio --shards 1 fresh boot, on legacy AOF generations, or on the WAL-authority path

CI stages are defined in three places that have silently diverged (ci-local.sh, ci.yml, the Linux bar)

tokio leg: --client-query-buffer-limit lets 5.29 MB through a 1 MiB cap

Seven of eight secondary write commands permanently flatten a small container — one HINCRBY/SREM/ZINCRBY/LPOP loses the compact encoding forever

The hosted per-PR memory gate runs --disk-offload disable, so it never measures the shipped default — which idles ~7x higher

EVAL/EVALSHA/FCALL/EVAL_RO inside MULTI are queued, then dropped at EXEC with 'unknown command' — the script's writes never apply

Warm vector segment file_id reuse makes boot recovery delete the segment directory it just attached

maxmemory footprint correction: two uncorrected consumers left over from #475 (Linux swap; db-maxmemory)

The production host costs 5.80x on reconcile — measured, same commit and same data, macOS vs Linux

Warm vector segments orphaned from their keymaps are re-encoded on every boot — 21% of segments, ~46% of the reconcile's re-encode

Uh oh!

Search results