v0.3.3

replited v0.3.3

Release date: 2026-02-06

Highlights

• Fixed multiple stream/restore edge cases that could cause replay gaps or repeated restore loops.
• Added monitor_interval_ms to tune primary-side WAL shadow sync polling interval.
• Reduced E2E replication latency by removing fixed 1-second idle delay in replica streaming.
• Improved log quality by reducing expected startup warning noise.

What's Changed

Stability

• Fixed restore --follow early return path in WAL apply loop.
• Fixed follow-mode generation transition to use correct start index.
• Fixed blocker/unblock lifecycle in sidecar managed-process WAL reset path.
• Improved rewind-floor handling to accept valid next-index boundary advance.
• Improved latest snapshot selection to compare (index, offset).

Configuration

• max_checkpoint_page_number = 0 is now accepted to disable forced checkpointing.
• New monitor_interval_ms (default: 1000) in [[database]]:
  • controls primary-side WAL shadow sync poll interval
  • lower values reduce replication lag with higher polling overhead

Logging

• Startup WAL-mode busy/locked condition on pinned primary connection is now debug-level.
• Cold-start WAL header missing ... rewinding is now info-level.
• Removed release-build warnings from debug-only function arguments.

Benchmark Snapshot (E2E)

Using tests/benchmarks/run_all_benchmarks.sh with benchmark config tuning:

• Latency: mean 79.30 ms, p90 104.45 ms, p99 123.29 ms
• Warning summary during benchmark run:
  • Primary WARN/ERROR: 0/0
  • Replica WARN/ERROR: 0/0
  • SnapshotBoundaryMismatch: 0

Verification

• cargo test -q
• cargo clippy --all-targets --all-features -- -D warnings
• E2E benchmark script:
  • tests/benchmarks/run_all_benchmarks.sh

Full Changelog: v0.3.2...v0.3.3