What happens after a write? Reworking Squirix's WAL in preview.6

Short summary

Squirix preview.6 reworks its write-ahead log with a pipelined binary WAL backend, a single journal thread, and bounded queue to handle storage backpressure. Journal-first ordering ensures no acknowledged-but-unrecoverable writes, while group commit lets multiple appends share one fsync for better throughput. The article details crash recovery semantics, commit-unknown states, and the tradeoffs between durability guarantees and response latency.

• •Pipelined binary WAL replaces JSON-based journal with CRC32C checksums and predictable record sizes
• •Single journal thread with bounded ring buffer prevents unbounded queue growth under storage pressure
• •Group commit batches multiple appends into one fsync, trading controlled latency for higher throughput

Generated with AI, which can make mistakes.