Optimize MySQL and PostgreSQL CI containers for test performance

[awln-temporal]

Summary

• Disable durability guarantees (fsync, doublewrite, binlog, full page writes) on MySQL and PostgreSQL CI containers
• Mount data directories on tmpfs to eliminate disk I/O entirely
• These settings only affect crash recovery (durability), not transaction isolation or consistency

MySQL flags added

• innodb-flush-log-at-trx-commit=0 - don't fsync redo log on every commit
• innodb-flush-method=nosync - skip fsync syscall
• innodb-doublewrite=0 - skip doublewrite buffer
• innodb-buffer-pool-size=512M - larger in-memory cache (up from 128M default)
• innodb-log-buffer-size=64M - larger redo log buffer
• sync-binlog=0 / skip-log-bin - disable binary logging
• tmpfs at /var/lib/mysql (2GB)

PostgreSQL flags added

• fsync=off - skip all fsync calls
• synchronous_commit=off - don't wait for WAL flush on commit
• full_page_writes=off - skip full page images in WAL
• wal_level=minimal / max_wal_senders=0 - minimal WAL overhead
• shared_buffers=256MB - larger shared memory cache
• work_mem=16MB - more memory for sorts/joins
• tmpfs at /var/lib/postgresql/data (2GB)

Test plan

• Local A/B test with smoke tests (MySQL: 75s -> 71s, PostgreSQL: 72s -> 71s — modest gains expected locally due to macOS Docker VM and fast NVMe)
• CI A/B comparison: compare this PR's job durations against recent main runs (baseline captured from runs 23058493365 and 23057581982)
• Verify no test failures introduced by the config changes

Made with Cursor