question on data loss on crash based on ai assessment

[sprappcom]

is this true as reviewed by claude:

CrossDB approach:

• Data is in OS page cache immediately (not application buffer)
• Another process can read it immediately via mmap
• Survives normal reboot (OS flushes on shutdown)
• Only loses data on power loss or kernel panic

This is exactly how CrossDB works - it's not "delayed persistence", it's "immediate to page cache" with catastrophic failure vulnerability.

[jcwangxp]

1-3 is right, 4 is not accurate as there's there's WAL which is similar with popular DB like MySQL/PGSQL. There's balance for persistence and performance. If no data loose for power cycle or kernel panic, then WAL must be flushed to disk after each commit which will be very slow. You can config the WAL sync mode

[sprappcom]

just curious, how bad is the performance with sync mode? i havent test it out. i didnt realise async is in 1s. do you roughly know what's the perf degradation with wal sync?