Untitled attachment

Notices where this attachment appears

1. Embed this notice
   Chris Siebenmann (cks@mastodon.social)'s status on Saturday, 01-Mar-2025 12:11:33 JST Chris Siebenmann

   I will admit that until I saw the CouchDB post, the scenario of 'program writes but crashes before fsync, restarted program re-reads the just written data and assumes it's safe on disk instead of just in the kernel buffer cache' had not occurred to me. I guess everything that reads a WAL or other recovery file should immediately fsync() it on startup.

   In re: https://narrativ.es/@janl/114076211520071258