Conversation

Notices

1. thurloat 📌 (thurloat@mastodon.club)'s status on Wednesday, 17-Apr-2019 11:45:15 EDT thurloat 📌

   #zfs users, what literally happens when a dedicated log device fills up before the rolling flush?

   1. thurloat 📌 (thurloat@mastodon.club)'s status on Wednesday, 17-Apr-2019 12:55:52 EDT thurloat 📌

      in reply to

      TIL: zfs doesn't like devices smaller than 64MB, and will silently shit the bed on it and refuse to add one as a log device.

      RIP my 10kb SLOG device for testing.

      1. thurloat 📌 (thurloat@mastodon.club)'s status on Wednesday, 17-Apr-2019 13:42:02 EDT thurloat 📌

         in reply to

         #zfs experiment update with 64MB log device.

         I think it may only effect throughput, the behaviour seems to be fill -> empty -> fill -> empty with `txg_sync` doing a lot of heavy lifting more often then every 5 secs.

         It seems like the zfs_txg_timeout default of 5 seconds if more of a "goal" for zfs to hit, and if there are extenuating circumstances it will trigger a txg sync