[HN Gopher] The Secret Life of Fsync
___________________________________________________________________
The Secret Life of Fsync
Author : todsacerdoti
Score : 19 points
Date : 2023-04-01 12:40 UTC (1 days ago)
(HTM) web link (puzpuzpuz.dev)
(TXT) w3m dump (puzpuzpuz.dev)
| hardwaresofton wrote:
| Actually ran into this while doing some experiments with ZFS &
| Postgres, some non-enterprise drives will pretend they wrote data
| when they didn't and you can corrupt a even a rock-steady
| database like Postgres.
|
| If you're not afraid to lose some perf, you can disable volatile
| write cache on your non enterprise drive:
|
| https://vadosware.io/post/everything-ive-seen-on-optimizing-...
| pyrolistical wrote:
| Feels like this doesn't actually answer the question? The
| hardware can still lie and say it was flushed so it benchmarks
| better.
|
| Would need to do real world test and yank the power after fsync.
| This https://news.ycombinator.com/item?id=10552218 article did
| just that. But unfortunately we would need this tested against
| all drive models/manufacturers
| emmelaich wrote:
| That's all covered in the last sentence.
|
| It addresses the main question well; is there a way to instruct
| the disk to flush its cache and is it used by fsync? Yes and
| yes.
___________________________________________________________________
(page generated 2023-04-02 23:00 UTC)