Sami LehtinenBtrfs - Most curious case, is this flaw or not. Can't really tell. - I've got lots of scattered data in dirty write buffers to disk, due to scattered writes and high system memory. I've also got long commit interval to allow efficient flushing (as contiguous as possible) when that scattered data is finally written to disk. It all works fine, if I sync before shutdown. But if I just shutdown the system, the flush process takes longer than the maximum allowed file system unmount time is. Even worse, some of the data is written to SSD and some to HDD. The state data is on SSD and actual data is on HDD. When the flush timeout is encountered, the system shuts down in state where state file says that this data is written to disk, and the data actually isn't on disk yet. - Flaw or not? - Of course this requires this curious combination of things to trigger, but the problem is very real. - Who's fault it is? <a class="hashtag" href="https://pleroma.envs.net/tag/btrfs" rel="nofollow noopener noreferrer" target="_blank">#btrfs</a> <a class="hashtag" href="https://pleroma.envs.net/tag/data" rel="nofollow noopener noreferrer" target="_blank">#data</a> <a class="hashtag" href="https://pleroma.envs.net/tag/corruption" rel="nofollow noopener noreferrer" target="_blank">#corruption</a> <a class="hashtag" href="https://pleroma.envs.net/tag/loss" rel="nofollow noopener noreferrer" target="_blank">#loss</a> <a class="hashtag" href="https://pleroma.envs.net/tag/cow" rel="nofollow noopener noreferrer" target="_blank">#cow</a> <a class="hashtag" href="https://pleroma.envs.net/tag/linux" rel="nofollow noopener noreferrer" target="_blank">#linux</a> <a class="hashtag" href="https://pleroma.envs.net/tag/filesystem" rel="nofollow noopener noreferrer" target="_blank">#filesystem</a> <a class="hashtag" href="https://pleroma.envs.net/tag/shutdown" rel="nofollow noopener noreferrer" target="_blank">#shutdown</a>