Re: Potential data loss of 2PC files

On Mon, Jan 30, 2017 at 10:52 PM, Heikki Linnakangas <hlinnaka(at)iki(dot)fi> wrote:
> So, if I understood correctly, the problem scenario is:
>
> 1. Create and write to a file.
> 2. fsync() the file.
> 3. Crash.
> 4. After restart, the file is gone.

Yes, that's a problem with fsync's durability, and we need to achieve
that at checkpoint. I find [1] a good read on the matter. That's
easier to decrypt than [2] or [3] in the POSIX spec..

[1]: http://blog.httrack.com/blog/2013/11/15/everything-you-always-wanted-to-know-about-fsync/
[2]: http://pubs.opengroup.org/onlinepubs/009695399/functions/fsync.html
[3]: http://pubs.opengroup.org/onlinepubs/009695399/functions/rename.html

> If that can happen, don't we have the same problem in many other places?
> Like, all the SLRUs? They don't fsync the directory either.

Right, pg_commit_ts and pg_clog enter in this category.

> Is unlink() guaranteed to be durable, without fsyncing the directory? If
> not, then we need to fsync() the directory even if there are no files in it
> at the moment, because some might've been removed earlier in the checkpoint
> cycle.

Hm... I am not an expert in file systems. At least on ext4 I can see
that unlink() is atomic, but not durable. So if an unlink() is
followed by a power failure, the previously unlinked file could be
here if the parent directory is not fsync'd.
--
Michael