On 2016-03-16 10:53:42 -0700, Joshua D. Drake wrote:
> Hello,
>
> fd.c[1] will remove files from pgsql_tmp on a restart but not a
> crash-restart per this comment:
>
> /*
> * NOTE: we could, but don't, call this during a post-backend-crash restart
> * cycle. The argument for not doing it is that someone might want to
> examine
> * the temp files for debugging purposes. This does however mean that
> * OpenTemporaryFile had better allow for collision with an existing temp
> * file name.
> */
>
> I understand that this is designed this way. I think it is a bad idea
> because:
>
> 1. The majority crash-restarts in the wild are going to be diagnosed rather
> easily within the OS itself. They fall into things like OOM killer and out
> of disk space.
I don't buy 1), like at all. I've seen numerous production instances
with crashes outside of os triggered things.
> 2. It can cause significant issues, we ran into this yesterday:
>
> -bash-4.1$ ls pgsql_tmp31227*|du -sh
> 250G
>
> There is no active process/backend with PID 31227. The database itself is
> only 55G, but we are taking up an 5x that with dead files.
>
> 3. The problem can get worse over time. If you have a very long running
> instance, any time the backend crash-restarts you have to potential to
> increase disk space used for no purpose.
But I think these outweigh the debugging benefit.
Andres
