Re: logical replication snapshots

On 2018-07-26 12:54:19 -0500, Dimitri Maziuk wrote:
> >> "LOG: logical decoding found consistent point at 19/E6942440"
> >> DETAIL: There are no running transactions."
> >>
> >> Others with more experience in this area would need to fill whether that
> >> might account for the 13 million files in the ~snapshot/ directory.
> >
> > That indicates there's some error handling issues to be resolved. Petr?
>
> It logs "COPY TO STDOUT", does that actually cache to disk (in ~snapshot/)?

No, it doesn't. The snapshots are internal to logical replication.

> Would it be safe to delete every file in ~snapshot/ older than some
> X-number of minutes? (And what a good X might be?)

You shouldn't manually delete them. But an explicit 'checkpoint;' (and
all automatically triggered ones), should do so for all that are
older. Is there a chance that you have some old leftover replication
slots around? Because that'd prevent cleanup?

What does SELECT * FROM pg_replication_slots ; return?

Greetings,

Andres Freund