Re: streaming replication master can fail to shut down

Hi,

On 2016-03-11 14:12:37 +0000, Nick Cleaton wrote:
> This patch is enough to eliminate the problem on my hardware

> diff -Nurd postgresql-9.5.1.orig/src/backend/replication/walsender.c postgresql-9.5.1/src/backend/replication/walsender.c
> --- postgresql-9.5.1.orig/src/backend/replication/walsender.c 2016-02-08 21:12:28.000000000 +0000
> +++ postgresql-9.5.1/src/backend/replication/walsender.c 2016-03-11 11:56:41.121361222 +0000
> @@ -2502,8 +2502,10 @@
>
> proc_exit(0);
> }
> - if (!waiting_for_ping_response)
> + if (!waiting_for_ping_response) {
> WalSndKeepalive(true);
> + waiting_for_ping_response = true;
> + }
> }

That looks (besides non-postges paren placement), reasonable. Will
commit & backpatch (to 9.4, where it looks like the bug has been
introduced).

> in this test the server sent only 29 keepalives during the shutdown:
> http://nick.cleaton.net/protodump-100k-nossl-patched.xz (47k)

I guess you have a fair amount of WAL traffic, and the receiver was
behind a good bit?

Greetings,

Andres Freund