2010/2/1 Andrew Dunstan <andrew(at)dunslane(dot)net>:
> Magnus Hagander wrote:
>>> Actually, on close inspection it looks to me like this commit: "Create
>>> typedef pgsocket for storing socket descriptors."
>>> could well be the culprit.
>> I'm not claiming it's not, but what exactly points to that? Does the
>> problem go away if you move to a version before that?
>> Because I'm 99% sure I saw it well before that commit, and we've had
>> reports on it from 8.4 as well, I'm not so sure... But it may be that
>> that commit made something more likely to happen...
> The buildfarm logs say otherwise. This was first seen in Jan 10, at least on my Windows animals:
> pgbfprod=# select sysname, min(snapshot) from build_status_log where
> sysname ~ 'bat' and log_stage ~ 'initdb' and log_text ~ 'pgstat'
> group by sysname;
> sysname | min ----------+---------------------
> dawn_bat | 2010-01-10 17:30:02
> red_bat | 2010-01-10 23:30:01
> (2 rows)
Well, that's clearly information I didn't have access to ;)
We *have* seen it before. But I guess that change either made it more
likely to happen, or it's a different issue with the same message.
But this one may well have been the -1 vs PGINVALID_SOCKET checks. It
didn't show up in my intial tests, but it also didn't show up entirely
consistently before :S The latest run of mastodon is also not showing
Did those two members produce consistent issues? Can you give them a
couple of kicks to get enough buidls out of them to figure out if this
solved the problem?
In response to
pgsql-hackers by date
|Next:||From: Magnus Hagander||Date: 2010-02-01 09:26:04|
|Subject: Re: mailing list archiver chewing patches|
|Previous:||From: Simon Riggs||Date: 2010-02-01 08:54:03|
|Subject: Re: Hot Standby and VACUUM FULL|