Re: streaming replication does not work across datacenter with 20ms latency?

Yan Chunlu <springrider(at)gmail(dot)com> writes:
> seems the Master server is compiled using 4.6.0:

> PostgreSQL 9.0.4 on x86_64-pc-linux-gnu, compiled by GCC gcc-4.6.real
> (Debian 4.6.0-6) 4.6.1 20110428 (prerelease), 64-bit

Hmm. Given the datestamp, that version of gcc almost certainly does
have the bug. I wonder whether Martin Pitt knows about this issue
and the workaround we put in --- I'd have thought he'd push updated
.debs with a workaround, as I did for Fedora ...

Martin: see
http://archives.postgresql.org/pgsql-hackers/2011-06/msg00890.php
http://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=c2ba0121c73b7461331104a46d140156e847572a

> and there is no way to know what slave is using since I have remove it.

If it was installed from the same .deb then it'd be the same build.

> I think the problem maybe is like Fujii said, does that bug only
> effect hot-stanby server? seems master is okay.

Well, actually, the bug affects WAL replay of any sort, which means
if your master were to crash and restart you'd be at risk of data
corruption on the master. I'd replace the master build too, ASAP.

regards, tom lane