On Mon, Aug 15, 2011 at 9:34 PM, Fujii Masao <masao.fujii@gmail.com> wrote:
> On Thu, Aug 11, 2011 at 7:19 AM, Lonni J Friedman <netllama@gmail.com> wrote:
>> First the problem. On *only* one of the two standby servers, I'm
>> seeing errors like the following whenever I issue any SQL commands on
>> the master which write (insert, update, etc) to the database:
>> LOG: invalid record length at 8/7A000020
>> FATAL: terminating walreceiver process due to administrator command
>> LOG: invalid record length at 8/7A0000B0
>> LOG: streaming replication successfully connected to primary
>> LOG: invalid record length at 8/7B000020
>> FATAL: terminating walreceiver process due to administrator command
>> LOG: record with zero length at 8/7B0000B0
>> LOG: streaming replication successfully connected to primary
>> LOG: record with incorrect prev-link 8/79000058 at 8/7D0000B0
>> LOG: streaming replication successfully connected to primary
>
> Did you use gcc4.6 or later to build PostgreSQL9.0? If yes, you would
> face the same problem reported before;
> http://archives.postgresql.org/pgsql-hackers/2011-06/msg00661.php
>
> This problem was fixed, and the fix will be included in next minor update
> (i.e., 9.0.5).
> http://archives.postgresql.org/pgsql-committers/2011-06/msg00101.php
Yes, that was the issue. I thought that I had replied earlier to
someone else speculating that this was the issue, but perhaps I had
not.
>> 1) Both of the wiki links above comment that the restore_command may
>> not be necessary if wal_keep_segments is large enough (mine is set to
>> 128). I was going to setup the restore_command anyway, as I'm not yet
>> confident enough about streaming replication and failover with
>> postgresql to take chances, although the fact that i have two standby
>> servers makes this setup a bit more complex. However, can anyone
>> comment about whether its ever truly safe 100% of the time to run
>> without a restore_command ?
>
> Specifically, what problem are you concerned about?
I wish I knew. All the documentation out there always focuses on
setting up a restore command, as if there would be a huge disaster if
it wasn't done. Is it safe to simply make wal_keep_segments really
large, and skip the restore_command altogether?