Re: Invalid redo in checkpoint record

On 27/11/2009 12:50 AM, Scott Marlowe wrote:
> On Thu, Nov 26, 2009 at 9:06 AM, Shak <sshaikh(at)hotmail(dot)com> wrote:
>> Hi,
>>
>> After a recent server crash after which the whole system hung and power had
>> to be pulled, I'm having the following error pop up when trying to start the
>> DB:
>>
>> * Starting PostgreSQL 8.4 database server
>> * The PostgreSQL server failed to start. Please check the log output:
>> 2009-11-26 16:00:23 GMT LOG: database system was interrupted; last known up
>> at 2009-11-25 17:37:07 GMT
>> 2009-11-26 16:00:23 GMT PANIC: invalid redo in checkpoint record
>> 2009-11-26 16:00:23 GMT LOG: startup process (PID 3126) was terminated by
>> signal 6: Aborted
>> 2009-11-26 16:00:23 GMT LOG: aborting startup due to startup process
>> failure
>>
>> [fail]
>>
>> This is on Ubuntu 9.10. I've since tried to reinstall Postgres via Synaptic
>> (using complete removal), but I'm getting the same error.
>>
>> The data is unimportant, so is there any way of getting the server itself to
>> run again? I've been advised to create a new cluster, but if I can somehow
>> reset or recreate the default one that'd be the best option. Otherwise I'm
>> considering reinstalling the whole shebang which seems a bit drastic.

Before you do re-create the cluster, if the data is unimportant is there
any chance you could take a copy of it so it can be examined to see what
happened? PostgreSQL should recover cleanly after a hard crash, and
unless there's a storage subsystem issue or fsync was off this sort of
thing might indicate an issue with Pg's crash recovery. Having a copy of
the database would be really handy.

The whole data directory would need to be tar'ed up and gzip'd.

--
Craig Ringer