Re: pg_rewind, a tool for resynchronizing an old master after failover

On 2013-05-24 10:57:22 -0400, Thom Brown wrote:
> By the way, without any data inserted I get:
> 
> thom@swift /tmp $ pg_rewind --target-pgdata=/tmp/primary
> --source-server='host=localhost port=5531 dbname=postgres' -v
> connected to remote server
> fetched file "global/pg_control", length 8192
> fetched file "pg_xlog/00000002.history", length 41
> Last common WAL position: 0/3000000 on timeline 1
> pg_rewind: xlogreader.c:214: XLogReadRecord: Assertion `((RecPtr) %
> 8192 >= (((intptr_t) ((sizeof(XLogPageHeaderData))) + ((8) - 1)) &
> ~((intptr_t) ((8) - 1))))' failed.
> Aborted (core dumped)
> 
> And this magically goes away with -k on initdb.

That looks like this is a bug independent of -k. pg_rewind tries to read
the beginning of a page but xlogreader doesn't allow that atm. The
likely reason this doesn't happen with -k is autovacuum probably hints
some pages which are logged when checksums are enabled.

Should we make that assert
Assert((RecPtr % XLOG_BLCKSZ == 0) ||XRecOffIsValid(RecPtr));
instead?

Greetings,

Andres Freund

-- Andres Freund                       http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services