Re: Invalid magic number in log file?

Gregory Stark wrote:
> 
> "Alvaro Herrera" <alvherre@commandprompt.com> writes:
> 
> > Gregory Stark wrote:
> >> 
> >> When starting up a postmaster from an older build on a database initialized
> >> with a newer build I'm getting the following errors. I think I saw the same
> >> thing earlier going in the opposite direction too. Shouldn't there be some
> >> earlier errors firing from the control file version number?
> >> 
> >> stark@oxford:~$ /usr/local/pgsql/bin/postgres -D /var/tmp/db
> >> LOG:  database system was shut down at 2007-05-14 19:47:51 BST
> >> LOG:  invalid magic number D061 in log file 0, segment 0, offset 0
> >> LOG:  invalid primary checkpoint record
> >> LOG:  invalid magic number D061 in log file 0, segment 0, offset 0
> >> LOG:  invalid secondary checkpoint record
> >> PANIC:  could not locate a valid checkpoint record
> >> LOG:  startup process (PID 9590) was terminated by signal 6: Aborted
> >> LOG:  aborting startup due to startup process failure
> >
> > Huh, works for me, what versions were those?
> 
> Well it's cvs HEAD and a checkout from May 1st.

It was changed in 1.20 of xlog_internal.h.  That particular changeset
didn't include a catversion bump nor pg_control magic version change.
I think the rationale here is that it was only a change in WAL format,
so it shouldn't affect either -- it did change the WAL magic.

Maybe the thing to do here is to disallow running a postmaster when the
data dir is using a different WAL magic (forcing you to pg_resetxlog or
initdb).  Does it work if you do a pg_resetxlog and restart?


2007-04-30 17:01  tgl
   * src/: backend/access/transam/twophase.c (1.30),     backend/access/transam/xact.c (1.242),
backend/access/transam/xlog.c(1.268),     backend/utils/adt/timestamp.c (1.176), include/access/xact.h     (1.87),
include/access/xlog_internal.h(1.20),     include/utils/timestamp.h (1.69):
 

Change the timestamps recorded in transaction commit/abort xlog records
from time_t to TimestampTz representation.  This provides full gettimeofday()
resolution of the timestamps, which might be useful when attempting to
do point-in-time recovery --- previously it was not possible to specify
the stop point with sub-second resolution.  But mostly this is to get
rid of TimestampTz-to-time_t conversion overhead during commit.  Per my
proposal of a day or two back.



-- 
Alvaro Herrera                                http://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support