Re: Recovery will take 10 hours - Mailing list pgsql-performance

Hi Tom,

I found it... it's called ktrace on OS X Server.

However, as I just finished posting to the list, the process died
with a PANIC error:

[2006-04-20 16:41:28 MDT] LOG: restored log file
"000000010000018F00000034" from archive
[2006-04-20 16:41:35 MDT] LOG: restored log file
"000000010000018F00000035" from archive
[2006-04-20 16:41:38 MDT] LOG: restored log file
"000000010000018F00000036" from archive
sh: line 1: /wal_archive/000000010000018F00000037.gz: No such file or
directory
[2006-04-20 16:41:46 MDT] LOG: could not open file "pg_xlog/
000000010000018F00000037" (log file 399, segment 55): No such file or
directory
[2006-04-20 16:41:46 MDT] LOG: redo done at 18F/36FFF254
sh: line 1: /wal_archive/000000010000018F00000036.gz: No such file or
directory
[2006-04-20 16:41:46 MDT] PANIC: could not open file "pg_xlog/
000000010000018F00000036" (log file 399, segment 54): No such file or
directory
[2006-04-20 16:41:46 MDT] LOG: startup process (PID 9190) was
terminated by signal 6
[2006-04-20 16:41:46 MDT] LOG: aborting startup due to startup
process failure
[2006-04-20 16:41:46 MDT] LOG: logger shutting down


Would turning off fsync make it go faster? Maybe it won't take 10
hours again if we start from scratch.

Also, what if we did just start it up again? Will postgres realize
that the existing wal_archive files have already been processed and
just skip along until it finds one it hasn't processed yet?

Thanks,

____________________________________________________________________
Brendan Duddridge | CTO | 403-277-5591 x24 |  brendan@clickspace.com

ClickSpace Interactive Inc.
Suite L100, 239 - 10th Ave. SE
Calgary, AB  T2G 0V9

http://www.clickspace.com

On Apr 20, 2006, at 3:19 PM, Tom Lane wrote:

> Brendan Duddridge <brendan@clickspace.com> writes:
>> Do you mean do a kill -QUIT on the postgres process in order to
>> generate a stack trace?
>
> Not at all!  I'm talking about tracing the kernel calls it's making.
> Depending on your platform, the tool for this is called strace,
> ktrace, truss, or maybe even just trace.  With strace you'd do
> something like
>
>     strace -p PID-of-process 2>outfile
>     ... wait 30 sec or so ...
>     control-C
>
> Not sure about the APIs for the others but they're probably roughly
> similar ... read the man page ...
>
>             regards, tom lane
>
> ---------------------------(end of
> broadcast)---------------------------
> TIP 9: In versions below 8.0, the planner will ignore your desire to
>        choose an index scan if your joining column's datatypes do not
>        match
>