[Bucardo-general] Long delay before changes on remote master propogate to local
Paul Theodoropoulos
paul at anastrophe.com
Mon Sep 23 20:17:43 UTC 2013
On 9/22/13 11:18 AM, Greg Sabino Mullane wrote:
> On Thu, Sep 19, 2013 at 12:59:55PM -0700, Paul Theodoropoulos wrote:
>> Eliding lots of repetitious log output, the times are bothersome:
>>
>> start Bucardo:
>> (23733) [Thu Sep 19 12:12:02 2013] MCP Starting Bucardo version 5.0.0
>> Almost ten minutes later, Bucardo begins inspecting the local db
>> groups on one dbgroup:
>> (23734) [Thu Sep 19 12:21:31 2013] MCP Inspecting source table
>> "public.alarm_notification_history" on database "local_trumgr_group"
>> Bucardo goes quiet again, then begin inspecting another dbgroup
>> seven minutes later:
> Yeah, that's pretty extreme. Even with an insane number of tables
> and/or columns and/or targets, it should not be taking that long.
> So Bucardo is doing nothing at all for those first ten minutes according
> to the logs? Can you check the database logs and see if there is any
> query that might be extra-long running or blocked? Barring that, you might
> try stracing the MCP process during those gaps and seeing if you can figure
> out exactly what it is doing.
>
Yeah, it was just sitting there doing nothing, at least according to the
logs. I hooked in via strace - unfortunately I don't recall what the
first couple of lines were, but it just sat there doing nothing. I note
right now - while I'm setting up another test sync - the Bucardo Master
is looping through checking for fullstopbucardo - but it wasn't even
doing that as I recall.
I don't think the database is particularly large. One table has 1.6M
rows, and one of the db's has 100 tables each with about 13k rows.
Doesn't seem particularly large in relative terms.
There were numerous errors in the logs, but I believe they were
unrelated to this. I'm keeping a closer eye on the postgresql log during
testing today.
Just to confirm - all interval values in the bucardo configs are in
seconds (or fractions thereof), correct? So for example,
ctl_checkonkids_time at the default of 10 is ten seconds, not ten minutes.
And thanks for your help, by the way. It is appreciated.
--
Paul Theodoropoulos
www.anastrophe.com
More information about the Bucardo-general
mailing list