Обсуждение: Hi- How frequently Postgres Poll for trigger file

On 01.12.2010 13:27, aaliya zarrin wrote:
> I want to know how frequently postgres search for trigger file to switch
> over.

In 9.0, every 100ms while streaming replication is active and connected. 
5 seconds otherwise. In current git master branch, it's always 5 s.

> Can this switch over time be reduced?

Not without hacking the sources and compiling.

Although, on many platforms, Linux included I believe, sending a signal 
to the startup process should wake it up from the sleep and make it 
check the trigger file immediately. "pg_ctl reload" for example should 
do it. So if ýou send a signal to the startup process immediately after 
creating the trigger file, it should take notice sooner.

> Plz let me know where postgres poll for trigger file. I could find it out in
> "backend/access/trans/xlog.c" ? am i right?

Yes. search for callers of CheckForStandbyTrigger() function.


--   Heikki Linnakangas  EnterpriseDB   http://www.enterprisedb.com

Heikki Linnakangas escreveu:
> On 01.12.2010 13:27, aaliya zarrin wrote:
>> I want to know how frequently postgres search for trigger file to switch
>> over.
> 
> In 9.0, every 100ms while streaming replication is active and connected.
> 5 seconds otherwise. In current git master branch, it's always 5 s.
> 
>> Can this switch over time be reduced?
> 
> Not without hacking the sources and compiling.
> 
> Although, on many platforms, Linux included I believe, sending a signal
> to the startup process should wake it up from the sleep and make it
> check the trigger file immediately. "pg_ctl reload" for example should
> do it. So if ýou send a signal to the startup process immediately after
> creating the trigger file, it should take notice sooner.
> 
Isn't it an ugly solution for stopping the replication immediately? At the top
of my head, I don't remember the reason for not turn the interval for pooling
trigger file into a configurable option. IMHO, high availability fits into
those cases (switchover immediately).

[poking the git history a little...]

I agree the a short polling interval is not energy efficient [1] but some
scenarios need this short interval. So if we want to stop the replication
immediately, we have two options: (i) advertise that we need to signal the
postmaster after creating a trigger file or (ii) made the pooling interval
configurable. As you said, there are platforms that a signal doesn't wake up a
process, so I suggest (ii) but I'm fine to include (i) at docs too.

Comments?


[1]
http://git.postgresql.org/gitweb?p=postgresql.git;a=commitdiff;h=723d0184e2972f21db0f85feef3d35f0cb9b3298


--  Euler Taveira de Oliveira http://www.timbira.com/

Thanks for quick response..<br />Can I change this 5 second time?<br />I have seen the postgres code as well.<br />What
isthe functionality of WaitLatch() function. I could not understand completely.<br />Plz help..<br /><br /><div
class="gmail_quote">On Wed, Dec 1, 2010 at 5:53 PM, Heikki Linnakangas <span dir="ltr"><<a
href="mailto:heikki.linnakangas@enterprisedb.com">heikki.linnakangas@enterprisedb.com</a>></span>wrote:<br
/><blockquoteclass="gmail_quote" style="margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204);
padding-left:1ex;"><div class="im">On 01.12.2010 13:27, aaliya zarrin wrote:<br /><blockquote class="gmail_quote"
style="margin:0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"> I want to know how
frequentlypostgres search for trigger file to switch<br /> over.<br /></blockquote><br /></div> In 9.0, every 100ms
whilestreaming replication is active and connected. 5 seconds otherwise. In current git master branch, it's always 5
s.<divclass="im"><br /><br /><blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid
rgb(204,204, 204); padding-left: 1ex;"> Can this switch over time be reduced?<br /></blockquote><br /></div> Not
withouthacking the sources and compiling.<br /><br /> Although, on many platforms, Linux included I believe, sending a
signalto the startup process should wake it up from the sleep and make it check the trigger file immediately. "pg_ctl
reload"for example should do it. So if ıou send a signal to the startup process immediately after creating the trigger
file,it should take notice sooner.<div class="im"><br /><br /><blockquote class="gmail_quote" style="margin: 0pt 0pt
0pt0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"> Plz let me know where postgres poll for
triggerfile. I could find it out in<br /> "backend/access/trans/xlog.c" ? am i right?<br /></blockquote><br /></div>
Yes.search for callers of CheckForStandbyTrigger() function.<br /><font color="#888888"><br /><br /> -- <br />  Heikki
Linnakangas<br/>  EnterpriseDB   <a href="http://www.enterprisedb.com"
target="_blank">http://www.enterprisedb.com</a><br/></font></blockquote></div><br /><br clear="all" /><br />-- <br
/>Thanks& Regards,<br /><br />Aaliya Zarrin<br />(+91)-9160665888<br /> 

On 01.12.2010 19:23, aaliya zarrin wrote:
> Thanks for quick response..
> Can I change this 5 second time?
> I have seen the postgres code as well.

You can, if you don't mind changing the sources.

> What is the functionality of WaitLatch() function. I could not understand
> completely.

The recoveryWakupLatch latch allows the startup process to wake up 
immediately when new WAL arrives via streaming replication, without 
waiting for the whole five seconds to elapse.

--   Heikki Linnakangas  EnterpriseDB   http://www.enterprisedb.com

On Thu, Dec 2, 2010 at 12:16 AM, Euler Taveira de Oliveira
<euler@timbira.com> wrote:
> As you said, there are platforms that a signal doesn't wake up a
> process, so I suggest (ii) but I'm fine to include (i) at docs too.

Can we use "pg_ctl kill" to send signal on such platforms?

Regards,

-- 
Fujii Masao
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center

On 03.12.2010 03:55, Fujii Masao wrote:
> On Thu, Dec 2, 2010 at 12:16 AM, Euler Taveira de Oliveira
> <euler@timbira.com>  wrote:
>> As you said, there are platforms that a signal doesn't wake up a
>> process, so I suggest (ii) but I'm fine to include (i) at docs too.
>
> Can we use "pg_ctl kill" to send signal on such platforms?

It won't make a difference, "pg_ctl kill" sends the signal just like 
regular "kill". Except that "pg_ctl kill" also works on Windows which 
doesn't have the concept of Unix signals.

--   Heikki Linnakangas  EnterpriseDB   http://www.enterprisedb.com