Обсуждение: Measurin the lag between a master and a replica

Поиск
Список
Период
Сортировка

Measurin the lag between a master and a replica

От
Ali Pouya
Дата:
Hi,
I am using a master and a replica with PostgreSQL 9.1.3.
I would like to measure the replication lag between my servers (in time units).

On the replica I can use the function pg_last_xact_replay_timestamp().
But on the master I only can use pg_current_xlog_location() which does not return a timestamp. I do not know any function returning the timestamp.

Is there any way to get the timestamp relative to the result of pg_current_xlog_location() ?

Thanks for your help
Best regards
Ali Pouya

Re: Measurin the lag between a master and a replica

От
Sergey Konoplev
Дата:
On Thu, Jul 19, 2012 at 7:17 PM, Ali Pouya <alipouya2@gmail.com> wrote:
> On the replica I can use the function pg_last_xact_replay_timestamp().
> But on the master I only can use pg_current_xlog_location() which does not
> return a timestamp. I do not know any function returning the timestamp.
>
> Is there any way to get the timestamp relative to the result of
> pg_current_xlog_location() ?

According to mine own research of this problem there are no way to get it.

What I would do to monitor the replication lag is

1. execute this on master SELECT txid_current(), that will create a
minimal wal entry
2. compare pg_last_xact_replay_timestamp() with current time on replica.

assuming that clocks are in sync on both servers of course.

--
Sergey Konoplev

a database architect, software developer at PostgreSQL-Consulting.com
http://www.postgresql-consulting.com

Jabber: gray.ru@gmail.com Skype: gray-hemp Phone: +79160686204

Re: Measurin the lag between a master and a replica

От
David Kerr
Дата:
On Jul 19, 2012, at 8:17 AM, Ali Pouya wrote:

> Hi,
> I am using a master and a replica with PostgreSQL 9.1.3.
> I would like to measure the replication lag between my servers (in time units).
>
> On the replica I can use the function pg_last_xact_replay_timestamp().
> But on the master I only can use pg_current_xlog_location() which does not return a timestamp. I do not know any
functionreturning the timestamp. 
>
> Is there any way to get the timestamp relative to the result of pg_current_xlog_location() ?
>
> Thanks for your help
> Best regards
> Ali Pouya
>

I was looking into this last month and some nice folks on the list pointed me to this formula
for calculating lag. (to be run on the slave) similar to what Sergey said.

    when pg_last_xlog_receive_location() = pg_last_xlog_replay_location()
    then 0
    else extract ('epoch' from now()-pg_last_xact_replay_timestamp()) end as log_delay;



Re: Measurin the lag between a master and a replica

От
Sergey Konoplev
Дата:
On Fri, Jul 20, 2012 at 1:02 AM, David Kerr <dmk@mr-paradox.net> wrote:
> I was looking into this last month and some nice folks on the list pointed me to this formula
> for calculating lag. (to be run on the slave) similar to what Sergey said.
>
>         when pg_last_xlog_receive_location() = pg_last_xlog_replay_location()
>         then 0
>         else extract ('epoch' from now()-pg_last_xact_replay_timestamp()) end as log_delay;

If you have a connection problem this will not reflect the situation.

--
Sergey Konoplev

a database architect, software developer at PostgreSQL-Consulting.com
http://www.postgresql-consulting.com

Jabber: gray.ru@gmail.com Skype: gray-hemp Phone: +79160686204

Re: Measurin the lag between a master and a replica

От
David Kerr
Дата:
On Jul 19, 2012, at 9:03 PM, Sergey Konoplev wrote:

> On Fri, Jul 20, 2012 at 1:02 AM, David Kerr <dmk@mr-paradox.net> wrote:
>> I was looking into this last month and some nice folks on the list pointed me to this formula
>> for calculating lag. (to be run on the slave) similar to what Sergey said.
>>
>>        when pg_last_xlog_receive_location() = pg_last_xlog_replay_location()
>>        then 0
>>        else extract ('epoch' from now()-pg_last_xact_replay_timestamp()) end as log_delay;
>
> If you have a connection problem this will not reflect the situation.


Good caveat! monitor the connection status separately as well!