Обсуждение: Re: [ADMIN] Too frequent warnings for wraparound failure

Поиск
Список
Период
Сортировка

Re: [ADMIN] Too frequent warnings for wraparound failure

От
Tom Lane
Дата:
I wrote:
> "Milen A. Radev" <milen@securax.org> writes:
>> I review the log every morning. In the beginning I got "wraparound
>> failure" warnings every third day. But from a week I got those warnings
>> every day. Well we have one table in one database where there are a lot
>> of inserts, but not that many - around 30-40 thousand per day.

> Are you really doing half a billion transactions a day?

I thought of another mechanism that wouldn't require such a preposterous
load, only half a billion transactions since initdb.  (How old is this
installation, anyway, and which PG version?)

If you are creating new databases every day and you do it by cloning
template0, then the new databases would come into existence with
datfrozenxid equal to template0's.  Once template0 is more than half a
billion transactions old, you'd start seeing the warning.

This is relatively harmless, but probably we should try to mask it.
We could make CREATE DATABASE set datfrozenxid to current time when
cloning a database that has datallowconn false, on the assumption that
the source DB is entirely frozen and so there's nothing to vacuum yet.

            regards, tom lane

Re: [ADMIN] Too frequent warnings for wraparound failure

От
"Milen A. Radev"
Дата:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Tom Lane wrote:
> I wrote:
>
>>"Milen A. Radev" <milen@securax.org> writes:
>>
>>>I review the log every morning. In the beginning I got "wraparound
>>>failure" warnings every third day. But from a week I got those warnings
>>>every day. Well we have one table in one database where there are a lot
>>>of inserts, but not that many - around 30-40 thousand per day.
>
>
>>Are you really doing half a billion transactions a day?
>
>
> I thought of another mechanism that wouldn't require such a preposterous
> load, only half a billion transactions since initdb.  (How old is this
> installation, anyway, and which PG version?)

It's created on February 16 and is 7.4.7 on Debian GNU/Linux 3.0 (Woody)
(kernel 2.6.10).

>
> If you are creating new databases every day and you do it by cloning
> template0, then the new databases would come into existence with
> datfrozenxid equal to template0's.  Once template0 is more than half a
> billion transactions old, you'd start seeing the warning.

No, we do not do anything like that. We imported all databases from a
backup created by using pg_dump.

>
> This is relatively harmless, but probably we should try to mask it.
> We could make CREATE DATABASE set datfrozenxid to current time when
> cloning a database that has datallowconn false, on the assumption that
> the source DB is entirely frozen and so there's nothing to vacuum yet.
>
>             regards, tom lane



- --
Milen A. Radev
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.0 (MingW32)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFCMHqzqGlhYx0/sboRAr1dAKCB4vGHvzwsQ9zsM20y3hLrOSfqsQCgqVai
8RrSaVHjsoktDriCwCRWjfc=
=ju16
-----END PGP SIGNATURE-----