Обсуждение: remove checkpoint_warning
the checkpoint_warning feature was added by commit 2986aa6a668bce3cfb836 in November 2002 when we didn't have any logging of checkpointing at all. I propose to remove it: surely anyone who cares about analyzing checkpointing behavior nowadays is using the log_checkpoint feature instead, which contains much more detail. The other one is just noise now, and probably ignored amidst the number of other warning traffic. -- Álvaro Herrera http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
On 2016-07-09 16:46:32 -0400, Alvaro Herrera wrote: > the checkpoint_warning feature was added by commit 2986aa6a668bce3cfb836 > in November 2002 when we didn't have any logging of checkpointing at > all. I propose to remove it: surely anyone who cares about analyzing > checkpointing behavior nowadays is using the log_checkpoint feature > instead, which contains much more detail. The other one is just noise > now, and probably ignored amidst the number of other warning traffic. I think we could do that, if we enable log_checkpoints by default, but right now we don't...
Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> the checkpoint_warning feature was added by commit 2986aa6a668bce3cfb836
> in November 2002 when we didn't have any logging of checkpointing at
> all. I propose to remove it: surely anyone who cares about analyzing
> checkpointing behavior nowadays is using the log_checkpoint feature
> instead, which contains much more detail. The other one is just noise
> now, and probably ignored amidst the number of other warning traffic.
Hmm, not sure. ISTM log_checkpoint is oriented to people who know what
they are doing, whereas checkpoint_warning is more targeted to trying
to help people who don't. Perhaps you could make an argument that
checkpoint_warning is useless because the people whom it's meant to help
won't notice the warning anyway --- but I doubt that it's been
"superseded" by log_checkpoint, because the latter would only be enabled
by people who already have a clue that checkpoint performance is something
to worry about.
Or in short, this may be a fine change to make, but I don't like your
argument for it.
regards, tom lane
* Tom Lane (tgl@sss.pgh.pa.us) wrote: > Alvaro Herrera <alvherre@2ndquadrant.com> writes: > > the checkpoint_warning feature was added by commit 2986aa6a668bce3cfb836 > > in November 2002 when we didn't have any logging of checkpointing at > > all. I propose to remove it: surely anyone who cares about analyzing > > checkpointing behavior nowadays is using the log_checkpoint feature > > instead, which contains much more detail. The other one is just noise > > now, and probably ignored amidst the number of other warning traffic. > > Hmm, not sure. ISTM log_checkpoint is oriented to people who know what > they are doing, whereas checkpoint_warning is more targeted to trying > to help people who don't. Perhaps you could make an argument that > checkpoint_warning is useless because the people whom it's meant to help > won't notice the warning anyway --- but I doubt that it's been > "superseded" by log_checkpoint, because the latter would only be enabled > by people who already have a clue that checkpoint performance is something > to worry about. > > Or in short, this may be a fine change to make, but I don't like your > argument for it. I don't agree that it's sensible to get rid of. Having just log_checkpoints will have the logs filled with checkpoints starting because of XLOG, but there's no indication of that being a bad thing. Thanks! Stephen
On Mon, Jul 11, 2016 at 7:25 AM, Stephen Frost <sfrost@snowman.net> wrote: >> Or in short, this may be a fine change to make, but I don't like your >> argument for it. > > I don't agree that it's sensible to get rid of. Having just > log_checkpoints will have the logs filled with checkpoints starting > because of XLOG, but there's no indication of that being a bad thing. I agree. checkpoint_warning exists for the benefit of novice DBAs. I've seen those warnings in customer logs on several occasions, at least back when I was a consultant. -- Peter Geoghegan
On 2016-07-11 11:14:29 -0700, Peter Geoghegan wrote: > On Mon, Jul 11, 2016 at 7:25 AM, Stephen Frost <sfrost@snowman.net> wrote: > >> Or in short, this may be a fine change to make, but I don't like your > >> argument for it. > > > > I don't agree that it's sensible to get rid of. Having just > > log_checkpoints will have the logs filled with checkpoints starting > > because of XLOG, but there's no indication of that being a bad thing. > > I agree. checkpoint_warning exists for the benefit of novice DBAs. > I've seen those warnings in customer logs on several occasions, at > least back when I was a consultant. Note that the situation changed a bit with 9.5, because our defaults aren't absurdly conservative (checkpoint_segments = 3) anymore.
* Andres Freund (andres@anarazel.de) wrote: > On 2016-07-11 11:14:29 -0700, Peter Geoghegan wrote: > > On Mon, Jul 11, 2016 at 7:25 AM, Stephen Frost <sfrost@snowman.net> wrote: > > >> Or in short, this may be a fine change to make, but I don't like your > > >> argument for it. > > > > > > I don't agree that it's sensible to get rid of. Having just > > > log_checkpoints will have the logs filled with checkpoints starting > > > because of XLOG, but there's no indication of that being a bad thing. > > > > I agree. checkpoint_warning exists for the benefit of novice DBAs. > > I've seen those warnings in customer logs on several occasions, at > > least back when I was a consultant. > > Note that the situation changed a bit with 9.5, because our defaults > aren't absurdly conservative (checkpoint_segments = 3) anymore. Agreed, but I don't think that means we'll never see that warning again.. Thanks! Stephen
On 11 July 2016 at 22:25, Stephen Frost <sfrost@snowman.net> wrote:
* Tom Lane (tgl@sss.pgh.pa.us) wrote:
> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> > the checkpoint_warning feature was added by commit 2986aa6a668bce3cfb836
> > in November 2002 when we didn't have any logging of checkpointing at
> > all. I propose to remove it: surely anyone who cares about analyzing
> > checkpointing behavior nowadays is using the log_checkpoint feature
> > instead, which contains much more detail. The other one is just noise
> > now, and probably ignored amidst the number of other warning traffic.
>
> Hmm, not sure. ISTM log_checkpoint is oriented to people who know what
> they are doing, whereas checkpoint_warning is more targeted to trying
> to help people who don't. Perhaps you could make an argument that
> checkpoint_warning is useless because the people whom it's meant to help
> won't notice the warning anyway --- but I doubt that it's been
> "superseded" by log_checkpoint, because the latter would only be enabled
> by people who already have a clue that checkpoint performance is something
> to worry about.
>
> Or in short, this may be a fine change to make, but I don't like your
> argument for it.
I don't agree that it's sensible to get rid of. Having just
log_checkpoints will have the logs filled with checkpoints starting
because of XLOG, but there's no indication of that being a bad thing.
Also, the warning is greppable-for and easily spotted by log ingesting tools. I see no real reason to remove it.
On Tue, Jul 12, 2016 at 3:32 AM, Craig Ringer <craig@2ndquadrant.com> wrote:
On 11 July 2016 at 22:25, Stephen Frost <sfrost@snowman.net> wrote:* Tom Lane (tgl@sss.pgh.pa.us) wrote:
> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> > the checkpoint_warning feature was added by commit 2986aa6a668bce3cfb836
> > in November 2002 when we didn't have any logging of checkpointing at
> > all. I propose to remove it: surely anyone who cares about analyzing
> > checkpointing behavior nowadays is using the log_checkpoint feature
> > instead, which contains much more detail. The other one is just noise
> > now, and probably ignored amidst the number of other warning traffic.
>
> Hmm, not sure. ISTM log_checkpoint is oriented to people who know what
> they are doing, whereas checkpoint_warning is more targeted to trying
> to help people who don't. Perhaps you could make an argument that
> checkpoint_warning is useless because the people whom it's meant to help
> won't notice the warning anyway --- but I doubt that it's been
> "superseded" by log_checkpoint, because the latter would only be enabled
> by people who already have a clue that checkpoint performance is something
> to worry about.
>
> Or in short, this may be a fine change to make, but I don't like your
> argument for it.
I don't agree that it's sensible to get rid of. Having just
log_checkpoints will have the logs filled with checkpoints starting
because of XLOG, but there's no indication of that being a bad thing.Also, the warning is greppable-for and easily spotted by log ingesting tools. I see no real reason to remove it.
+1, this is a useful warning.
I'd flip the original argument over instead and say that for *most* people, log_checkpoint output is mostly noise, and it's checkpoint_warnings that's actually interesting. That on tells them if they have to look at anything at all. It's true that those that care about *analyzing* their checkpointing behaviour need to turn on log_checkpoint, but it's checkpoint_warnings that tells them that they have to do this.
And the argument about it getting lost amongst other log traffic to me is an argument for not turning on log_checkpoints by default. That generates a lot of log entries that most people don't care for, and in doing so hides other things in the log. It's excellent to have it once you need it, but it shouldn't be turned on by default.
On 07/09/2016 11:12 PM, Tom Lane wrote:
> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
>> the checkpoint_warning feature was added by commit 2986aa6a668bce3cfb836
>> in November 2002 when we didn't have any logging of checkpointing at
>> all. I propose to remove it: surely anyone who cares about analyzing
>> checkpointing behavior nowadays is using the log_checkpoint feature
>> instead, which contains much more detail. The other one is just noise
>> now, and probably ignored amidst the number of other warning traffic.
> Hmm, not sure. ISTM log_checkpoint is oriented to people who know what
> they are doing, whereas checkpoint_warning is more targeted to trying
> to help people who don't. Perhaps you could make an argument that
> checkpoint_warning is useless because the people whom it's meant to help
> won't notice the warning anyway --- but I doubt that it's been
> "superseded" by log_checkpoint, because the latter would only be enabled
> by people who already have a clue that checkpoint performance is something
> to worry about.
>
> Or in short, this may be a fine change to make, but I don't like your
> argument for it.
>
> regards, tom lane
>
>
i think tom is right here.
log_checkpoint and checkpoint_warning are for totally different people.
we might just want to do one thing: we might want to state explicitly
that the database cannot break down if this warning shows up.
many people are scared to death that this warning somehow indicates that
PostgreSQL is about to go up in flames, which is of course not true.
maybe we could do "consider increasing .... to ensure good performance"
or so ...
regards,
hans
--
Hans-Jürgen Schönig
Cybertec Schönig & Schönig GmbH
Gröhrmühlgasse 26
A-2700 Wiener Neustadt
Web: http://www.postgresql-support.de, http://www.cybertec.at
On Tue, Jul 12, 2016 at 3:19 AM, Magnus Hagander <magnus@hagander.net> wrote: > +1, this is a useful warning. +1. I'd like to see more people turn log_checkpoints=on, and I often ask customers to do that when they're having systemic performance problems. But this warning regularly alerts me to cases where I've failed to configure max_wal_size to an adequately large value. It's true that the new values are less conservative than the old ones, but they're still pretty conservative. Anybody who still thinks that 1GB is a lot of disk space might be due for a hardware upgrade. I regularly need to increase that value substantially on my *laptop*. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company