From: Tomas Vondra <tomas.vondra@2ndquadrant.com>
To: Scott Marlowe <scott.marlowe@gmail.com>
To: Joshua D. Drake <jd@commandprompt.com>
Cc: pgsql-performance@postgresql.org <pgsql-performance@postgresql.org>
Subject: Re: How to reduce writing on disk ? (90 gb on pgsql_tmp)
Date: Thu, 04 Jun 2015 16:43:13 +0200
Message-ID: <55706401.4080707@2ndquadrant.com> (raw)
In-Reply-To: <CAOR=d=3WjFfjwK-ne2P-J4senyoLwxUKy5GSX3+wmXSKzwH8Sg@mail.gmail.com>
References: <1433334092106-5852321.post@n5.nabble.com>
<1433337082772-5852328.post@n5.nabble.com>
<CA+RSxMiSwse+YfMoFT8128PoWG3kifmrvg4rhdBHxUTE3HPH-g@mail.gmail.com>
<556F0649.9000402@2ndquadrant.com>
<9DED5F42-2C51-4CDE-A179-56A8A9370338@playrion.com>
<556F15AD.4040801@2ndquadrant.com>
<CAOR=d=20bFbC-LnrzqCL2-c79q-8hmHAf6zv0Roh8r=tmxafrA@mail.gmail.com>
<556F547D.5040206@2ndquadrant.com>
<CAOR=d=2bMfcv9d5Lok5WddVm7ztM4a1QNcxm9Tcpx+PChjQunQ@mail.gmail.com>
<556F7CAA.2080003@2ndquadrant.com>
<556F7FD8.5070802@commandprompt.com>
<CAOR=d=2vMYtYfJhRQoHMPzPBi22omk46XRDa2EcTVJYT8YCFpw@mail.gmail.com>
<CAOR=d=3WjFfjwK-ne2P-J4senyoLwxUKy5GSX3+wmXSKzwH8Sg@mail.gmail.com>
List-Unsubscribe: <mailto:majordomo@postgresql.org?body=unsub%20pgsql-performance>
On 06/04/15 02:58, Scott Marlowe wrote:
> On Wed, Jun 3, 2015 at 6:53 PM, Scott Marlowe <scott.marlowe@gmail.com> wrote:
>> On Wed, Jun 3, 2015 at 4:29 PM, Joshua D. Drake <jd@commandprompt.com> wrote:
>>>
>>> On 06/03/2015 03:16 PM, Tomas Vondra wrote:
>>>
>>>> Cache is not free memory - it's there for a purpose and usually
>>>> plays a significant role in performance. Sure, it may be freed
>>>> and used for other purposes, but that has consequences - e.g.
>>>> it impacts performance of other queries etc. You generally
>>>> don't want to do that onproduction.
>>>
>>>
>>> Exactly. If your cache is reduced your performance is reduced
>>> because less things are in cache. It is not free memory. Also the
>>> command "free" is not useful in this scenario. It is almost
>>> alwaysbetter to use sar so you can see where the data points are
>>> thatfree is using.
>>
>> But if that WAS happening he wouldn't still HAVE 60G of cache!
>> That's my whole point. He's NOT running out of memory. He's not
>> even having to dump cache right now.
No one claimed he's running out of memory ...
What I claimed is that considering page cache equal to free memory is
not really appropriate, because it is used for caching data, which plays
a significant role.
Regarding the "free" output, we have no clue when the "free" command was
executed. I might have been executed while the query was running, right
after it failed or long after that. That has significant impact on
interpretation of the output.
Also, we have no clue what happens on the machine, so it's possible
there are other queries competing for the page cache, quickly filling
reusing free memory (used for large query moments ago) for page cache.
And finally, we have no clue what plan the query is using, so we don't
know how much memory it's using before it starts spilling to disk. For
example it might easily be a single sort node, taking only 384MB (the
work_mem) of RAM before it starts spilling to disk.
> Further if he started using a few gig here for this one it wouldn't
> have a big impact on cache (60G-1G etc) but might make it much
> faster, as spilling to disk is a lot less intrusive when you've got a
> bigger chunk of ram to work in. OTOH doing something like setting
> work_mem to 60G would likely be fatal.
It'd certainly be fatal, because this query is spilling >95G to disk,
and keeping that in memory would easily require 2-3x more space.
>
> But he's not down to 3GB of memory by any kind of imagination. Any
> working machine will slowly, certainly fill its caches since it's
> not using the memory for anything else. That's normal. As long as
> you're not blowing out the cache you're fine.
Once again, what about the 15GB shared buffers? Not that it'd change
anything, really.
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
--
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-performance@postgresql.org
Cc: tomas.vondra@2ndquadrant.com, scott.marlowe@gmail.com, jd@commandprompt.com
Subject: Re: How to reduce writing on disk ? (90 gb on pgsql_tmp)
In-Reply-To: <55706401.4080707@2ndquadrant.com>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox