On 2020/08/20 10:43, Andres Freund wrote:
> Hi,
>
> On 2020-08-19 21:29:06 -0400, Tom Lane wrote:
>> Andres Freund <andres@anarazel.de> writes:
>>> On 2020-08-19 11:01:37 -0400, Tom Lane wrote:
>>>> I agree with that, but I think this patch has a bigger problem:
>>>> why bother at all? It seems like a waste of code space and future
>>>> maintenance effort, because there is no use-case.
>>
>>> I don't agree with this at all. I think there's plenty use cases. It's
>>> e.g. very common to try to figure out why the memory usage of a process
>>> is high. Is it memory not returned to the OS? Is it caches that have
>>> grown too much etc.
>>
>> Oh, I agree completely that there are lots of use-cases for finding
>> out what a process' memory map looks like. But this patch fails to
>> address any of them in a usable way.
>
> Even just being able to see the memory usage in a queryable way is a
> huge benefit.
+1
> The difference over having to parse the log, then parse
> the memory usage dump, and then aggregate the data in there in a
> meaningful way is *huge*. We've been slacking around lowering our
> memory usage, and I think the fact that it's annoying to analyze is a
> partial reason for that.
Agreed.
> I totally agree that it's not *enough*, but in contrast to you I think
> it's a good step. Subsequently we should add a way to get any backends
> memory usage.
> It's not too hard to imagine how to serialize it in a way that can be
> easily deserialized by another backend. I am imagining something like
> sending a procsignal that triggers (probably at CFR() time) a backend to
> write its own memory usage into pg_memusage/<pid> or something roughly
> like that.
Sounds good. Maybe we can also provide the SQL-callable function
or view to read pg_memusage/<pid>, to make the analysis easier.
Regards,
--
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION