Re: Causeless CPU load waves in backend, on windows, 9.5.5 (EDB binary).

From: Nikolai Zhubr <n-a-zhubr(at)yandex(dot)ru>
To: pgsql-general(at)postgresql(dot)org
Subject: Re: Causeless CPU load waves in backend, on windows, 9.5.5 (EDB binary).
Date: 2017-02-08 12:50:48
Message-ID: 589B1428.7070100@yandex.ru
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

05.02.2017 22:05, I wrote:
[...]
>> And yes, running Process Explorer gave some new and unexpected input.
>> During the period of this strange high load it claims 40% CPU is used by
>> interrupts (normally 0.01%) and 3% used by backend postgres.exe
>> (normally approx 0%). I'd guess this means some problem happening in the
>> OS (which hosts this postgres.exe), probably related to network
>> communication? (Because nothing else seems likely related to interrupts
>> in such scenario?)

Ok, I've got a working example as simple as the following:

do {
for (i=0; i<40; i++) {
Sleep(15);
res = PQexec(conn, "SELECT localtimestamp ");
if (PQresultStatus(res) != PGRES_TUPLES_OK) return 1;
PQclear(res);
}
fprintf(stdout, "*");
Sleep(350);
} while(1);

Here, Sleep(15) and Sleep(350) are very important, removing or changing
them can cause the effect to disappear.
Looks like some timing-sensitive issue in windows TCP/IP implementation?
But then, how can it be that no-one have noticed it yet? Puzzling.

Thank you.

Regards,
Nikolai

>
> Additionally, I've now got a kernrate viewer reports, one for high load,
> and one for normal load period, attached below. Here, tcpip shows some
> more kernel activity during the problematic period, but the difference
> is not so huge.
>
> Time 36348 hits, 25000 events per hit --------
> Module Hits msec %Total Events/Sec
> intelppm 35048 45640 96 % 19198071
> hal 1030 45640 2 % 564198
> ntkrnlpa 204 45640 0 % 111744
> tcpip 20 45640 0 % 10955
> win32k 18 45640 0 % 9859
> afd 6 45640 0 % 3286
> ipnat 6 45640 0 % 3286
> NDIS 4 45640 0 % 2191
> ......
>
> Time 37227 hits, 25000 events per hit --------
> Module Hits msec %Total Events/Sec
> intelppm 35856 46828 96 % 19142393
> hal 1089 46828 2 % 581382
> ntkrnlpa 229 46828 0 % 122255
> win32k 29 46828 0 % 15482
> tcpip 9 46828 0 % 4804
> NDIS 4 46828 0 % 2135
> afd 3 46828 0 % 1601
> psched 3 46828 0 % 1601
> ipnat 2 46828 0 % 1067
>
>
> Thank you.
>
> Nikolai
>
>>
>> The screenshot: https://yadi.sk/i/hC8FMZCE3CyzPs
>>
>>
>> Thank you.
>>
>> Nikolai
>>
>>
>>>
>>> Andres
>>>
>>
>
>
>

In response to

Browse pgsql-general by date

  From Date Subject
Next Message Tom Lane 2017-02-08 15:44:24 Re: Running out of memory the hard way ...
Previous Message Albe Laurenz 2017-02-08 09:53:10 Re: Running out of memory the hard way ...