Subject: Re: Add client connection check during the execution of the query
To: Thomas Munro <thomas.munro@gmail.com>,
 "tsunakawa.takay@fujitsu.com" <tsunakawa.takay@fujitsu.com>
Cc: pgsql-hackers <pgsql-hackers@postgresql.org>,
 Tatsuo Ishii <ishii@sraoss.co.jp>, Tom Lane <tgl@sss.pgh.pa.us>,
 Andres Freund <andres@anarazel.de>, Zhihong Yu <zyu@yugabyte.com>
References: 
 <CA+hUKG+TGsHUZ4S6QOXr_ueKJ9-8d-TaoDpk9zdgjB_vHqotzQ@mail.gmail.com>
 <CA+hUKGJ3-N0F7T_tcRnGZv0_g_DJwcYM2=XTYCHrvcFG2YrrWg@mail.gmail.com>
 <20190718.121906.432767475415875714.t-ishii@sraoss.co.jp>
 <79f5f12b-184f-dcdb-b22f-4f95b6a89f2d@postgrespro.ru>
 <CA+hUKGL9LLNE8_+rBkRzBhmXA1XHEzJVqQr4T4fqzngqp66=-g@mail.gmail.com>
 <6bd6945e-f02e-8ee8-b8c4-e8d74e283efd@gmail.com>
 <CA+hUKGLyj5Aqt6ojYfSc+qSeB1x=3RbU61hnus5sL0BKqEBsLw@mail.gmail.com>
 <CA+hUKG+OuHdowPJ--cdN0+ZhPHiFqw0TDHp__N=Eu9yPsJtmyw@mail.gmail.com>
 <TYAPR01MB2990F4C319FC852CCED60394FE7B9@TYAPR01MB2990.jpnprd01.prod.outlook.com>
 <CA+hUKGLJ3PmhunHO4B+nFoxtJtodDo5H2GhP0zH2TVTfnmkJ7Q@mail.gmail.com>
 <TYAPR01MB2990E5F57212DBFD41F61983FE7A9@TYAPR01MB2990.jpnprd01.prod.outlook.com>
 <CA+hUKG++KitzNUOxW2-koB1pKWD2cyUqA9vLj5bf0g_i7L1M0w@mail.gmail.com>
 <CA+hUKG+tg5fWWvF5WbwEt4VO69Tj9o6F9mq-wd9kHHxDhUiBzw@mail.gmail.com>
 <CA+hUKGLYmBkROj=tUdXNeymxSdpBtfb=L5VcQKqUiZdvk2cBgg@mail.gmail.com>
From: Maksim Milyutin <milyutinma@gmail.com>
Message-ID: <f8547534-47ec-0f7c-fb6a-a7eb347f5a7e@gmail.com>
Date: Mon, 11 Oct 2021 17:10:29 +0300
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.13.0
MIME-Version: 1.0
In-Reply-To: 
 <CA+hUKGLYmBkROj=tUdXNeymxSdpBtfb=L5VcQKqUiZdvk2cBgg@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Content-Language: en-US
Archived-At: 
 <https://www.postgresql.org/message-id/f8547534-47ec-0f7c-fb6a-a7eb347f5a7e%40gmail.com>
Precedence: bulk

On 12.06.2021 07:24, Thomas Munro wrote:

> On Fri, Apr 30, 2021 at 2:23 PM Thomas Munro <thomas.munro@gmail.com> wrote:
>> Here's something I wanted to park here to look into for the next
>> cycle:  it turns out that kqueue's EV_EOF flag also has the right
>> semantics for this.  That leads to the idea of exposing the event via
>> the WaitEventSet API, and would the bring
>> client_connection_check_interval feature to 6/10 of our OSes, up from
>> 2/10.  Maybe Windows' FD_CLOSE event could get us up to 7/10, not
>> sure.
> Rebased.  Added documentation tweak and a check to reject the GUC on
> unsupported OSes.


Good work. I have tested your patch on Linux and FreeBSD on three basic 
cases: client killing, cable breakdown (via manipulations with firewall) 
and silent closing client connection before completion of previously 
started query in asynchronous manner. And all works fine.

Some comments from me:

  bool
  pq_check_connection(void)
  {
-#if defined(POLLRDHUP)
+    WaitEvent events[3];


3 is looks like as magic constant. We might to specify a constant for 
all event groups in FeBeWaitSet.


+    ModifyWaitEvent(FeBeWaitSet, FeBeWaitSetLatchPos, WL_LATCH_SET, NULL);
+    ModifyWaitEvent(FeBeWaitSet, FeBeWaitSetSocketPos, 
WL_SOCKET_CLOSED, NULL);
+    rc = WaitEventSetWait(FeBeWaitSet, 0, events, lengthof(events), 0);
+    ModifyWaitEvent(FeBeWaitSet, FeBeWaitSetLatchPos, WL_LATCH_SET, 
MyLatch);

AFAICS, side effect to 
(FeBeWaitSet->events[FeBeWaitSetSocketPos]).events by setting 
WL_SOCKET_CLOSED value under calling of pq_check_connection() function 
doesn't have negative impact later, does it? That is, all 
WaitEventSetWait() calls have to setup socket events on its own from 
scratch.


-- 
Regards,
Maksim Milyutin