MIME-Version: 1.0
References: 
 <CAJhEC07OK8J7tLUbyiccnuOXRE7UKxBNqD2-pLfeFXa=tBoWtw@mail.gmail.com>
 <acNIQaeIg88xkCNC@paquier.xyz>
 <6ec95d6e.15a854.19d22f55f00.Coremail.18663776784@163.com>
In-Reply-To: <6ec95d6e.15a854.19d22f55f00.Coremail.18663776784@163.com>
From: Robert Pang <robertpang@google.com>
Date: Wed, 25 Mar 2026 21:05:34 -0700
Message-ID: 
 <CAJhEC05t1JHWEWtmji2u5n8g_+WmROuF5YuN7UEbOA=cL16zXQ@mail.gmail.com>
Subject: Re: [PATCH] Fix premature timeout in pg_promote() caused by signal
 interruptions
To: getiancheng_2012 <18663776784@163.com>
Cc: Michael Paquier <michael@paquier.xyz>,
 pgsql-hackers <pgsql-hackers@postgresql.org>
Content-Type: multipart/alternative; boundary="000000000000ccd3a4064de57f2a"
Archived-At: 
 <https://www.postgresql.org/message-id/CAJhEC05t1JHWEWtmji2u5n8g_%2BWmROuF5YuN7UEbOA%3DcL16zXQ%40mail.gmail.com>
Precedence: bulk

--000000000000ccd3a4064de57f2a
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Hi Michael and Tiancheng

On Tue, Mar 24, 2026 at 8:06=E2=80=AFPM getiancheng_2012 <18663776784@163.c=
om>
wrote:

> ---- Replied Message ----
> From Michael Paquier<michael@paquier.xyz> <michael@paquier.xyz>
> Date 3/25/2026 10:28
> To Robert Pang<robertpang@google.com> <robertpang@google.com>
> Cc pgsql-hackers<pgsql-hackers@postgresql.org>
> <pgsql-hackers@postgresql.org>
> Subject Re: [PATCH] Fix premature timeout in pg_promote() caused by
> signal interruptions
>
> On Wed, Mar 11, 2026 at 09:44:07AM -0700, Robert Pang wrote:
>
> The current implementation of pg_promote() calculates a fixed number
>  of loop iterations based on the timeout value, assuming each loop
>  waits exactly 100 ms for the backend latch. However, if the backend
>  receives an unrelated signal (e.g., from
>  client_connection_check_interval), it wakes up early. These repeated,
>  unrelated wakeups cause the loop counter to deplete much faster than
>  intended, leading to a premature timeout.
>
>
> It is true that we can do better here, and your proposal about having
> a more precise timeout calculation looks like a sensible improvement
> for this case.
>
> No objections regarding your patch.  I would like to apply it on HEAD,
> if there are no objections.
> --
> Michael
>
>
> Hi all,
>
> Overall LGTM. Just a small comment:
>
> "+  end_time =3D GetCurrentTimestamp() + wait_seconds * 1000000L;"
>
> I think we can use TimestampTzPlusSeconds(GetCurrentTimestamp(), wait_sec=
onds).
>
>
> Best regards
>
> Tiancheng Ge
>
>
Thank you for reviewing this patch. The use of TimestampTzPlusSeconds()
will be good.

Best regards
Robert Pang

--000000000000ccd3a4064de57f2a
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div dir=3D"ltr">Hi Michael and=C2=A0Tiancheng</div><br><d=
iv class=3D"gmail_quote gmail_quote_container"><div dir=3D"ltr" class=3D"gm=
ail_attr">On Tue, Mar 24, 2026 at 8:06=E2=80=AFPM getiancheng_2012 &lt;<a h=
ref=3D"mailto:18663776784@163.com">18663776784@163.com</a>&gt; wrote:<br></=
div><blockquote class=3D"gmail_quote" style=3D"margin:0px 0px 0px 0.8ex;bor=
der-left:1px solid rgb(204,204,204);padding-left:1ex"><div style=3D"line-he=
ight:1.5;font-family:=E5=BE=AE=E8=BD=AF=E9=9B=85=E9=BB=91"><div style=3D"fo=
nt-family:=E5=BE=AE=E8=BD=AF=E9=9B=85=E9=BB=91;line-height:1.5"><div><span =
style=3D"color:rgb(120,120,122);font-size:14px">---- Replied Message ----</=
span></div></div><div style=3D"padding-top:1px;padding-bottom:1px;font-fami=
ly:=E5=BE=AE=E8=BD=AF=E9=9B=85=E9=BB=91;line-height:1.5"><div style=3D"marg=
in-bottom:1em;font-size:12px;line-height:1.25;color:rgb(35,35,36);padding:0=
.5em 0.25em;border-radius:0.67em;background-color:rgb(240,240,240)"><table =
width=3D"100%" cellpadding=3D"0" cellspacing=3D"9" border=3D"0" id=3D"m_583=
6655200720777922ntes_editor_table_10010"><tbody><tr><td valign=3D"top" styl=
e=3D"width:4.25em;font-size:12px;line-height:1.25;color:rgb(120,120,122)">F=
rom</td><td valign=3D"top" style=3D"font-size:12px;line-height:1.25;color:r=
gb(35,35,36);word-break:break-all"><a style=3D"color:rgb(48,97,242);text-de=
coration:none" href=3D"mailto:michael@paquier.xyz" target=3D"_blank">Michae=
l Paquier&lt;michael@paquier.xyz&gt;</a></td></tr><tr><td valign=3D"top" st=
yle=3D"width:4.25em;font-size:12px;line-height:1.25;color:rgb(120,120,122)"=
>Date</td><td valign=3D"top" style=3D"font-size:12px;line-height:1.25;color=
:rgb(35,35,36);word-break:break-all">3/25/2026 10:28</td></tr><tr><td valig=
n=3D"top" style=3D"width:4.25em;font-size:12px;line-height:1.25;color:rgb(1=
20,120,122)">To</td><td valign=3D"top" style=3D"font-size:12px;line-height:=
1.25;color:rgb(35,35,36);word-break:break-all"><a style=3D"color:rgb(48,97,=
242);text-decoration:none" href=3D"mailto:robertpang@google.com" target=3D"=
_blank">Robert Pang&lt;robertpang@google.com&gt;</a></td></tr><tr><td valig=
n=3D"top" style=3D"width:4.25em;font-size:12px;line-height:1.25;color:rgb(1=
20,120,122)">Cc</td><td valign=3D"top" style=3D"font-size:12px;line-height:=
1.25;color:rgb(35,35,36);word-break:break-all"><a style=3D"color:rgb(48,97,=
242);text-decoration:none" href=3D"mailto:pgsql-hackers@postgresql.org" tar=
get=3D"_blank">pgsql-hackers&lt;pgsql-hackers@postgresql.org&gt;</a></td></=
tr><tr><td valign=3D"top" style=3D"width:4.25em;font-size:12px;line-height:=
1.25;color:rgb(120,120,122)">Subject</td><td valign=3D"top" style=3D"font-s=
ize:12px;line-height:1.25;color:rgb(35,35,36);word-break:break-all">Re: [PA=
TCH] Fix premature timeout in pg_promote() caused by signal interruptions</=
td></tr></tbody></table></div><pre>On Wed, Mar 11, 2026 at 09:44:07AM -0700=
, Robert Pang wrote:<br><blockquote>The current implementation of pg_promot=
e() calculates a fixed number<br> of loop iterations based on the timeout v=
alue, assuming each loop<br> waits exactly 100 ms for the backend latch. Ho=
wever, if the backend<br> receives an unrelated signal (e.g., from<br> clie=
nt_connection_check_interval), it wakes up early. These repeated,<br> unrel=
ated wakeups cause the loop counter to deplete much faster than<br> intende=
d, leading to a premature timeout.<br></blockquote><br>It is true that we c=
an do better here, and your proposal about having<br>a more precise timeout=
 calculation looks like a sensible improvement<br>for this case.<br><br>No =
objections regarding your patch.=C2=A0 I would like to apply it on HEAD,<br=
>if there are no objections.<br>--<br>Michael<br></pre><pre><br></pre><pre>=
<pre>Hi all,</pre><pre>Overall LGTM. Just a small comment:</pre><pre>&quot;=
+  end_time =3D GetCurrentTimestamp() + wait_seconds * 1000000L;&quot;</pre=
><pre>I think we can use TimestampTzPlusSeconds(GetCurrentTimestamp(), wait=
_seconds).</pre><pre><br></pre><pre>Best regards</pre><pre>Tiancheng Ge<br>=
</pre></pre></div></div></blockquote><div><br></div><div>Thank you for revi=
ewing this patch. The use of=C2=A0TimestampTzPlusSeconds() will be good.</d=
iv><div><br></div><div>Best regards</div><div>Robert Pang</div></div></div>

--000000000000ccd3a4064de57f2a--