Re: recoveryCheck/008_fsm_truncation is failing on dodo in v14- (due to slow fsync?)

From: Robins Tharakan <tharakan(at)gmail(dot)com>
To: Alexander Lakhin <exclusion(at)gmail(dot)com>
Cc: pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: recoveryCheck/008_fsm_truncation is failing on dodo in v14- (due to slow fsync?)
Date: 2024-06-22 11:52:51
Message-ID: CAEP4nAxQizWfM+TpK_0WBBZD2CDtutFeKg-_0+Jpa080-=gDBw@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Sat, 22 Jun 2024 at 18:30, Alexander Lakhin <exclusion(at)gmail(dot)com> wrote:

> So it doesn't seem impossible for this operation to last for more than two
> minutes.
>
> The facts that SyncDataDirectory() is executed between these two messages
> logged, 008_fsm_truncation is the only test which turns fsync on, and we
> see no such failures in newer branches (because of a7f417107), make me
> suspect that dodo is slow on fsync.
>

Not sure if it helps but I can confirm that dodo is used for multiple tasks
and that
it is using a (slow) external USB3 disk. Also, while using dodo last week
(for
something unrelated), I noticed iotop at ~30MB/s usage & 1-min CPU around
~7.

Right now (while dodo's idle), via dd I see ~30MB/s is pretty much the max:

pi(at)pi4:/media/pi/250gb $ dd if=/dev/zero of=./test count=1024 oflag=direct
bs=128k
1024+0 records in
1024+0 records out
134217728 bytes (134 MB, 128 MiB) copied, 4.51225 s, 29.7 MB/s

pi(at)pi4:/media/pi/250gb $ dd if=/dev/zero of=./test count=1024 oflag=dsync
bs=128k
1024+0 records in
1024+0 records out
134217728 bytes (134 MB, 128 MiB) copied, 24.4916 s, 5.5 MB/s

-
robins

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Bertrand Drouvot 2024-06-22 12:48:33 Re: Track the amount of time waiting due to cost_delay
Previous Message Amit Kapila 2024-06-22 09:47:03 Re: New standby_slot_names GUC in PG 17