Hi again,
thank you for pinpointing the issue.
I have now updated the table with "ALTER TABLE $table SET LOGGED" (actually
it's a partitioned table and I've altered both the main table and the
partitions).
I wanted to double-check the result and what I found out using "select
relpersistence, relname from pg_class" that, e.g., ${table_name}_383_pkey still
has its relpersistence set to "u" whereas ${table_name}_383 has its
relpersistence set to "p" now.
Does anyone have an idea what I'm doing wrong here?
Thank you!
Manuel
________________________________
From: Holtgrewe, Manuel <[email protected]>
Sent: Tuesday, June 15, 2021 5:53:54 PM
To: Tom Lane
Cc: [email protected]
Subject: Re: [ext] Re: Losing data because of problematic configuration?
>> < 2021-06-15 12:33:04.537 CEST > DEBUG: resetting unlogged relations:
>> cleanup 1 init 0
>
> Are you perhaps keeping your data in an UNLOGGED table? If so, resetting
> it to empty after a crash is exactly what's supposed to happen. The
> entire point of UNLOGGED is that the performance benefits come at the
> cost of losing the data on crash.
D'oh!
Yes, that is the case. I once used this but I was convinced that I took this
back in some version. It is not in my main deployment, though. Now I have to
find out (a) why I have diverging deployment and (b) how that bug came about to
be.
Thanks a lot!
________________________________
From: Tom Lane <[email protected]>
Sent: Tuesday, June 15, 2021 3:39:31 PM
To: Holtgrewe, Manuel
Cc: [email protected]
Subject: [ext] Re: Losing data because of problematic configuration?
"Holtgrewe, Manuel" <[email protected]> writes:
> So it looks as if the database jumps back "half an hour" to ensure consistent
> data. Everything in between is lost.
Postgres does not lose committed data --- if it did, we'd consider that a
fairly serious bug. (Well, there are caveats of course. But most of them
have to do with operating-system crashes or power loss, neither of which
are at stake here.)
I am wondering about this though:
> < 2021-06-15 12:33:04.537 CEST > DEBUG: resetting unlogged relations:
> cleanup 1 init 0
Are you perhaps keeping your data in an UNLOGGED table? If so, resetting
it to empty after a crash is exactly what's supposed to happen. The
entire point of UNLOGGED is that the performance benefits come at the
cost of losing the data on crash.
regards, tom lane
