On 6/27/20 1:37 AM, Matthias Apitz wrote:
El día Donnerstag, Juni 25, 2020 a las 04:25:00 -0400, Tom Lane escribió:
Matthias Apitz <guru@xxxxxxxxxxx> writes:
On Thursday, 25 June 2020 21:41:54 CEST, Tom Lane <tgl@xxxxxxxxxxxxx>
wrote:
regardless of the exact details, it seems like the most likely theory
about what is happening is that the dump file is corrupt and the
corruption is causing the de-gzipped output to be missing or
duplicating chunks of text.
I think, that a
gzip -dc foo
will work, regardless of the file names extensiom not beeing .gz
Agreed, that gzip command won't care. I was concerned about what
might've happened in between.
...
I restarted the server, created a new database with:
$ /usr/local/sisis-pap/pgsql/bin/createdb -U sisis -T template0 testdump
and loaded the DUMP while no other processes were connected to the
server with:
$ gzip -dc newanna_export.dmp | /usr/local/sisis-pap/pgsql/bin/psql -U sisis -dtestdump
all went fine (table creations, COPYs into them, creating indexs)
without any error message in the terminal. In the serverlog only these
messages came out:
2020-06-27 10:16:38.778 CEST [18094] LOG: database system is ready to accept connections
2020-06-27 10:21:26.062 CEST [18096] LOG: checkpoints are occurring too frequently (25 seconds apart)
2020-06-27 10:21:26.062 CEST [18096] HINT: Consider increasing the configuration parameter "max_wal_size".
2020-06-27 10:21:52.436 CEST [18096] LOG: checkpoints are occurring too frequently (26 seconds apart)
2020-06-27 10:21:52.436 CEST [18096] HINT: Consider increasing the configuration parameter "max_wal_size".
2020-06-27 10:22:07.424 CEST [18302] ERROR: canceling autovacuum task
2020-06-27 10:22:07.424 CEST [18302] CONTEXT: automatic analyze of table "testdump.public.d01buch"
2020-06-27 10:22:14.545 CEST [18302] ERROR: canceling autovacuum task
2020-06-27 10:22:14.545 CEST [18302] CONTEXT: automatic analyze of table "testdump.public.d02zus"
2020-06-27 10:22:16.051 CEST [18249] ERROR: canceling autovacuum task
2020-06-27 10:22:16.051 CEST [18249] CONTEXT: automatic analyze of table "testdump.public.d03geb"
2020-06-27 10:22:18.331 CEST [18249] ERROR: canceling autovacuum task
2020-06-27 10:22:18.331 CEST [18249] CONTEXT: automatic analyze of table "testdump.public.d11loesch"
2020-06-27 10:22:19.718 CEST [18302] ERROR: canceling autovacuum task
2020-06-27 10:22:19.718 CEST [18302] CONTEXT: automatic analyze of table "testdump.public.d11rueck"
2020-06-27 10:22:23.034 CEST [18249] ERROR: canceling autovacuum task
2020-06-27 10:22:23.034 CEST [18249] CONTEXT: automatic analyze of table "testdump.public.d35buchnotiz"
2020-06-27 10:22:38.902 CEST [18302] ERROR: canceling autovacuum task
2020-06-27 10:22:38.902 CEST [18302] CONTEXT: automatic analyze of table "testdump.public.idm_statistic"
2020-06-27 10:22:59.332 CEST [18302] ERROR: canceling autovacuum task
2020-06-27 10:22:59.332 CEST [18302] CONTEXT: automatic analyze of table "testdump.public.idm_targetlog"
2020-06-27 10:23:01.493 CEST [18302] ERROR: canceling autovacuum task
2020-06-27 10:23:01.493 CEST [18302] CONTEXT: automatic analyze of table "testdump.public.idm_targettab"
2020-06-27 10:23:02.631 CEST [18302] ERROR: canceling autovacuum task
2020-06-27 10:23:02.631 CEST [18302] CONTEXT: automatic analyze of table "testdump.public.idm_tasktab"
The messages about 'ERROR: canceling autovacuum task' appeared while
the indexes were created. The database has around 400 tables and the
uncompressed dump is around 3.6 GByte:
$ gzip -dc newanna_export.dmp | wc -c
3643850597
So, the syntax error messages were caused by some other process (and we
have already an idea which cron job it was). What must be done re/ the
messages about 'ERROR: canceling autovacuum task', or can we just
ignore them?
From what I understand they are occurring because the machine is to
busy doing the restore to get to the autovacuum task in a timely manner.
So I would say ignore and check back later to see that the autovacuum is
working. Given that it is ANALYZE that is being cancelled I would run a
manual ANALYZE after the restore is done to update the database statistics.
Thanks
matthias
--
Adrian Klaver
adrian.klaver@xxxxxxxxxxx
