Thanks tom.
But what I think is we may provide a better experience. Consider the
below example:
[jet@halodev-jet-01 data]$ psqlpsql (16.6)
Type "help" for help.
postgres=# CREATE TABLE a_test (n INT);
CREATE TABLE
postgres=# INSERT INTO a_test VALUES (1);
INSERT 0 1
postgres=# 2024-12-23 16:56:11.023 CST [1356476] FATAL: terminating
connection due to unexpected postmaster exit
postgres=#
postgres=# \q
[jet@halodev-jet-01 data]$
### Here we simulate crash and clog file corrupt (delete the clog file).
[jet@halodev-jet-01 data]$ pg_ctl start
pg_ctl: another server might be running; trying to start server anyway
waiting for server to start....2024-12-23 16:57:24.036 CST [1356495] LOG:
starting PostgreSQL 16.6 on x86_64-pc-linux-gnu, compiled by gcc (GCC)
8.5.0 20210514 (Red Hat 8.5.0-4), 64-bit
2024-12-23 16:57:24.036 CST [1356495] LOG: listening on IPv6 address
"::1", port 5432
2024-12-23 16:57:24.036 CST [1356495] LOG: listening on IPv4 address
"127.0.0.1", port 5432
2024-12-23 16:57:24.046 CST [1356495] LOG: listening on Unix socket
"/tmp/.s.PGSQL.5432"
2024-12-23 16:57:24.055 CST [1356498] LOG: database system was
interrupted; last known up at 2024-12-23 16:54:56 CST
2024-12-23 16:57:24.147 CST [1356498] LOG: database system was not
properly shut down; automatic recovery in progress
2024-12-23 16:57:24.151 CST [1356498] LOG: redo starts at 0/14E4D20
2024-12-23 16:57:24.152 CST [1356498] LOG: file "pg_xact/0000" doesn't
exist, reading as zeroes
2024-12-23 16:57:24.152 CST [1356498] CONTEXT: WAL redo at 0/14FCAB0 for
Transaction/COMMIT: 2024-12-23 16:55:13.531244+08; inval msgs: catcache 80
catcache 79 catcache 80 catcache 79 catcache 55 catcache 54 catcache 7 catcache
6 catcache 7 catcache 6 catcache 7 catcache 6 catcache 7 catcache 6 catcache 7
catcache 6 catcache 7 catcache 6 catcache 7 catcache 6 snapshot 2608 relcache
16384
2024-12-23 16:57:24.152 CST [1356498] LOG: invalid record length at
0/14FCD20: expected at least 24, got 0
2024-12-23 16:57:24.152 CST [1356498] LOG: redo done at 0/14FCCE8 system
usage: CPU: user: 0.00 s, system: 0.00 s, elapsed: 0.00 s
2024-12-23 16:57:24.157 CST [1356496] LOG: checkpoint starting:
end-of-recovery immediate wait
2024-12-23 16:57:24.184 CST [1356496] LOG: checkpoint complete: wrote 27
buffers (0.2%); 0 WAL file(s) added, 0 removed, 0 recycled; write=0.005 s,
sync=0.014 s, total=0.030 s; sync files=22, longest=0.006 s, average=0.001 s;
distance=96 kB, estimate=96 kB; lsn=0/14FCD20, redo lsn=0/14FCD20
2024-12-23 16:57:24.188 CST [1356495] LOG: database system is ready to
accept connections
done
server started
[jet@halodev-jet-01 data]$ psql
psql (16.6)
Type "help" for help.
postgres=# SELECT * FROM a_test;
n
---
1
(1 row)
postgres=# \q
We can see that when database restart, it will try to recover. So I think we
may improve database reliable in some scenarios if just clog
file corrupted.
Regards,
Jet
Tom Lane<t...@sss.pgh.pa.us> 在 2024年12月23日 周一 14:50
写道：
"=?utf-8?B?56ug5pmo5pumQOaYk+aZr+enkeaKgA==?=" <zhangche...@halodbtech.com>
writes:
> And after a while, a system error occurred and&nbsp;unfortunately,
just caused clog file corrupted.&nbsp;&nbsp;
> So we need to restore the database from backup just because of the tiny
clog file corrupted.
I'm not seeing a large difference between this complaint
and whining because Unix doesn't have a way to recover from
"sudo rm -rf /". clog is critical data: if you mess with
it you will destroy your database. It is not the only
critical data in the system, either.
> Is there any chance to improve this?
We're not in the business of building doubly- or triply-redundant
storage. The cost/benefit just isn't attractive for very many people.
If you don't trust your hardware, you can put your storage on RAID,
or replicate the database, etc. If you have a DBA who thinks it's
cool to remove files they don't understand the purpose of, the answer
is to fire that DBA.
regards, tom lane
