kill other backends when timed out about pg_squeeze

[ahouska]

Is the problem that no one can access the table on which pg_squeeze is trying to acquire an exclusive lock (and that lock cannot be acquired due to the idle transactions)? If this is the case, then it's not specific to pg_squeeze: any DDL command will probably cause the same situation. Please describe your problem more in detail if I misunderstood.

[Dsolik]

Squeeze is waiting for the completion of transactions in the status of idle in transaction, as a result of which a huge blocking tree can be formed, I would like squeeze to be able to shoot such transactions, for example, as it does pg_repack -D

[Dsolik]

Now in crontab we run: select pg_terminate_backend (pid),(now() - pg_stat_activity.xact_start) as duration,query from pg_stat_activity where pid=(select pid from pg_locks where relation=(select relation from pg_locks where pid=(select pid from pg_stat_activity where backend_type ='squeeze worker') and granted='f') and granted ='t') and (now() - pg_stat_activity.xact_start)> interval '1 minutes'

[ahouska]

SELECT squeeze.stop_worker();

[ahouska]

Squeeze is waiting for the completion of transactions in the status of idle in transaction, as a result of which a huge blocking tree can be formed, I would like squeeze to be able to shoot such transactions, for example, as it does pg_repack -D

You still haven't explained in detail what happens. The contents of pg_locks system table at the time the problem exists might give me a clue.

[Dsolik]

pg_locks.log
pg_stat_activity_.log

[ahouska]

I don't see any call of the squeeze.squeeze_table() function in pg_stat_activity. Only this function does the actual work, including locking and waiting for locks. I was curious about the "lock tree" that you mentioned earlier.