This output was captured after stopping all writes to our postgres database and the only thing talking to postgres was our webserver that only does metadata reads. As you can see from the above gist, even the 'SET' statements are blocked waiting for the ALTER statement to finish.
when you say 'almost every query in our application starts getting blocked'...
'blocked' as in 'they are stuck and can be seen in pg_stat_activity'? (in this case, please post the full content of pg_stat_activity)
or
'blocked' as in 'they are waiting in pgbouncer pool?
regards,
fabio pardi
On 31/05/18 12:38, Ashu Pachauri wrote: > We have been using Postgres 9.5.12 behind PGBouncer and facing some weird issues. Whenever we running long running DDL statements (e.g. 'add index concurently' or 'Alter table alter column type'), after some time, we start seeing that almost every query in our application starts getting blocked. > I understand that the operations I mentioned cab be unsafe, but the queries being blocked are on completely unrelated tables. I used the instructions given on Postgres wiki (https://wiki.postgresql.org/wiki/Lock_Monitoring) to correlate the blocking and blocked statements and there seems to be absolutely no correlation. > > Thanks, > Ashu
