LWLock optimization

Hi.

I was asked to bring up this topic on maling lists after asking question on IRC.

The issue is the following.
On one of our servers (10.5), I see a lot of queries with wait_event_type='LWLock', wait_event='lock_manager'

This is a stat gathering/analyzing application with tables which have > 100 partitions.
The queries itself are mostly simple (select by pkey, update by primary key and so on).

select count(*) from pg_locks shows about 40K records (all granted) and

select count(*) from (select distinct locktype,database,relation,page,tuple,virtualxid,transactionid,classid,objid,objsubid from pg_locks)

is varying from about 1K to 25K (I haven't stored this value in monitoring system, just observed manually).

Currently I was adviced to increase LOG2_NUM_LOCK_PARTITIONS at compile time. If it bothers us enough, we would likely do it, but I have a question, if this (or NUM_LOCK_PARTITIONS) value should be increased by default or set tunable?

С уважением,
Александр Пыхалов,
программист отдела телекоммуникационной инфраструктуры
управления информационно-коммуникационной инфраструктуры ЮФУ