Re: [PATCH] Reduce EXPLAIN ANALYZE overhead for row counting - Mailing list pgsql-hackers

From Ryo Matsumura (Fujitsu)
Subject Re: [PATCH] Reduce EXPLAIN ANALYZE overhead for row counting
Date
Msg-id TYRPR01MB13457E2998F52D94C51422AFDE847A@TYRPR01MB13457.jpnprd01.prod.outlook.com
Whole thread Raw
In response to [PATCH] Reduce EXPLAIN ANALYZE overhead for row counting  (Hironobu SUZUKI <hironobu@interdb.jp>)
List pgsql-hackers
Hi Suzuki-san, hackers,

I am supportive of efforts to improve EXPLAIN ANALYZE.
At a minimum, I would like to see the Phase 1 patch merged.
I have reviewed only the Phase 1 patch, and I did not notice any problems with it.
It's very simple.

In practice, when we encounter performance issues in customer systems, it is often not feasible to ask them to run EXPLAIN ANALYZE on the live production system with the real application workload. The overhead can be high enough to risk destabilizing the system.
While fully resolving a performance issue may ultimately require enabling BUFFERS, TIMING, and WAL, the small overhead demonstrated here makes focusing only on row counting a very reasonable first step. In some cases, if the issue is primarily related to the execution plan itself, this information alone may be sufficient to identify the problem.
From that perspective, I believe customers would be much more willing to accept running EXPLAIN ANALYZE with this reduced overhead.

Thanks for working on this.
Just my perspective based on support experience with customer systems.

Best Regards
Ryo Matsumura

pgsql-hackers by date:

Previous
From: shveta malik
Date:
Subject: Re: synchronized_standby_slots behavior inconsistent with quorum-based synchronous replication
Next
From: Lukas Fittl
Date:
Subject: Re: Reduce timing overhead of EXPLAIN ANALYZE using rdtsc?