On Wed, 2005-12-07 at 16:59 -0500, Tom Lane wrote:
> I've now seen actual evidence of that in
> profiling pgbench: using a modified backend that counts LWLock-related
> wait operations,
> So it seems it's time to start thinking about how to reduce contention
> for the LockMgrLock
You're right to be following up this thought.
My concern, longer term is on our ability to determine contention issues
in an agreed way. I've long been thinking about wait-time measurement -
I think its the only way to proceed.
There's always a next-bottleneck, so I'd like to first agree the
diagnostic probes so we can decide how to determine that. That way we
can all work on solutions for various workloads, and prove that they
work, in those cases.
My view would be that the LockMgrLock is not relevant for all workloads,
but I want even more to be able to discuss whether it is, or is not, on
an accepted basis before discussions begin.
Best Regards, Simon Riggs