Re: Patch: fix lock contention for HASHHDR.mutex

On Fri, Dec 18, 2015 at 8:46 AM, Teodor Sigaev <teodor(at)sigaev(dot)ru> wrote:
>> Oh, that's an interesting idea. I guess the problem is that if the
>> freelist is unshared, then users might get an error that the lock
>> table is full when some other partition still has elements remaining.
>
> Could we split one freelist in hash to NUM_LOCK_PARTITIONS freelists?
> Each partition will have its own freelist and if freelist is empty then
> partition should search an entry in freelists of other partitions. To
> prevent concurrent access it's needed to add one LWLock to hash, each
> partition should lock LWlock in share mode to work with its own freelist and
> exclusive to work with other freelists.
>
> Actually, I'd like to improve all partitioned hashes instead of improve only
> one case.

Yeah. I'm not sure that should be an LWLock rather than a spinlock,
but we can benchmark it both ways.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company