Re: Proposal for CSN based snapshots

Hi,

On Fri, Jun 13, 2014 at 7:24 PM, Heikki Linnakangas
<hlinnakangas@vmware.com> wrote:
> Yeah, that seems like a better design, after all.
>
> Attached is a new patch. It now keeps the current pg_clog unchanged, but
> adds a new pg_csnlog besides it. pg_csnlog is more similar to pg_subtrans
> than pg_clog: it's not WAL-logged, is reset at startup, and segments older
> than GlobalXmin can be truncated.
>
> This addresses the disk space consumption, and simplifies pg_upgrade.
>
> There are no other significant changes in this new version, so it's still
> very much WIP. But please take a look!
>

Thanks for working on this important patch. I know this patch is still
largely a WIP but I would like to report an observation.

I applied this patch and did a few pgbench runs with 32 clients (this
is on a not so powerful VM, by the way) . Perhaps you suspect such a
thing already but I observed a relatively larger percentage of time
being spent in XLogInsert().

Perhaps XLogCtlInsert.insertpos_lck contention via GetXLogInsertRecPtr()?

--
Amit