On Thu, Sep 23, 2004 at 08:29:25AM -0700, Mr Pink wrote:
> Not knowing anything about the internals of pg, I don't know how this relates, but in theory,
> query plan caching is not just about saving time re-planning queries, it's about scalability.
> Optimizing queries requires shared locks on the database metadata, which, as I understand it
> causes contention and serialization, which kills scalability.
One of the guru's can correct me if I'm wrong here, but AFAIK metadata
lookups use essentially the same access methods as normal queries. This
means MVCC is used and no locking is required. Even if locks were
required, they would be shared read locks which wouldn't block each
Jim C. Nasby, Database Consultant decibel(at)decibel(dot)org
Give your computer some brain candy! www.distributed.net Team #1828
Windows: "Where do you want to go today?"
Linux: "Where do you want to go tomorrow?"
FreeBSD: "Are you guys coming, or what?"
In response to
pgsql-performance by date
|Next:||From: Tom Lane||Date: 2004-09-27 19:26:32|
|Subject: Re: best statistic target for boolean columns |
|Previous:||From: Gregory Stark||Date: 2004-09-27 19:13:45|
|Subject: Re: best statistic target for boolean columns|