Message-ID: <4E01B0B8.2020205@enterprisedb.com>
Date: Wed, 22 Jun 2011 12:07:04 +0300
From: Heikki Linnakangas <heikki.linnakangas@enterprisedb.com>
User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; fi-FI;
	rv:1.9.2.17) Gecko/20110606 Icedove/3.1.10
MIME-Version: 1.0
To: Dan Ports <drkp@csail.mit.edu>
CC: pgsql-hackers@postgresql.org, Kevin Grittner <Kevin.Grittner@wicourts.gov>
Subject: Re: Repeated PredicateLockRelation calls during seqscan
References: <20110622045850.GN83336@csail.mit.edu>
In-Reply-To: <20110622045850.GN83336@csail.mit.edu>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

On 22.06.2011 07:58, Dan Ports wrote:
> I was looking at ExecSeqScan today and noticed that it invokes
> PredicateLockRelation each time it's called, i.e. for each tuple
> returned. Any reason we shouldn't skip that call if
> rs_relpredicatelocked is already set, as in the attached patch?
>
> That would save us a bit of overhead, since checking that flag is
> cheaper than doing a hash lookup in the local predicate lock table
> before bailing out.

Hmm, I wonder if we should move this logic to heapam.c. The optimization 
to acquire a relation lock straight away should apply to all heap scans, 
not only those coming from ExecSeqScan. The distinction is academic at 
the moment, because that's the only caller that uses a regular MVCC 
snapshot, but it seems like a modularity violation for nodeSeqscan.c to 
reach into the HeapScanDesc to set the flag and grab the whole-relation 
lock, while heapam.c contains the PredicateLockTuple and 
CheckForSerializableConflictOut() calls.

BTW, isn't bitgetpage() in nodeBitmapHeapscan.c missing 
PredicateLockTuple() and CheckForSerializableConflictOut() calls in the 
codepath for a lossy bitmap? In the non-lossy case, 
heap_hot_search_buffer() takes care of it, but not in the lossy case.

-- 
   Heikki Linnakangas
   EnterpriseDB   http://www.enterprisedb.com