On 10/26/2012 6:35 AM, Amit Kapila wrote:
> On Friday, October 26, 2012 11:50 AM Jan Wieck wrote:
>> On 10/26/2012 1:29 AM, Amit Kapila wrote:
>> > One other way could be to check after every few pages for a
>> conflicting
>> > lock request.
>>
>> How is this any different from what my patch does?
> The difference is that in the patch it checks for waiters by using 2
> parameters autovacuum_truncate_lock_check and blkno%32 and what I
> had mentioned was to check only based on blkno.
> Will it effect too much if we directly check for waiters after every 32
> (any feasible number) blocks?
The blkno%32 is there to not do the gettimeofday() call too often. But
relying on the blkno alone is IMHO not a good idea. It had to be a
number small enough so that even on a busy system and when the pages
have to be read from disk, vacuum checks and releases the lock quickly.
But large enough so that it doesn't create a significant amount of
spinlock calls in the lmgr. We would end up with another parameter,
number of blocks, that is a lot harder to estimate a good value for.
Jan
--
Anyone who trades liberty for security deserves neither
liberty nor security. -- Benjamin Franklin
