Hi, On Mon, 31 Mar 2025 at 21:15, Melanie Plageman <melanieplage...@gmail.com> wrote: > > On Mon, Mar 31, 2025 at 12:27 PM Nazir Bilal Yavuz <byavu...@gmail.com> wrote: > > > > I worked on an alternative approach, I refactored code a bit. It does > > not traverse the list two times and I think the code is more suitable > > to use read streams now. I simply get how many blocks are processed by > > read streams and move the list forward by this number, so the actual > > loop skips these blocks. This approach is attached with 'alternative' > > prefix. > > I am leaning toward the refactored approach because I don't think we > want to go through the array twice and I think it is hard to get it > right with incrementing p.pos in both places and being sure we > correctly close the relation etc.
I liked it more as well.
> Looking at your alternative approach, I don't see how the innermost
> while loop in autoprewarm_database_main() is correct
>
> /* Check whether blocknum is valid and within fork file size. */
> while (cur_filenumber == blk->filenumber &&
> blk->blocknum >= nblocks_in_fork)
> {
> /* Move to next forknum. */
> pos++;
> continue;
> }
>
> Won't this just infinitely loop?
Oops, you are right. It should be an if statement instead of a while
loop, fixed now. Also, moved 'dsm_detach(seg);' to its previous place
to reduce diff. Attached as v7. Do you think that I should continue to
attach both approaches?
--
Regards,
Nazir Bilal Yavuz
Microsoft
v7-0001-Optimize-autoprewarm-with-read-streams.patch
Description: Binary data
v7-0002-Count-free-buffers-at-the-start-of-the-autoprewar.patch
Description: Binary data
