> BTW, it appears to me that doing it this way is O(N^2) in the number
> of active temp tables. So it's not hard to believe that the patch
> as-presented would actually be a fairly serious performance drag for
> some use cases with lots of temp tables. There are certainly ways
> we could do better than that (hash table, bloom filter, etc) but
> there would be even more engineering effort needed.
Yes, you're right. I also consider using like hash table to do more better and try
to merge the in_use list and on_commits list into one hashtable. But, as just you
said, it needs much more effort. Thanks any way.
Regards,
Jet
Halo Tech (www.halodbtech.com)
openHalo (www.openhalo.org)