On 12 November 2014 07:54, David Rowley <dgrowleyml@gmail.com> wrote: > On Tue, Nov 11, 2014 at 9:29 PM, Simon Riggs <simon@2ndquadrant.com> wrote: >> >> >> This plan type is widely used in reporting queries, so will hit the >> mainline of BI applications and many Mat View creations. >> This will allow SELECT count(*) FROM foo to go faster also. >> > > We'd also need to add some infrastructure to merge aggregate states together > for this to work properly. This means that could also work for avg() and > stddev etc. For max() and min() the merge functions would likely just be the > same as the transition functions.
Do you mean something like a "subtotal" or "intermediate combination" functon?
If you had 4 parallel workers performing a seqscan, say the relation was 4000 pages in total, you could say that worker 1 would scan blocks 0-999, worker 2, 1000-1999 etc. After each worker had finished, there would then be 4 sets of records then needed to be merged into 1 set.
Take int4_avg_accum() for example it does:
transdata->count++;
transdata->sum += newval;
The merge function would need to perform something like:
transdata->count += transdata2merge.count;
transdata->sum += transdata2merge.sum;
Then the final function could be called on the merged aggregate state.
The same can be applied when the query contains a GROUP BY clause, just we'd need pay attention to which groups we merge together for that to work Any HAVING clause would have to be applied after the groups have been merged.
This whole topic is pretty exciting for data warehouse type workloads.
