On MNT-801 we discovered that the search service now implements a feature to "bulk fetch" node properties when searching for them.
For a customer with Oracle, this has caused the search time to increase linearly with the size of the resultset, meaning that searches on 4.1.3 are slower than on 4.1.2
In tests and profiling attached on MNT-801 we observed that most of this time was spent doing DB queries.
The customer is requesting that there is either a way to disable the bulk-fetching, or improve it so that there is no speed degradation fro Explorer users.
To reproduce the problem:
on a big 4.1.2 environment using Oracle and few million nodes with Solr as search subsystem
open Alfresco explorer and search for "doc" or "pdf" that would give you a big resultset.
Repeat the searches with 4.1.3 and compare timings.
Warm caches, times from Firebug
search for DOC 1.18s
Search for PDF 1.22s
Search for TXT 0.83s
search for DOC 3.3s
Search for PDF 2.2s
Search for TXT 1.49s
These timings are for tests on Support's systems. The tests on the customers systems show a similar percentage change, but all times are longer, so the impact is much greater.
4.1.4 shows a small, almost negligible improvement.
Andy Hind proposed a rethink/reimplementation of the bulk fetch: