ESPs are created serially, should be created in parallel
Bug #1336271 reported by
Mike Hanlon
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Trafodion |
Fix Released
|
High
|
Mike Hanlon |
Bug Description
Parallel creation of ESPs was disabled in October '13 to fix a problem with some regression tests failing. It would be good to figure out the specific problem(s), if any, with parallel creation and fix those, so that we don't have to take the performance hit of serial creation. It costs about 15 seconds to fixup a 64 d-o-p plan when ESPs must be started serially, versus ~0.5 seconds if they are started in parallel.
Changed in trafodion: | |
assignee: | nobody → Mike Hanlon (mike-hanlon) |
description: | updated |
Changed in trafodion: | |
status: | New → In Progress |
tags: | added: sql-exe |
Changed in trafodion: | |
importance: | Undecided → High |
Changed in trafodion: | |
milestone: | none → r0.8 |
Changed in trafodion: | |
status: | Fix Committed → Fix Released |
To post a comment you must log in.
I re-enabled parallel ESP creation, and verified the expected behavior, i.e., that fixup time decreases noticeably when ESPs need to be created. Peter tested the change with ED and found that it had a positive impact. I ran all SQL regression tests on a workstation. The only persistent failure was udr/TEST101, but it also failed in the same way when I temporarily backed out my change. There were two non-repeatable failures as well: debug flavor usr/TEST001 and release flavor seabase/TEST012.
I don't think enabling parallel ESP creation causes any regression test failure so I will check-in this fix.