Glance deployed with Single worker for API and Registry
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
puppet-glance |
Invalid
|
Undecided
|
Unassigned |
Bug Description
In the glance configuration files, "workers" is set to 0. Every other service I know of, treats workers=0 as wokers= number of cores on the machine. But seems like in glance's case setting the workers to None does workers=# of cores. So either, we should be setting workers=None when deploying(have huge performance gains) or change glance so that it treats workers=0 as workers= number of cores.
Some rally tests on a 32 core machine with api and registry workers set to 0(1 worker) vs 32.
Glance create and delete image at 64 concurrency.
1 worker:
+------
| Response Times (sec) |
+------
| Action | Min (sec) | Median (sec) | 90%ile (sec) | 95%ile (sec) | Max (sec) | Avg (sec) | Success | Count |
+------
| glance.create_image | 8.612 | 14.652 | 18.135 | 19.215 | 23.974 | 14.847 | 100.0% | 1000 |
| glance.delete_image | 0.646 | 8.095 | 10.229 | 11.135 | 17.047 | 8.224 | 100.0% | 1000 |
| total | 10.251 | 22.996 | 26.547 | 27.984 | 39.831 | 23.072 | 100.0% | 1000 |
+------
32 workers:
+------
| Response Times (sec) |
+------
| Action | Min (sec) | Median (sec) | 90%ile (sec) | 95%ile (sec) | Max (sec) | Avg (sec) | Success | Count |
+------
| glance.create_image | 4.211 | 9.405 | 10.76 | 11.089 | 15.997 | 9.318 | 100.0% | 1000 |
| glance.delete_image | 0.544 | 4.741 | 5.807 | 6.037 | 7.047 | 4.631 | 100.0% | 1000 |
| total | 6.758 | 14.038 | 15.651 | 16.157 | 21.642 | 13.949 | 100.0% | 1000 |
+------
Changed in puppet-glance: | |
importance: | Undecided → Medium |
Formatting is borke in the Rally results, here is a link to much bettter looking formatting: http:// paste.openstack .org/show/ 545640/