smart server speed regression on older (e.g. Ubuntu 6.06) kernels
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Bazaar |
Fix Released
|
High
|
Andrew Bennetts |
Bug Description
On a machine with an Ubuntu 6.06 or 6.10 machine (i.e. 2.6.15 or
2.6.17 kernel):
| $ ./bzr --no-plugins selftest 'clone_
| No handlers could be found for logger "bzr"
| testing: /home/james/
| /home/james/
|
| running 1 tests...
| ...implementati
|
| -------
| Ran 1 test in 18.006s
|
| OK
| [ 2669] 2008-05-23 11:02:45.387 INFO: tests passed
| tests passed
| [ 2669] 2008-05-23 11:02:45.387 WARNING: bzrlib.
| bzrlib.
| $
On the same machine, running Ubuntu 6.06 but with an Ubuntu 8.04
(2.6.24) kernel:
| $ ./bzr --no-plugins selftest 'clone_
| No handlers could be found for logger "bzr"
| testing: /home/james/
| /home/james/
|
| running 1 tests...
| ...implementati
|
| -------
| Ran 1 test in 0.440s
|
| OK
| [22234] 2008-05-23 11:03:31.060 INFO: tests passed
| tests passed
| [22234] 2008-05-23 11:03:31.061 WARNING: bzrlib.
| bzrlib.
| $
Also, if I drop back 100 revisions, the time for the runs becomes, on
a 2.6.17 kernel:
| $ bzr revert -r -100
[...]
| $ ./bzr --no-plugins selftest 'clone_
| No handlers could be found for logger "bzr"
| testing: /home/james/
| /home/james/
|
| running 1 tests...
| ...implementati
|
| -------
| Ran 1 test in 5.214s
|
| OK
| [ 3195] 2008-05-23 11:48:49.714 INFO: tests passed
| tests passed
| $
And on a 2.6.24 kernel:
| $ ./bzr --no-plugins selftest 'clone_
| No handlers could be found for logger "bzr"
| testing: /home/james/
| /home/james/
|
| running 1 tests...
| ...implementati
|
| -------
| Ran 1 test in 0.316s
|
| OK
| [22972] 2008-05-23 11:50:34.750 INFO: tests passed
| tests passed
| $
So this appears to be a regression in bzr.
Any chance you could run the comparison with '--lsprof' to see a breakdown of where the time is spent?
I wonder if it is actually our changes which track leaking threads.
Certainly there is somewhat of a kernel issue here, since even the old one is 5000ms => 314ms. Even if there is a regression of 5000 => 18000ms from bzr.