[Swan-dev] test suite slower

Andrew Cagney andrew.cagney at gmail.com
Wed Aug 3 15:09:34 UTC 2016


Perhaps it's related to the way the file systems are now created (see
earlier e-mail about clone).  The disk images were set up with backing
store:

  swanfedorabase <- east, west ....

but now there is an extra level vis:

  swanfedorabase <- clone <- east, west, ....

For instance:

$ qemu-img info ../pool/east.qcow2
...
disk size: 343M
cluster_size: 65536
backing file: /home/libreswan/pool/clone.qcow2
$ qemu-img info ../pool/clone.qcow2
...
disk size: 19M
cluster_size: 65536
backing file: /home/libreswan/pool/swanfedorabase.qcow2

The boot time presumably roughly correlates to the number of backing stores.

I've pushed some changes to more clearly identify each test's
boot-time and script time (look for stop booting domains after ...).

Can you try the following:

- find a test with lots of domains and time it (setting KVM_WORKERS=1
so that the boots are serialized might help magnify the problem).

- apply the attached patch (it creates clone's image stand-alone
getting us back to the old set up), and then re-create the clones
(kvm-uninstall-clones kvm-install); ,and then re-run the test

if it shows promise then the following are likely also worth exploring

- adding '-c' to qemu-img convert so that the qcow2 image is compressed
- try the same change to the east, west, ..., disks, this would only
be practical with compression

Andrew

On 2 August 2016 at 20:38, Andrew Cagney <andrew.cagney at gmail.com> wrote:
> On 2 August 2016 at 17:42, D. Hugh Redelmeier <hugh at mimosa.com> wrote:
>> The test suite used to take about 12 hours to run on my machine.
>>
>> After Andrew added a couple of kinds of parallelism, it went down to about
>> 5.75 hours, a very useful improvement.
>>
>> In the last little while it has gone up to 8.5 to 9 hours.  The first slow
>> run was on 2016-07-24.
>>
>> What happened?  Can this be fixed?
>
> One data point,  http://testing.libreswan.org/results/ consistently
> takes <6:00 and it was last rebuilt around July 21st.  it's invocation
> is:
>
> /home/build/libreswan-cagney/testing/utils/kvmrunner.py  --prefix t1.
> --prefix t2.  --prefix t3.  --prefix t4. --workers 2  --test-result
> "good" --skip-passed --test-result "good|wip"  testing/pluto
> kvmrunner 0.00: Options:
> kvmrunner 0.00:   directories: ['testing/pluto']
> kvmrunner 0.00:   verbose: 0
> kvmrunner 0.00:   dry-run: False
> kvmrunner 0.00: Testsuite arguments:
> kvmrunner 0.00:   testing-directory: 'testing'
> kvmrunner 0.00:   testsuite-output: 'None'
> kvmrunner 0.00: Test Runner arguments:
> kvmrunner 0.00:   workers: 2
> kvmrunner 0.00:   prefix: ['t1.', 't2.', 't3.', 't4.']
> kvmrunner 0.00:   parallel: False
> kvmrunner 0.00:   stop-at: None
> kvmrunner 0.00:   backup-directory: BACKUP/2016-08-03-001310
> kvmrunner 0.00: Postmortem arguments
> kvmrunner 0.00:   ignore-all-spaces: False
> kvmrunner 0.00:   ignore-blank-lines: False
> kvmrunner 0.00: Logging arguments:
> kvmrunner 0.00:   log-level: 'info'
> kvmrunner 0.00:   debug: 'None'
> kvmrunner 0.00: Skip arguments:
> kvmrunner 0.00:   skip-passed: True
> kvmrunner 0.00:   skip-failed: False
> kvmrunner 0.00:   skip-incomplete: False
> kvmrunner 0.00:   skip-untested: False
> kvmrunner 0.00: Test filter arguments:
> kvmrunner 0.00:   test-kind: 'kvmplutotest'
> kvmrunner 0.00:   test-name: ''
> kvmrunner 0.00:   test-result: 'good|wip'
> kvmrunner 0.00:   test-exclude: ''
> kvmrunner 0.00:   include-ignored: 'False'
>
>
> Andrew
-------------- next part --------------
A non-text attachment was scrubbed...
Name: diffs
Type: application/octet-stream
Size: 644 bytes
Desc: not available
URL: <https://lists.libreswan.org/pipermail/swan-dev/attachments/20160803/abcd427a/attachment.obj>


More information about the Swan-dev mailing list