mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From haosdent huang <haosd...@gmail.com>
Subject Re: Review Request 43558: Speed up ExamplesTest.PersistentVolumeFramework.
Date Fri, 12 Aug 2016 13:46:42 GMT


> On Aug. 10, 2016, 7:23 p.m., Benjamin Mahler wrote:
> > src/examples/persistent_volume_framework.cpp, lines 402-410
> > <https://reviews.apache.org/r/43558/diff/3/?file=1470240#file1470240line402>
> >
> >     Can you give some context on why lowering these speeds up the test? How much
does it speed it up?
> 
> haosdent huang wrote:
>     Hi, @bmahler The most effective snippet is
>     
>     ```
>     export MESOS_AUTHENTICATION_TIMEOUT=200ms
>     ```
>     
>     which we committed before. I test this patch again, seems only bring little speed
up.
>     
>     Before apply the patch,
>     
>     ```
>     [       OK ] ExamplesTest.PersistentVolumeFramework (3505 ms)
>     [----------] 1 test from ExamplesTest (3510 ms total)
>     ```
>     
>     After apply the patch,
>     
>     ```
>     [       OK ] ExamplesTest.PersistentVolumeFramework (2493 ms)
>     [----------] 1 test from ExamplesTest (2497 ms total)
>     ```
>     
>     Do you think we should discard this?
> 
> Anindya Sinha wrote:
>     I think the improvement is not due to the reduction in number of shards (since the
ACCEPT call consolidates offer operations for all shards), but due to the reduction in number
of tasks per shard (since each of the tasks are launched in separate ACCEPT calls).
> 
> Benjamin Mahler wrote:
>     Why not have defaults of 1 shard and 1 task per shard? Seems the default should either
be: 1 or run forever. Also, we should probably explicitly set these in the persistent_volume_framework_test.sh.
> 
> Anindya Sinha wrote:
>     Default of 1 shard is fine, but the way is structured I think default for tasks per
shard should be 2. I think the 1st task in a shard does a "touch volume/persisted", and subsequent
tasks does a "test -f volume/persisted". So with 2 tasks per shard, we can test content of
persistent volume (which have a life cycle across task boundaries).

Seems nothing different after change to 1 shard.

```
[       OK ] ExamplesTest.PersistentVolumeFramework (2495 ms)
[----------] 1 test from ExamplesTest (2500 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test case ran. (2529 ms total)
[  PASSED  ] 1 test.
```


- haosdent


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/43558/#review145390
-----------------------------------------------------------


On Aug. 10, 2016, 5:21 p.m., haosdent huang wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/43558/
> -----------------------------------------------------------
> 
> (Updated Aug. 10, 2016, 5:21 p.m.)
> 
> 
> Review request for mesos, Alexander Rukletsov, Benjamin Mahler, and Jie Yu.
> 
> 
> Bugs: MESOS-4663
>     https://issues.apache.org/jira/browse/MESOS-4663
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Speed up ExamplesTest.PersistentVolumeFramework.
> 
> 
> Diffs
> -----
> 
>   src/examples/persistent_volume_framework.cpp fe2837cfffb6dd251ccb9c93197f623d0c55fb36

> 
> Diff: https://reviews.apache.org/r/43558/diff/
> 
> 
> Testing
> -------
> 
> Repeat test in CentOS 7.1
> ```
> sudo ./bin/mesos-tests.sh --gtest_filter="ExamplesTest.PersistentVolumeFramework" --gtest_repeat=200
--gtest_break_on_failure
> ```
> 
> 
> Thanks,
> 
> haosdent huang
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message