Automation #8
Automation #8
Conversation
|
At first glance everything here seems clear to me. However I haven't actually run through things myself yet. @TomAugspurger and @pitrou might also have thoughts. |
| DASK_CONFIG=${DASK_ASV_CONFIG:-$HOME/asv.dask.conf.json} | ||
| DISTRIBUTED_CONFIG=${DISTRIBUTED_ASV_CONFIG:-$HOME/asv.distributed.conf.json} | ||
|
|
||
| source activate dask-asv |
There was a problem hiding this comment.
How is this environment created?
There was a problem hiding this comment.
ah yes, I forgot i wanted to add in here steps that will create the environment if not already present.
|
|
||
| echo "Running dask benchmarks..." | ||
| cd $DASK_DIR | ||
| asv --config $DASK_CONFIG run NEW |
There was a problem hiding this comment.
Should also run new benchmarks, see airspeed-velocity/asv#278
| # can get it to work. However directory structure gets messed up and machine.json | ||
| # is not in the correct location to generate the graphs. Thus this hack to copy it | ||
| # to the right locations before running publish. | ||
| find /home/ec2-user/results/distributed/aws-ec2-c4.xlarge -type d -exec cp /home/ec2-user/results/distributed/aws-ec2-c4.xlarge/machine.json {} \; |
There was a problem hiding this comment.
Is there a way to avoid hardcoding aws-ec2-c4.xlarge?
There was a problem hiding this comment.
This whole line is just a workaround to install dask master. I have an open issue for asv here: airspeed-velocity/asv#494. Plan to take a look at applying pv's fix.
|
@pitrou OK I added in the conda env creation and running new benchmarks. It always reruns benchmarks for distributed at the |
|
No comments from me, other than a thanks to @postelrich for setting this up. |
This PR contains a bash script and relevant asv configuration files to run benchmarking as a cron script. It is setup to run for both dask and distributed and to push to the gh-pages branch. You can already see the published benchmarks for both:
https://dask.github.io/dask-benchmarks/results/distributed
https://dask.github.io/dask-benchmarks/results/dask
Distributed was a little tricky because of it's heavy dependency on dask. To work around this I set it so asv install dask github master. Unfortunately this is not supported with asv and a little hack was added to ensure a file is in the right place for html generation. I was also unable to run benchmarks for previous versions of distributed as
distributed.Clientdid not exist yet, which the tests depend on.