Dropping the py35 testing
All the integration testing has been moved to Bionic now and py3.5 is not tested runtime for Train or stable/stein. As per below ML thread, we are good to drop the py35 testing now: http://lists.openstack.org/pipermail/openstack-discuss/2019-April/005097.html  http://lists.openstack.org/pipermail/openstack-discuss/2019-April/004647.html  https://governance.openstack.org/tc/reference/runtimes/stein.html https://governance.openstack.org/tc/reference/runtimes/train.html Change-Id: Idc1f3561f775de39166ef048a0d84b060258299e
|1 month ago|
|app||4 months ago|
|doc/source||10 months ago|
|etc||3 years ago|
|gulp||1 year ago|
|openstack_health||5 months ago|
|test||10 months ago|
|.coveragerc||3 years ago|
|.eslintignore||2 years ago|
|.eslintrc||3 years ago|
|.gitignore||1 year ago|
|.gitreview||1 month ago|
|.jshintrc||3 years ago|
|.stestr.conf||1 year ago|
|.testr.conf||3 years ago|
|.zuul.yaml||1 month ago|
|CONTRIBUTING.rst||3 years ago|
|HACKING.rst||7 months ago|
|LICENSE||3 years ago|
|README.rst||9 months ago|
|gulpfile.js||3 years ago|
|package.json||1 year ago|
|requirements.txt||1 year ago|
|setup.cfg||1 month ago|
|setup.py||1 year ago|
|test-requirements.txt||1 year ago|
|tox.ini||1 month ago|
webclient for visualizing test results of OpenStack CI jobs.
Make sure the python dependencies are installed preferably in a virtualenv if doing development work:
$ pip install -r requirements.txt
Installation of the frontend requires Node.js and Gulp.
$ sudo apt-get install nodejs npm nodejs-legacy $ sudo npm -g install npm@2 $ sudo npm -g config set prefix /usr/local $ sudo npm -g install npm $ sudo npm -g install gulp
OSX (via HomeBrew, note no sudo):
$ brew install nodejs $ npm install -g gulp
Then, install the Node modules by running, from the project directory:
$ npm install
To run the REST API for development you can install the openstack_health python package in development mode and start the API service with:
$ python setup.py develop $ openstack-health-api <config_file>
or alternatively just can just run the api.py file manually. For example, from the top of the repo you would run:
$ python2 openstack_health/api.py <config_file>
A sample of
<config_file> can be found in
etc/openstack-health-api.conf. This will start up a local webserver listening on localhost. You can then send requests to the specified port on stdout to see the response.
A development server can be run as follows:
$ gulp dev
This will open a web browser and reload code automatically as it changes on the filesystem.
The rest api is a flask application so any of the methods for deploying a flask application can be used. The standalone entrypoint used for development isn't suitable for production because it's single threaded. You should use a wsgi container, something like uwsgi, gunicorn, or mod_wsgi to deploy it for real. For example, running the API with uwsgi standalone you can do something like:
$ uwsgi -s /tmp/uwsgi.sock --module openstack_health.api --callable app --pyargv config_file --http :5000
That will startup a uwsgi server running the rest api on port 5000.
Depending on how your backend subunit2sql database is configured and how you've configured the WSGI app deployed you might want to adjust the access patterns used for querying the database. There are 3 options related to database access. The first is mandatory, db_uri is the sqlalchemy connection string for connecting to the database. The second option is pool_size which sets the size of the connection pool to use, the default being 20. This option is used to set the maximum number of persistent connections allowed per process. It can be set to 0 to indicate there is no limit and -1 is used to disable pooling (which is normally a really bad idea). The last option is pool_recycle which is used to prevent the pool from using a connection which has passed a certain age. This is a value in seconds and it defaults to 3600 (1 hour) for more info about this option refer to the sqlalchemy documentation: http://docs.sqlalchemy.org/en/latest/core/pooling.html#setting-pool-recycle
There are certain API queries that return an RSS feed on the wire. As part of the events generated on these feed there are links to pages on the JS frontend. To generate the correct links you have to tell the api server the base url of the frontend. This is set with the frontend_url option and it defaults to 'http://status.openstack.org/openstack-health' the OpenStack community instance of OpenStack-Health.
There are certain API operations which will use the elastic-recheck project to pull in additional information about failures that occur during a run. However, since elastic-recheck is not widely deployed this is an optional feature and is only enabled if elastic-recheck is installed. (and importable by the API server) Also note that elastic-recheck is not published on pypi and must be manually installed via git. Additionally, after you install elastic-recheck you also need to configure the location of the queries by using the query_dir configuration option. If this is not set than the elastic-recheck support will be disabled. Optionally, if you need to set the url of you elasticsearch API endpoint you can set this with the es_url configuration option. By default it is configured to talk to openstack-infra's elasticsearch server at http://logstash.openstack.org/elasticsearch
Since the introduction of elastic recheck querying dogpile.cache has been used to cache any request that hits elasticsearch. This is because the query times for using elastic-recheck are quite variable and often very slow. (at least for talking to openstack-infra's elasticsearch) To enable reasonable interactive response times we cache the api response from requests using elasticsearch data. Note, that this caching is enabled regardless of whether elastic-recheck is enabled or not.
There are four configuration options available around configuring caching. While the defaults were picked to work in most situations depending on your specific deployment specifics there are other choices that might make more sense.
The first is cache_backend which is used to set the python class for the dogpile.cache.api.CacheBackend to use. By default this is set to dogpile.cache.dbm which uses a DBM file on disk. You can effectively disable all caching by setting this value to dogpile.cache.null.
The second option is cache_expiration which is used to set the timeout value to use for any cached responses. This is an integer for the number of seconds to keep a response cached. By default this is set to 30mins.
The third option is cache_file which is used to set the file path when using the DBM backend is used. By default this is configured to use TEMPDIR/openstack-health.dbm
The fourth option is cache_url which is used to provide the url to an external service, like memcached, for storing the cache data. This only needs to be set if you're using a backend that requires this.
It also should be noted that when configuring caching using a non-default backend the API server will attempt to configure refreshing the cache asynchronously with a background thread. This makes the end user response near instantaneous in all cases because the cache is updated in the background instead of on an incoming request.
Recommended Production Cache Configuration: The recommend way to configure your cache is to have memcached setup to use for distributed locking and then use the default dbm file store for the actual caching. This enables using an async worker that will update the cache in the background ensuring that users will never receive an uncached response. To set this up you need to have memcached installed and running, then set the cache_url option set to the hostname for that server. After that the defaults to use the dogpile.cache.dbm backend are sufficient, however you can change the cache_file to live somewhere else. You can also set the cache_expiration to be a much lower value because the async worker updates the cache in the background, so you don't have to worry about a stale cache having a user facing performance impact.
The production application can be build using:
$ gulp prod
The result will be written to
./build and should be appropriate for distribution. Note that all files are not required:
css/main.css): required unless gzipped versions are used.
*.gz): not required, but preferred. Use instead of plain core files to save on disk usage and bandwidth.
js/main.js.map.gz): only required for debugging purposes.
To test python code, run:
$ tox -e py27
The frontend tests
npm test and
npm run unit use a headless chrome driver and the driver requires the chromium-browser package to be installed on the system, you can do this by running:
$ sudo apt-get install chromium-browser
When using an operating system that is not ubuntu 16.04 the process.env.CHROME_BIN variable may need to be updated in openstack-health/test/karma.conf.js to reflect your system's chromium path.
For example on SUSE Leap OS, process.env.CHROME_BIN = '/usr/bin/chromium' must be set.
$ npm test
This will execute both unit and end-to-end tests, and will write coverage reports to
./cover. To individually run unit tests and generate coverage reports, run:
$ npm run unit
Similarly, to run only end-to-end tests, run:
$ npm run protractor
Alternatively, you can start the karma server and have it watch for changes in your files so that unit tests are run every time they change, allowing for much faster feedback:
./node_modules/karma/bin/karma start test/karma.conf.js --no-single-run