fa82de131c
In an effort to being really supporting using subunit2sql in a python3 environment this patch fixes all the issues with python3 incompatibility that show up when running tests. However, because the test coverage is limited for certain parts there are still probably pieces which don't work with python 3. Change-Id: I160d1675069836fa514547c31866b22bcaf29dcb |
||
---|---|---|
config-generator | ||
doc/source | ||
etc | ||
subunit2sql | ||
.gitignore | ||
.gitreview | ||
.testr.conf | ||
CONTRIBUTING.rst | ||
LICENSE | ||
MANIFEST.in | ||
README.rst | ||
requirements.txt | ||
setup.cfg | ||
setup.py | ||
test-requirements.txt | ||
TODO.rst | ||
tox.ini |
subunit2SQL README
subunit2SQL is a tool for storing test results data in a SQL database. Like it's name implies it was originally designed around converting subunit streams to data in a SQL database and the packaged utilities assume a subunit stream as the input format. However, the data model used for the DB does not preclude using any test result format. Additionally the analysis tooling built on top of a database is data format agnostic. However if you choose to use a different result format as an input for the database additional tooling using the DB api would need to be created to parse a different test result output format. It's also worth pointing out that subunit has several language library bindings available. So as a user you could create a small filter to convert a different format to subunit. Creating a filter should be fairly easy and then you don't have to worry about writing a tool like :ref:_subunit2sql to use a different format.
For multiple distributed test runs that are generating subunit output it is useful to store the results in a unified repository. This is the motivation for the _testrepository project which does a good job for centralizing the results from multiple test runs.
However, imagine something like the OpenStack CI system where the same basic test suite is normally run several hundreds of times a day. To provide useful introspection on the data from those runs and to build trends over time the test results need to be stored in a format that allows for easy querying. Using a SQL database makes a lot of sense for doing this, which was the original motivation for the project.
At a high level subunit2SQL uses alembic migrations to setup a DB
schema that can then be used by the subunit2sql
tool to parse subunit streams and populate
the DB. Then there are tools for interacting with the stored data in the
subunit2sql-graph
command as well as the sql2subunit
command to create a subunit stream from
data in the database. Additionally, subunit2sql provides a Python DB API
that can be used to query information from the stored data to build
other tooling.
Usage
DB Setup
The usage of subunit2sql is split into 2 stages. First you need to prepare a database with the proper schema; subunit2sql-db-manage should be used to do this. The utility requires db connection info which can be specified on the command or with a config file. Obviously the sql connector type, user, password, address, and database name should be specific to your environment. subunit2sql-db-manage will use alembic to setup the db schema. You can run the db migrations with the command:
subunit2sql-db-manage --database-connection mysql://subunit:pass@127.0.0.1/subunit upgrade head
or with a config file:
subunit2sql-db-manage --config-file subunit2sql.conf upgrade head
This will bring the DB schema up to the latest version for subunit2sql. Also, it is worth noting that the schema migrations used in subunit2sql do not currently support sqlite. While it is possible to fix this, sqlite only supports a subset of the necessary sql calls used by the migration scripts. As such, maintaining support for sqlite will be a continual extra effort, so if support is added back in the future, it is no guarantee that it will remain. In addition, the performance of running, even in a testing capacity, subunit2sql with MySQL or Postgres make it worth the effort of setting up one of them to use subunit2sql.
subunit2sql
Once you have a database setup with the proper database schema you can then use the subunit2sql command to populate the database with data from your test runs. subunit2sql takes in a subunit v2 either through stdin or by passing it file paths as positional arguments to the script. If only a subunit v1 stream is available, it can be converted to a subunit v2 stream using the subunit-1to2 utility.
There are several options for running subunit2sql, they can be listed with:
subunit2sql --help
The only required options are the state_path and the database-connections. These options and the other can either be used on the CLI, or put in a config file. If a config file is used you need to specify the location on the CLI.
Most of the optional arguments deal with how subunit2sql interacts with the SQL DB. However, it is worth pointing out that the artifacts option and the run_meta option are used to pass additional metadata into the database for the run(s) being added. The artifacts option should be used to pass in a url or path that points to any logs or other external test artifacts related to the run being added. The run_meta option takes in a dictionary which will be added to the database as key value pairs associated with the run being added.
sql2subunit
The sql2subunit utility is used for taking a run_id and creating a subunit v2 stream from the data in the DB about that run. To create a new subunit stream run:
sql2subunit $RUN_ID
along with any options that you would normally use to either specify a config file or the DB connection info. Running this command will print to stdout the subunit v2 stream for the run specified by $RUN_ID, unless the --out_path argument is specified to write it to a file instead.