[SPARK-7721][PYTHON][TESTS] Adds PySpark coverage generation script
authorhyukjinkwon <gurwls223@gmail.com>
Mon, 22 Jan 2018 13:12:50 +0000 (22:12 +0900)
committerhyukjinkwon <gurwls223@gmail.com>
Mon, 22 Jan 2018 13:12:50 +0000 (22:12 +0900)
commit87ffe7adddf517541aac0d1e8536b02ad8881606
treeef234206c1293271f8cb66e7a5e9b395bf571fac
parent5d680cae486c77cdb12dbe9e043710e49e8d51e4
[SPARK-7721][PYTHON][TESTS] Adds PySpark coverage generation script

## What changes were proposed in this pull request?

Note that this PR was made based on the top of https://github.com/apache/spark/pull/20151. So, it almost leaves the main codes intact.

This PR proposes to add a script for the preparation of automatic PySpark coverage generation. Now, it's difficult to check the actual coverage in case of PySpark. With this script, it allows to run tests by the way we did via `run-tests` script before. The usage is exactly the same with `run-tests` script as this basically wraps it.

This script and PR alone should also be useful. I was asked about how to run this before, and seems some reviewers (including me) need this. It would be also useful to run it manually.

It usually requires a small diff in normal Python projects but PySpark cases are a bit different because apparently we are unable to track the coverage after it's forked. So, here, I made a custom worker that forces the coverage, based on the top of https://github.com/apache/spark/pull/20151.

I made a simple demo. Please take a look - https://spark-test.github.io/pyspark-coverage-site.

To show up the structure, this PR adds the files as below:

```
python
├── .coveragerc  # Runtime configuration when we run the script.
├── run-tests-with-coverage  # The script that has coverage support and wraps run-tests script.
└── test_coverage  # Directories that have files required when running coverage.
    ├── conf
    │   └── spark-defaults.conf  # Having the configuration 'spark.python.daemon.module'.
    ├── coverage_daemon.py  # A daemon having custom fix and wrapping our daemon.py
    └── sitecustomize.py  # Initiate coverage with COVERAGE_PROCESS_START
```

Note that this PR has a minor nit:

[This scope](https://github.com/apache/spark/blob/04e44b37cc04f62fbf9e08c7076349e0a4d12ea8/python/pyspark/daemon.py#L148-L169) in `daemon.py` is not in the coverage results as basically I am producing the coverage results in `worker.py` separately and then merging it. I believe it's not a big deal.

In a followup, I might have a site that has a single up-to-date PySpark coverage from the master branch as the fallback / default, or have a site that has multiple PySpark coverages and the site link will be left to each pull request.

## How was this patch tested?

Manually tested. Usage is the same with the existing Python test script - `./python/run-tests`. For example,

```
sh run-tests-with-coverage --python-executables=python3 --modules=pyspark-sql
```

Running this will generate HTMLs under `./python/test_coverage/htmlcov`.

Console output example:

```
sh run-tests-with-coverage --python-executables=python3,python --modules=pyspark-core
Running PySpark tests. Output is in /.../spark/python/unit-tests.log
Will test against the following Python executables: ['python3', 'python']
Will test the following Python modules: ['pyspark-core']
Starting test(python): pyspark.tests
Starting test(python3): pyspark.tests
...
Tests passed in 231 seconds
Combining collected coverage data under /.../spark/python/test_coverage/coverage_data
Reporting the coverage data at /...spark/python/test_coverage/coverage_data/coverage
Name                         Stmts   Miss Branch BrPart  Cover
--------------------------------------------------------------
pyspark/__init__.py             41      0      8      2    96%
...
pyspark/profiler.py             74     11     22      5    83%
pyspark/rdd.py                 871     40    303     32    93%
pyspark/rddsampler.py           68     10     32      2    82%
...
--------------------------------------------------------------
TOTAL                         8521   3077   2748    191    59%
Generating HTML files for PySpark coverage under /.../spark/python/test_coverage/htmlcov
```

Author: hyukjinkwon <gurwls223@gmail.com>

Closes #20204 from HyukjinKwon/python-coverage.
.gitignore
python/.coveragerc [new file with mode: 0644]
python/run-tests-with-coverage [new file with mode: 0755]
python/run-tests.py
python/test_coverage/conf/spark-defaults.conf [new file with mode: 0644]
python/test_coverage/coverage_daemon.py [new file with mode: 0644]
python/test_coverage/sitecustomize.py [new file with mode: 0644]