mosesdecoder/contrib/moses-speedtest/README.md

Moses speedtesting framework

Description

This is an automatic test framework that is designed to test the day to day performance changes in Moses.

Set up

Set up a Moses repo

Set up a Moses repo and build it with the desired configuration.

git clone https://github.com/moses-smt/mosesdecoder.git
cd mosesdecoder
./bjam -j10 --with-cmph=/usr/include/

You need to build Moses first, so that the testsuite knows what command you want it to use when rebuilding against newer revisions.

Create a parent directory.

Create a parent directory where the runtests.py and related scripts and configuration file should reside. This should also be the location of the TEST_DIR and TEST_LOG_DIR as explained in the next section.

Set up a global configuration file.

You need a configuration file for the testsuite. A sample configuration file is provided in testsuite_config

MOSES_REPO_PATH: /home/moses-speedtest/moses-standard/mosesdecoder
DROP_CACHES_COMM: sys_drop_caches 3
TEST_DIR: /home/moses-speedtest/phrase_tables/tests
TEST_LOG_DIR: /home/moses-speedtest/phrase_tables/testlogs
BASEBRANCH: RELEASE-2.1.1

The MOSES_REPO_PATH is the place where you have set up and built moses. The DROP_CACHES_COMM is the command that would beused to drop caches. It should run without needing root access. TEST_DIR is the directory where all the tests will reside. TEST_LOG_DIR is the directory where the performance logs will be gathered. It should be created before running the testsuite for the first time. BASEBRANCH is the branch against which all new tests will be compared. It should normally be set to be the latest Moses stable release.

Creating tests

In order to create a test one should go into the TEST_DIR and create a new folder. That folder will be used for the name of the test. Inside that folder one should place a configuration file named config. The naming is mandatory. An example such configuration file is test_config

Command: moses -f ... -i fff #Looks for the command in the /bin directory of the repo specified in the testsuite_config
LDPRE: ldpreloads #Comma separated LD_LIBRARY_PATH:/, 
Variants: vanilla, cached, ldpre #Can't have cached without ldpre or vanilla

The Command: line specifies the executable (which is looked up in the /bin directory of the repo.) and any arguments necessary. Before running the test, the script cds to the current test directory so you can use relative paths. The LDPRE: specifies if tests should be run with any LD_PRELOAD flags. The Variants: line specifies what type of tests should we run. This particular line will run the following tests:

1. A Vanilla test meaning just the command after Command will be issued.
2. A vanilla cached test meaning that after the vanilla test, the test will be run again without dropping caches in order to benchmark performance on cached filesystem.
3. A test with LD_PRELOAD ldpreloads moses -f command. For each available LDPRELOAD comma separated library to preload.
4. A cached version of all LD_PRELOAD tests.

Running tests.

Running the tests is done through the runtests.py script.

Running all tests.

To run all tests, with the base branch and the latests revision (and generate new basebranch test data if such is missing) do a:

python3 runtests.py -c testsuite_config

Running specific tests.

The script allows the user to manually run a particular test or to test against a specific branch or revision:

moses-speedtest@crom:~/phrase_tables$ python3 runtests.py --help
usage: runtests.py [-h] -c CONFIGFILE [-s SINGLETESTDIR] [-r REVISION]
                   [-b BRANCH]

A python based speedtest suite for moses.

optional arguments:
  -h, --help            show this help message and exit
  -c CONFIGFILE, --configfile CONFIGFILE
                        Specify test config file
  -s SINGLETESTDIR, --singletest SINGLETESTDIR
                        Single test name directory. Specify directory name,
                        not full path!
  -r REVISION, --revision REVISION
                        Specify a specific revison for the test.
  -b BRANCH, --branch BRANCH
                        Specify a branch for the test.

Generating HTML report.

To generate a summary of the test results use the html_gen.py script. It places a file named index.html in the current script directory.

python3 html_gen.py testsuite_config

You should use the generated file with the style.css file provided in the html directory.

Command line regression testing.

Alternatively you could check for regressions from the command line using the check_fo\r_regression.py script:

python3 check_for_regression.py TESTLOGS_DIRECTORY

Alternatively the results of all tests are logged inside the the specified TESTLOGS directory so you can manually check them for additional information such as date, time, revision, branch, etc...

Create a cron job:

Create a cron job to run the tests daily and generate an html report. An example cronjob is available.

#!/bin/sh
cd /home/moses-speedtest/phrase_tables

python3 runtests.py -c testsuite_config #Run the tests.
python3 html_gen.py testsuite_config #Generate html

cp index.html /fs/thor4/html/www/speed-test/ #Update the html

Place the script in /etc/cron.daily for dayly testing

Author

Nikolay Bogoychev, 2014

License

This software is licensed under the LGPL.