Python package with perftracker client, tools, tests and libraries
Project description
# perftracker-lib
A client library for the [perftracker](https://github.com/perfguru87/perftracker) and a set of libraries for performance testing
## Supported python version
python2.7
python3.0+
## Building and installing the perftracker-client python package
Installing from pypi.org:
```
pip install perftrackerlib
```
Installing from sources:
```
python3 ./setup.py build
python3 ./setup.py install
```
## Usage Examples
### Python-written Test Suites
Minimalistic test suite:
```
python3 ./examples/pt_suite_example_minimal.py --pt-title="Website suite run" --pt-url http://perftracker.localdomain:9000
```
Simulate a 'website' suite run and upload results:
```
python3 ./examples/pt_suite_example_fake.py -v --pt-title="Website suite run" --pt-project="Default project" --pt-url http://perftracker.localdomain:9000
```
Use code like `examples/pt_suite_example_populate.sh` to mass populate perftracker with fake data
### Control Panel Crawler
Run selenium-based test on a real WordPress Admin panel:
```
python3 ./examples/pt-wp-crawler.py -m -U user -P user https://demo.wpjobboard.net/wp-login.php
```
### Upload pre-generated files with tests results:
Sometimes you don't want to write a python suite and just grab some files and export results. In this case
you can use the pt-suite-uploader.py tool to parse test/json files (or even launch an external tool) and then
upload results:
```
python3 ./tools/pt-suite-uploader.py -f ./examples/data/sample.txt
python3 ./tools/pt-suite-uploader.py -f -j ./examples/data/sample.json
python3 ./tools/pt-suite-uploader.py -- /bin/echo "tag: my test; score: 2.3;"
...
```
### Manage artifacts (i.e. jobs and tests attachments)
The perftracker server supports [artifact management](https://github.com/perfguru87/perftracker)
An artifact is a file which can be stored as blob file and linked to test or job run, for example
it can be test or job log, dump or some test data. Many to many links are allowed
There are three ways how clients can managet the artifacts:
1. perftracker REST API
2. perftrackerlib/client.py - ptArfitact() class
3. the ./tools/pt-artifact-ctl.py tool (see --help)
Short introuduction to pt-artifact-ctl.py:
a) Help
```
pt-artifact-ctl.py --help
Usage: pt-artifact-ctl.py [options] command [command parameters]
Description:
pt-artifact-ctl.py [options] upload ARTIFACT_FILE_TO_UPLOAD [ARTIFACT_UUID]
pt-artifact-ctl.py [options] update ARTIFACT_UUID
pt-artifact-ctl.py [options] delete ARTIFACT_UUID
pt-artifact-ctl.py [options] info ARTIFACT_UUID
pt-artifact-ctl.py [options] link ARTIFACT_UUID OBJECT_UUID
pt-artifact-ctl.py [options] unlink ARTIFACT_UUID OBJECT_UUID
pt-artifact-ctl.py [options] list [LIMIT]
pt-artifact-ctl.py [options] download ARTIFACT_UUID ARTIFACT_FILE_TO_SAVE
Options:
-h, --help show this help message and exit
-v, --verbose enable verbose mode
-p PT_SERVER_URL, --pt-server-url=PT_SERVER_URL
perftracker url, default http://127.0.0.1:9000
-d DESCRIPTION, --description=DESCRIPTION
artifact description (i
-m MIME, --mime=MIME artifact mime type, default is guessed or
'application/octet-stream'
-f FILENAME, --filename=FILENAME
override artifact file name by given name
-z, --compression inline decompression on every file view or
download
-i, --inline inline view in browser (do not download on click)
-t TTL, --ttl=TTL time to live (days), default=180, 0 - infinite
```
b) Upload an artifact and link in to the test with uuid = $TEST_UUID
```
./pt-artifact-ctl.py upload ~/my_test.log
./pt-artifact-ctl.py link $ARTIFACT_UUID $TEST_UUID
```
c) Upload an artifact, set infinite time to live, enable dynamic compression and enable inline view in the browser
```
./pt-artifact-ctl.py upload ~/my_test.log -iz -t 0
```
## Contributing a patch
Make a change and test your code before commit:
```
python ./test.py
```
## Release notes
See [http://www.perftracker.org/client/#Release_Notes](http://www.perftracker.org/client/#Release_Notes)
A client library for the [perftracker](https://github.com/perfguru87/perftracker) and a set of libraries for performance testing
## Supported python version
python2.7
python3.0+
## Building and installing the perftracker-client python package
Installing from pypi.org:
```
pip install perftrackerlib
```
Installing from sources:
```
python3 ./setup.py build
python3 ./setup.py install
```
## Usage Examples
### Python-written Test Suites
Minimalistic test suite:
```
python3 ./examples/pt_suite_example_minimal.py --pt-title="Website suite run" --pt-url http://perftracker.localdomain:9000
```
Simulate a 'website' suite run and upload results:
```
python3 ./examples/pt_suite_example_fake.py -v --pt-title="Website suite run" --pt-project="Default project" --pt-url http://perftracker.localdomain:9000
```
Use code like `examples/pt_suite_example_populate.sh` to mass populate perftracker with fake data
### Control Panel Crawler
Run selenium-based test on a real WordPress Admin panel:
```
python3 ./examples/pt-wp-crawler.py -m -U user -P user https://demo.wpjobboard.net/wp-login.php
```
### Upload pre-generated files with tests results:
Sometimes you don't want to write a python suite and just grab some files and export results. In this case
you can use the pt-suite-uploader.py tool to parse test/json files (or even launch an external tool) and then
upload results:
```
python3 ./tools/pt-suite-uploader.py -f ./examples/data/sample.txt
python3 ./tools/pt-suite-uploader.py -f -j ./examples/data/sample.json
python3 ./tools/pt-suite-uploader.py -- /bin/echo "tag: my test; score: 2.3;"
...
```
### Manage artifacts (i.e. jobs and tests attachments)
The perftracker server supports [artifact management](https://github.com/perfguru87/perftracker)
An artifact is a file which can be stored as blob file and linked to test or job run, for example
it can be test or job log, dump or some test data. Many to many links are allowed
There are three ways how clients can managet the artifacts:
1. perftracker REST API
2. perftrackerlib/client.py - ptArfitact() class
3. the ./tools/pt-artifact-ctl.py tool (see --help)
Short introuduction to pt-artifact-ctl.py:
a) Help
```
pt-artifact-ctl.py --help
Usage: pt-artifact-ctl.py [options] command [command parameters]
Description:
pt-artifact-ctl.py [options] upload ARTIFACT_FILE_TO_UPLOAD [ARTIFACT_UUID]
pt-artifact-ctl.py [options] update ARTIFACT_UUID
pt-artifact-ctl.py [options] delete ARTIFACT_UUID
pt-artifact-ctl.py [options] info ARTIFACT_UUID
pt-artifact-ctl.py [options] link ARTIFACT_UUID OBJECT_UUID
pt-artifact-ctl.py [options] unlink ARTIFACT_UUID OBJECT_UUID
pt-artifact-ctl.py [options] list [LIMIT]
pt-artifact-ctl.py [options] download ARTIFACT_UUID ARTIFACT_FILE_TO_SAVE
Options:
-h, --help show this help message and exit
-v, --verbose enable verbose mode
-p PT_SERVER_URL, --pt-server-url=PT_SERVER_URL
perftracker url, default http://127.0.0.1:9000
-d DESCRIPTION, --description=DESCRIPTION
artifact description (i
-m MIME, --mime=MIME artifact mime type, default is guessed or
'application/octet-stream'
-f FILENAME, --filename=FILENAME
override artifact file name by given name
-z, --compression inline decompression on every file view or
download
-i, --inline inline view in browser (do not download on click)
-t TTL, --ttl=TTL time to live (days), default=180, 0 - infinite
```
b) Upload an artifact and link in to the test with uuid = $TEST_UUID
```
./pt-artifact-ctl.py upload ~/my_test.log
./pt-artifact-ctl.py link $ARTIFACT_UUID $TEST_UUID
```
c) Upload an artifact, set infinite time to live, enable dynamic compression and enable inline view in the browser
```
./pt-artifact-ctl.py upload ~/my_test.log -iz -t 0
```
## Contributing a patch
Make a change and test your code before commit:
```
python ./test.py
```
## Release notes
See [http://www.perftracker.org/client/#Release_Notes](http://www.perftracker.org/client/#Release_Notes)
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
perftrackerlib-0.0.39.tar.gz
(215.7 kB
view hashes)