Continuous testing and deployment in Perl (London.pm Technical Meeting October 2012)

Continuous Deployment in PerlPresentation by Alex Balhatchet (KAORU)

Intro

What is continuous deployment?

Commit Changes

Test Changes

Deploy Changes As quickly as possible! :-)

Committing Changes

● Use your favourite VCS

● Push the changes somewhere centralized (staging/testing environment)

● This should kick off a "build and test" cycle if one is not currently running

Testing Changes

Perl is absolutely fantastic for testing.

● TAP

● TAP::Harness / App::Prove

● Test::Builder / Test::More

Deploying Changes

● Pick your favourite deployment method

● Make it as fast as possible

● Make it as easy as possible

Challenges

● Fast tests

● Reliable tests

● Fast deployment

Challenges

● Fast tests

● Reliable tests

● Fast deployment

Focus of this talk

Perl Testing

TAP::Harness and App::Prove

TAP::Harness and App::Prove

/usr/bin/prove

uses App::Prove

uses TAP::Harness

App::Prove features - basics

● Run lots of tests with one command

● Hide non-failure output

● Run tests in parallel

● Show timing information

App::Prove features - basics

alex@karin$ prove -l -r -j 16 t/t/00-load.t .......... ok t/01-backup.t ........ ok t/02-delete_files.t .. ok All tests successful.Files=3, Tests=51, 1 wallclock secs ( 0.06 usr 0.01 sys + 0.78 cusr 0.20 csys = 1.05 CPU)Result: PASS

App::Prove features - advanced

● Load plugins from App::Prove::Plugin::*

● Save the TAP results to a .tgz archive

● Save state between runs and use that state to change the order for future runs

App::Prove features - advancedalex@karin$ prove -l -r --state=hot,all,save t/No saved state, selection will be emptyt/00-load.t .. okt/01-good.t .. okt/02-bad.t ... 1/1

alex@karin$ prove -l -r --state=hot,all,save t/t/02-bad.t ... 1/1 <===== Runs first!t/00-load.t .. okt/01-good.t .. ok

App::Prove features - advanced

Other "state" options:

● slow - very useful with -j N

● fresh - only run the tests you need to

● last - same as last run, even with --shuffle

Test::Aggregate

Test::Aggregate

● Combine multiple .t files into a single file

● Avoids compilation overhead

● Therefore speeds up your tests!

Test::Aggregate::Nested

● Alpha code

● Uses the fairly recently added syntax for subtests in TAP

● Works really well for us!

Test::Aggregate caveats

● Breaks BEGIN and END blocks

(use Scope::Guard instead!)

● exit() is also a bad idea!

● Be aware of Test modules which use these

(Test::NoWarnings is the most common!)

Parallelisation of Aggregated Tests

Currently closed source - sorry! The logic is:

i. Split up tests into N groups

ii. Copy .t files into temporary directory

iii. Write temporary .t file that uses Test::Aggregate::Nested

iv. Run those tests with prove -j N

TAP::Harness::Archive

TAP::Harness::Archive

● "isa" TAP::Harness

● Stores the TAP output in a tar archive

● Gives you full TAP even when running in parallel

● Allows you to run your tests once and then format the results in multiple ways

How we format our TAP

● Process the output from our parallelised aggregated test runs

● Match up the failures with the .t files

● Write a summary in Markdown format that's later converted to HTML and emailed to us

How you might format your TAP

● TAP::Formatter::JUnit (eg. for Jenkins)

● TAP::Formatter::TeamCity

● TAP::Formatter::HTML

TAP::Formatter::HTML example

TAP::Formatter::HTML example

Putting it all together

Continuous Deployment at Lokku

svn commit

Perl daemon notices commit

Parallelised and aggregated test run starts running

Nicely formatted results are emailed on success or failure

On test success new code is automatically pushed

Deployment is one command and very quick

What's next?

What's next?

● Test::WWW::Selenium::More

● Faster tests - more unit tests, fewer integration tests

● Parallelised deployment

Any questions?

Thanks!