Accelerating CI with AWS CodeBuild: Parallel check execution now obtainable

I’m excited to announce that AWS CodeBuild now helps parallel check execution, so you possibly can run your check suites concurrently and scale back construct instances considerably.

With the demo undertaking I wrote for this submit, the full check time went down from 35 minutes to six minutes, together with the time to provision the environments. These two screenshots from the AWS Administration Console present the distinction.

Sequential execution of the check suite

Parallel execution of the check suite

Very lengthy check instances pose a major problem when working steady integration (CI) at scale. As initiatives develop in complexity and group measurement, the time required to execute complete check suites can enhance dramatically, resulting in prolonged pipeline execution instances. This not solely delays the supply of latest options and bug fixes, but in addition hampers developer productiveness by forcing them to attend for construct outcomes earlier than continuing with their duties. I’ve skilled pipelines that took as much as 60 minutes to run, solely to fail on the final step, requiring a whole rerun and additional delays. These prolonged cycles can erode developer belief within the CI course of, contribute to frustration, and in the end decelerate all the software program supply cycle. Furthermore, long-running assessments can result in useful resource competition, elevated prices due to wasted computing energy, and decreased general effectivity of the event course of.

With parallel check execution in CodeBuild, now you can run your assessments concurrently throughout a number of construct compute environments. This function implements a sharding method the place every construct node independently executes a subset of your check suite. CodeBuild supplies setting variables that establish the present node quantity and the full variety of nodes, that are used to find out which assessments every node ought to run. There isn’t a management construct node or coordination between nodes at construct time—every node operates independently to execute its assigned portion of your assessments.

To allow check splitting, configure the batch fanout part in your buildspec.xml, specifying the specified parallelism degree and different related parameters. Moreover, use the codebuild-tests-run utility in your construct step, together with the suitable check instructions and the chosen splitting technique.

The assessments are cut up primarily based on the sharding technique you specify. codebuild-tests-run presents two sharding methods:

• Equal-distribution. This technique kinds check recordsdata alphabetically and distributes them in chunks equally throughout parallel check environments. Modifications within the names or amount of check recordsdata would possibly reassign recordsdata throughout shards.
• Stability. This technique fixes the distribution of assessments throughout shards by utilizing a constant hashing algorithm. It maintains present file-to-shard assignments when new recordsdata are added or eliminated.

CodeBuild helps automated merging of check stories when working assessments in parallel. With automated check report merging, CodeBuild consolidates assessments stories right into a single check abstract, simplifying outcome evaluation. The merged report consists of aggregated cross/fail statuses, check durations, and failure particulars, lowering the necessity for guide report processing. You possibly can view the merged ends in the CodeBuild console, retrieve them utilizing the AWS Command Line Interface (AWS CLI), or combine them with different reporting instruments to streamline check evaluation.

Let’s have a look at the way it works
Let me exhibit the way to implement parallel testing in a undertaking. For this demo, I created a really fundamental Python undertaking with tons of of assessments. To hurry issues up, I requested Amazon Q Developer on the command line to create a undertaking and 1,800 check instances. Every check case is in a separate file and takes one second to finish. Operating all assessments in a sequence requires half-hour, excluding the time to provision the setting.

On this demo, I run the check suite on ten compute environments in parallel and measure how lengthy it takes to run the suite.

To take action, I added a buildspec.yml file to my undertaking.

model: 0.2

batch:
  fast-fail: false
  build-fanout:
    parallelism: 10 # ten runtime environments 
    ignore-failure: false

phases:
  set up:
    instructions:
      - echo 'Putting in Python dependencies'
      - dnf set up -y python3 python3-pip
      - pip3 set up --upgrade pip
      - pip3 set up pytest
  construct:
    instructions:
      - echo 'Operating Python Exams'
      - |
         codebuild-tests-run 
          --test-command 'python -m pytest --junitxml=report/test_report.xml' 
          --files-search "codebuild-glob-search 'assessments/test_*.py'" 
          --sharding-strategy 'equal-distribution'
  post_build:
    instructions:
      - echo "Check execution accomplished"

stories:
  pytest_reports:
    recordsdata:
      - "*.xml"
    base-directory: "report"
    file-format: JUNITXML 

There are three elements to focus on within the YAML file.

First, there’s a build-fanout part below batch. The parallelism command tells CodeBuild what number of check environments to run in parallel. The ignore-failure command signifies if failure in any of the fanout construct duties may be ignored.

Second, I take advantage of the pre-installed codebuild-tests-run command to run my assessments.

This command receives the whole checklist of check recordsdata and decides which of the assessments should be run on the present node.

• Use the sharding-strategy argument to decide on between equally distributed or steady distribution, as I defined earlier.
• Use the files-search argument to cross all of the recordsdata which might be candidates for a run. We advocate to make use of the offered codebuild-glob-search command for efficiency causes, however any file search software, similar to discover(1), will work.
• I cross the precise check command to run on the shard with the test-command argument.

Lastly, the stories part instructs CodeBuild to gather and merge the check stories on every node.

Then, I open the CodeBuild console to create a undertaking and a batch construct configuration for this undertaking. There’s nothing new right here, so I’ll spare you the small print. The documentation has all the small print to get you began.  Parallel testing works on batch builds. Be certain to configure your undertaking to run in batch.

Now, I’m able to set off an execution of the check suite. I can commit new code on my GitHub repository or set off the construct within the console.

After a couple of minutes, I see a standing report of the totally different steps of the construct; with a standing for every check setting or shard.

When the check is full, I choose the Studies tab to entry the merged check stories.

The Studies part aggregates all check knowledge from all shards and retains the historical past for all builds. I choose my most up-to-date construct within the Report historical past part to entry the detailed report.

As anticipated, I can see the aggregated and the person standing for every of my 1,800 check instances. On this demo, they’re all passing, and the report is inexperienced.

The 1,800 assessments of the demo undertaking take one second every to finish. Once I run this check suite sequentially, it took 35 minutes to finish. Once I run the check suite in parallel on ten compute environments, it took 6 minutes to finish, together with the time to provision the environments. The parallel run took 17.9 % of the time of the sequential run. Precise numbers will fluctuate together with your initiatives.

Further issues to know
This new functionality is appropriate with all testing frameworks. The documentation consists of examples for Django, Elixir, Go, Java (Maven), Javascript (Jest), Kotlin, PHPUnit, Pytest, Ruby (Cucumber), and Ruby (RSpec).

For check frameworks that don’t settle for space-separated lists, the codebuild-tests-run CLI supplies a versatile various via the CODEBUILD_CURRENT_SHARD_FILES setting variable. This variable comprises a newline-separated checklist of check file paths for the present construct shard. You should utilize it to adapt to totally different check framework necessities and format check file names.

You possibly can additional customise how assessments are cut up throughout environments by writing your personal sharding script and utilizing the CODEBUILD_BATCH_BUILD_IDENTIFIER setting variable, which is mechanically set in every construct. You should utilize this method to implement framework-specific parallelization or optimization.

Pricing and availability
With parallel check execution, now you can full your check suites in a fraction of the time beforehand required, accelerating your growth cycle and bettering your group’s productiveness.

Parallel check execution is obtainable on all three compute modes supplied by CodeBuild: on-demand, reserved capability, and AWS Lambda compute.

This functionality is obtainable as we speak in all AWS Areas the place CodeBuild is obtainable, with no further value past the usual CodeBuild pricing for the compute assets used.

I invite you to attempt parallel check execution in CodeBuild as we speak. Go to the AWS CodeBuild documentation to be taught extra and get began with parallelizing your assessments.

— seb

PS: Right here’s the immediate I used to create the demo software and its check suite: “I’m writing a weblog submit to announce codebuild parallel testing. Write a quite simple python app that has tons of of assessments, every check in a separate check file. Every check takes one second to finish.”

How is the Information Weblog doing? Take this 1 minute survey!

(This survey is hosted by an exterior firm. AWS handles your info as described within the AWS Privateness Discover. AWS will personal the info gathered through this survey and won’t share the data collected with survey respondents.)