Conda Windows CI migration: execution plan

[j-rivero]

Related to #373

As context: the buildfarm expects to run the nightly generation at 10:00am Spain time and the DSL regenerate for libraries run at 10:39am. Should be fine to disable the job until the migration is done to have the morning clean.

Migration plan

• ACTION: Disable nightly scheduler for Jan 28th https://build.osrfoundation.org/job/ignition-jetty-nightly-scheduler/
  • DSL should return it back automatically next day
• CHECK: Smoke testing in testing agents on Conda setup in the master branch of release-tools
  • Looking good. Uses wrong environment but that is expected, cancelling.
• ACTION: disable Windows testing agents (manually in AWS console)
• ACTION: Switch agents to use the GPU AMI in production https://github.com/osrf/osrf-terraform/pull/192
• ACTION: refresh agents in production
• CHECK: Run and the ign-rendering6 testing Conda job in production nodes.
  • Results should be the same than current job
• ACTION: Merge release-tools change Switch to use Conda for Windows jobs #1245
• ACTION: Run DSL gazebo_libs https://build.osrfoundation.org/job/_dsl_gazebo_libs/
  • https://build.osrfoundation.org/job/_dsl_gazebo_libs/295/
• CHECK: Run and the ign-rendering6 and gz-sim -ci-clwin from Fortress
  • Conda is being really used
  • Hot fix applied in Fix typo in checking Conda paths #1249
• CHECK: Run and the gz-rendering8 and gz-rendering9 from Harmonic / Ionic
  • Conda is being really used
  • gz-cmake as smoke test
• CHECK: existing PR, re-triggered jobs
  • Two jobs should appear: green old one with pr-win and new one with
  • Doing it in: Add example about using joint control components gazebosim/gz-sim#2728 (comment)
  • Hotfix Disable testing and CMake on pr-win stub #1250
• ACTION: launch the full testing

Full testing

Launching all the new jobs:

def jenkins = Jenkins.instance

jenkins.items.each { job ->
    // Check if the job name ends with -cwin or -clowin, does not include -pr-, and has no builds yet
    if ((job.name.endsWith("-clwin") || job.name.endsWith("-clowin")) && 
        !job.name.contains("-pr-") && 
        job.builds.isEmpty()) {
        
        println "Triggering job: ${job.name}"
        job.scheduleBuild(0)
    }
}

Post migration

• Prepare the local builder for the Conda Windows jobs
  • Code is in jrivero/support_conda
• ACTION Improve the testing AMI according to https://github.com/osrf/osrf-terraform/issues/190
• WAIT 2 weeks: clean references to USE_PIXI and vcpkg

Revert to current state

If things do not work well, the migration can be reverted. What would happen following the next steps:

• CI: previous -ci-*-win jobs will be enabled again.
• CI: new -ci-*-clwin or -ci-*-clowinjobs will be disabled.
• PR: the -ci-pr-win jobs will changed again to use
• PR: new -pr-*-clwin or -pr-*-clowinjobs will be disabled.

Executing the rescue

• Revert the main release-tools change Switch to use Conda for Windows jobs #1245
•  Revert the changes https://github.com/osrf/osrf-terraform/pull/192
  • Not mandatory, will enable back the vcpkg cold cache