codewind-jenkins-agents.md

Jenkins agents in Eclipse Codewind builds

Intro to Agents

An agent is a slave machine that is used to run a pipeline or a stage of a jenkins build. Only one build can be using a given agent at a given time.

Agents can be top-level ('pipeline' level) but can be further broken down by stages. So, one build can use multiple agents for its different stages. For VS Code, the top-level agent is a kube one from a node image (Jenkinsfile#57). This agent is used for any stage which does not override the agent clause.

If you have to run docker commands, use the docker-build agent for that stage. For VS Code and Eclipse, this is only the test stage (Jenkinsfile#134).

3 agent types

1. Kube agents
   • Described using regular kubernetes yaml (Jenkinsfile#58).
     • Pick any public docker image (you could build your own)
     • Use whatever user the dockerfile specifies
   • Basically unlimited resources to run these through Eclipse infrastructure
   • You can spin off as many as you want as part of your build
     • But the pods will share a filesystem, for some reason - watch out for this especially when using parallel stages.
     • In VS Code the parallel stages execute in two different copies of the code to work around this (Jenkinsfile#156)
   • Can't run docker commands (docker in docker too messy)
2. docker-build agents
   • These are our 4 VMs, so limited resources (see https://ci.eclipse.org/codewind/computer/)
   • Just regular VMs, not run in docker, so they can run docker commands
   • Use genie.codewind user who is part of docker group but not part of sudoers
   • We have ssh access and root password to all machines except s8p2d-ubuntu1804 which is managed by eclipse team so we have to ask them to install stuff
3. Basic agent
   • Used if agent clause is agent any
   • Used by the 'Upload to download.eclipse.org' stages (Jenkinsfile#200)

Important notes

• Since the kube agent is unlimited, use the kube agent as much as possible to take load off our VMs.
• When switching between agents, the filesystem is not preserved. Use stashes to persist changes between stages. (Jenkinsfile#115)