WAT

Helps running azure (multiple) azure self-hosted agents on one machine (scoped).

Features

One-line command to set up n-agents (including indivdual os users per agent, systemd service). Downloads, setups and keeps them running.
Set's up a user and home folder for each agent automatically.
Supports running docker based jobs.
Uses --once mode to run the agents with automatic workdir cleanup and restart after job-end
offers batch-uninstalling agents
Helps setting up MTU vars

This is not

an orchestrator for k8s
this does not use docker-in-docker (though it can be used to run jobs in docker containers)

Details

The run-once mode is based on Microsoft's ./run.sh --once which ensures that an agents only runs 1 job and then stops. This is used to

cleanup the workdir in a safe manner after each job (ensuring no other job is scheduled yet)
ensures each job on an agent runs in a clean workdir
starts the agent right after cleanup up (few seconds) to be available for the next job
Use the original microsoft tools, binaries and all the bits. Be agent-upgrade ready.

This fixes issues like

microsoft/azure-pipelines-agent#1895, microsoft/azure-pipelines-agent#708, microsoft/azure-pipelines-agent#1506 and microsoft/azure-pipelines-yaml#453
Security: A non-related, following up job has access to data of a prior job (leak)
Jobs artifact downloads getting multiplied due to left-overs of prior jobs (pollution)

Requirements

For examples, sets up and starts one agent called agent0 for the pool Default enabling the run-once mode

./agent-setup agent0 <PAT> Default 1

You can also add an additional agent, with an MTU (for the docker network) of 1400

./agent-setup agent1 <PAT> Default 1 1400

X agents, creates 15 agents (agent0....agent14) in the pool Default with an MTU of 1400

./batch-setup.sh 0 15 <PAT> Default 1 1400

The below commands does uninstall

So there should be nothing left after running this for one agent.

./agent-uninstall agent0 <PAT>

Uninstalls agent 0..14 from the Default pool

./batch-uninstall.sh 0 15 <PAT> Default

Bug in the agent, see microsoft/azure-pipelines-agent#4215 workaround by blocking the network request

sudo ip route add blackhole 169.254.169.254

This sends the stalling requests into a direct-drop gw and speeds up the start by 100s.

Anytime, just open PRs. Happy to extend whatever we have here.