A new requirement came up in relation to #2957. We need a way to discern between a workspace being ready and not. This is also relevant for vscode-coder where we might start a workspace for a user and want to wait for it to become ready.

This requirement has also come up in #5171.

After discussing with @kylecarbs potential implementation, we came up with the following proposal:

• Introduce new property describing workspace agent state, WorkspaceAgentLifecycle: (created,) starting, start_timeout, start_error, ready, (stopping, stop_timeout, stop_error, stopped)
  feat: Add workspace agent lifecycle state reporting #5785
  • This property describes the current state of the agent and is independent of status (which describes the connection status), it can be thought of as a lifecycle status / readyness status
  • And additional created state could be added to indicate an initial state of the agent, before it has reported it's initial state
  • Example usage of state:
    • If state is starting, keep users waiting e.g. during ssh coder.workspace (loading indicator)
    • If state is start_timeout, show users a warning
    • If startup script exits with non-zero status, the start_error state lets us direct users towards inspecting the startup log (Add startup logs to build log UI #2957)
• Introduce a new provider parameter coder_agent.startup_script_timeout with a default value (say 5min) (shutdown_script_timeout for feat: Add support for shutdown_script #5171)
  feat: Add startup_script_timeout and delay_login_until_ready terraform-provider-coder#84
  • This is a soft-timeout after which we warn
  • Show warning in dashboard
    refactor(site): update agent status to include the lifecycle #5835
    fix(site): Hide agent lifecycle unless opted in via delay_login_until_ready #5850
  • Show warning in CLI, e.g. when SSH:ing into the workspace
    NOTE: Not applicable until Final steps
    Partially handled by: feat(cli): Add support for delay_login_until_ready #5851
• Introduce a new provider parameter coder_agent.delay_login_until_ready
  feat: Add startup_script_timeout and delay_login_until_ready terraform-provider-coder#84
  • This will delay logins until agent is ready, will default to false and changed to true in a future release
  • Delay login in web Terminal
  • Delay login in coder ssh
    feat(cli): Add support for delay_login_until_ready #5851
  • Delay login in ssh (coder config-ssh)
    feat(cli): Add support for delay_login_until_ready #5851
  • Add --skip-delay-login-until-ready flag
    feat(cli): Add support for delay_login_until_ready #5851
• Add example(s) with non-blocking startup scripts
  chore(examples): Add login_before_ready and startup_script_timeout #5880
  • Consider /etc/init.d/svc start, nohup /path/to/svc &, "fake systemd", etc.
• Add documentation for troubleshooting agent lifecycles
  docs: Document agent readiness issues (startup script) #5877

Final steps:

• Create a breaking PR:
  • Revert fix(site): Hide agent lifecycle unless opted in via delay_login_until_ready #5850 to show agents with timed out startup scripts by default

We will focus on implementing the start-states first, stop will come later.