Unclear Whether this orderly shutdown is used by default or only when the orchestrator is started with it is not known.
What
A runner orchestrator is the program that starts Claude Code sessions on machines you host yourself when work comes in. It can now shut down in an orderly way when told to stop:
- A check for new work already under way is allowed to finish.
- Sessions that check brings back are started if there is enough time left, and otherwise handed back so they can be tried again elsewhere.
- Runners kept on standby are dropped when time is short.
- A start-up hook killed by a signal during shutdown is not handed back.
- On exit, it logs a line beginning
[runner:orchestrator] stop summary:with counts of what happened.
Before, the loop simply stopped when cancelled.
Why
Sessions the orchestrator had already claimed are no longer left stuck until their claim runs out, so a restart or redeploy loses less work.
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
Whether this orderly shutdown is used by default or only when the orchestrator is started with it is not known.