Expose container registry for centralized cleanup #25

New issue

Open

opened 2025-11-19 20:50:20 +00:00 by glenux · 0 comments

glenux commented

2025-11-19 20:50:20 +00:00

Owner

Summary

Provide a shared registry that maps logical container names/engine pairs to running processes so cleanup can be coordinated.
Allow ProcessManager and runners to sweep the registry on shutdown to remove dangling docker/podman workloads.

Current behavior

ProcessManager in src/process_manager.cr only tracks OS process IDs and has no awareness of the container that each subprocess created. When a docker/podman child dies unexpectedly or the CLI exits before the child is gone, there is no metadata to call engine-specific cleanup helpers, so containers continue running until the operator intervenes.

Desired behavior

Introduce APIs to register and unregister container metadata (name, engine type) alongside tracked processes.
Extend the at_exit handler to iterate over registry entries and call the appropriate engine cleanup helpers so containers are killed/removed reliably.
Update container/build runners to register their containers and rely on the registry during normal exits, signal handling, and crashes.

Technical notes

Store the registry inside ProcessManager or a new helper under DocMachine::Container, and guard it with a mutex so concurrent runners stay safe.
Entries should tolerate engines that already shut down and log when cleanup isn't needed.
This work implements AUDIT-2025-014 follow-up actions and enables future features that need centralized cleanup orchestration.

Local task: 11aa04

## Summary - Provide a shared registry that maps logical container names/engine pairs to running processes so cleanup can be coordinated. - Allow ProcessManager and runners to sweep the registry on shutdown to remove dangling docker/podman workloads. ## Current behavior ProcessManager in `src/process_manager.cr` only tracks OS process IDs and has no awareness of the container that each subprocess created. When a docker/podman child dies unexpectedly or the CLI exits before the child is gone, there is no metadata to call engine-specific cleanup helpers, so containers continue running until the operator intervenes. ## Desired behavior - Introduce APIs to register and unregister container metadata (name, engine type) alongside tracked processes. - Extend the `at_exit` handler to iterate over registry entries and call the appropriate engine cleanup helpers so containers are killed/removed reliably. - Update container/build runners to register their containers and rely on the registry during normal exits, signal handling, and crashes. ## Technical notes - Store the registry inside ProcessManager or a new helper under `DocMachine::Container`, and guard it with a mutex so concurrent runners stay safe. - Entries should tolerate engines that already shut down and log when cleanup isn't needed. - This work implements AUDIT-2025-014 follow-up actions and enables future features that need centralized cleanup orchestration. Local task: 11aa04