crane / zot launcher: kill registry serve process on shell exit

[malt3]

Fixes #419

[thesayyn]

I believe the cleanup procedure doesn't need to leak into call site rules. the launcher script can easily keep a global state and expose a cleanup function that kills the process.

[malt3]

[...] the launcher script can easily keep a global state and expose a cleanup function that kills the process
This is not correct. The start_registry function is invoked in a subshell like this:

readonly REGISTRY=$(start_registry "${STORAGE_DIR}" "${STDERR}" "${REGISTRY_PIDFILE}")

The subshell (and any global state that may be preserved inside) does not survive the subshell invocation.

Or to be more precise:

function start_registry() {
    # [...]
    "${REGISTRY_BIN}" >> $output 2>&1 &
    readonly REGISTRY_PID=$!
    # [...]
     return 0
}

function stop_registry() {
    if [[ -z "${REGISTRY_PID+x}" ]]; then
        echo "Registry not started"
        return 0
    fi
    echo "Stopping registry process ${REGISTRY_PID}"
    kill -9 "${REGISTRY_PID}" || true
    return 0
}

# if I do this, the variable will survive but we cannot easily capture the registry address without writing to a file:
start_registry # [...] args
stop_registry

# if I do this, the variable is only part of the subshell
readonly REGISTRY=$(start_registry # [...] args)
stop_registry

Proposed solutions

• What do you think about passing a PIDFILE to start_registry and passing the same PIDFILE to stop_registry? Would look a bit cleaner and works with subshells.
• alternatively, we could have three functions:
  • start_registry (called in same shell as function)
  • stop_registry (called in same shell as function)
  • get_registry (called after start_registry as subshell)

[thesayyn]

Ah true, we can write the pid into storage_dir/proc.pid or call pkill -P $$ kill all subprocesses. I'd rather keep the diff minimal.

[malt3]

I think pkill should not be used since it is not part of POSIX and not even in coreutils or util-linux.
I used the fixed path inside the storage dir, as suggested.

[thesayyn]

diff is still big and there seems to be some formatting changes. Can you revert those?

[thesayyn]

Sorry @malt3 going to look into this soon.