Module execution_recovery 

Launch-time recovery for killed detached docker sessions (issue #176).

$ --isolated docker --detached --on-kill-resume 3 --recovery-command B -- A runs A; when the main process is killed (exit 137, or OOMKilled with no exit status of its own), the detached completion watcher hands the inspected facts to this module, which restarts the same container so B continues on the same filesystem, with the same resource limits (they live in the container’s HostConfig), under the same execution UUID and appending to the same log file.

How B replaces A inside the same container: a container launched with a recovery command runs a tiny selector as its command. The selector runs A normally, but runs B when the marker file exists. Before docker start, the marker (holding the attempt number) is copied into the stopped container with docker cp. Without a recovery command the original command is simply started again.

Like detached_finalize, the entry point runs as a short-lived child of the watcher on a host whose CLI invocation is long gone, so it never panics on bad input and never prints anything.

Structs

RecoveryDelayHooks

Where the --on-kill-resume-delay wait gets its randomness and sleeps; tests swap both out (issue #181).

RecoveryFacts

The docker facts the watcher hands over.

RecoveryOutcome

Outcome of a recovery attempt; recovered is true only when the container was started again.

RecoverySeparator

What the [Recovery k/N] separator reports.

Constants

INTERNAL_RECOVER_FLAG

Hidden argument the detached watcher re-invokes this binary with.

KILLED_EXIT_CODE

Exit code of a process killed with SIGKILL (the OOM killer’s signal).

RECOVERY_ATTEMPT_ENV

Environment variable carrying the attempt number into the recovery command.

RECOVERY_MARKER_PATH

File whose presence makes the selector run the recovery command.

Functions

build_recovery_selector_args

Wrap the container command so a later docker start can switch to the recovery command.

build_recovery_snippet

Shell condition run by the completion watcher: true when the container was killed and the recovery entry point resumed it. Mirrors is_killed_exit: OOMKilled only counts without a non-negative exit code (issue #178).

format_recovery_separator

The [Recovery k/N] separator written into the session log.

is_killed_exit

Whether the inspected facts describe a killed main process.

recover_killed_execution

Resume a killed execution in its own container.

recover_killed_execution_with_delay

recover_killed_execution after the random --on-kill-resume-delay wait (issue #181), with injectable randomness and sleep.

recovery_selector

POSIX selector run as the container command when a recovery command is set: sh -c SELECTOR start-command <shell> <flag> <recovery> <main argv...>.

run_internal_recover

Entry point for --internal-recover-detached-docker <uuid> <exit> <oom> <started> <finished> <error> <cgroup>. Returns 0 only when the container was resumed; any other outcome lets the watcher continue with its normal cleanup, footer and finalization.

start_system_watcher

The real watcher: a detached shell following the resumed container.

Type Aliases

WatcherStarter

Starts the completion watcher that follows a resumed container.