Objective
Recover a Docker container stuck in a restart loop without hiding the root cause by inspecting configuration, logs, exit codes and OOM events, validating dependencies, controlling the restart policy, and relaunching deliberately.
Prerequisites
- Root or sudo access and the service, filesystem, container or timer name.
- A backup or configuration copy when the procedure changes persistent service state.
- Administrative access appropriate to the system being changed or diagnosed.
- A clearly identified scope: affected users, systems, addresses, services and the time of the observed problem.
- A maintenance or test window when the procedure can affect production traffic or availability.
- A copy of the current configuration or other recovery material before any irreversible action.
Step-by-step procedure
Establish the baseline and scope
Before changing anything, reproduce the issue or document the requested change on a representative system. Record the affected users or services, exact time, current configuration, recent changes and a known-good comparison point. This baseline is the reference used to decide whether each later step improves the situation.
- The scope and current state are documented well enough to reproduce or verify the procedure.
Inspecting configuration
Examine this diagnostic area: inspecting configuration. Use the narrowest reproducible test, compare the observed state with the documented or expected state, record timestamps and exact errors, and correlate the result with adjacent dependencies. Do not change several settings at once: the purpose of this step is to establish whether this area is healthy, misconfigured, unavailable, or simply not involved in the failure.
- Evidence for this area is explicit, reproducible and consistent with the intended design.
Logs
Examine this diagnostic area: logs. Use the narrowest reproducible test, compare the observed state with the documented or expected state, record timestamps and exact errors, and correlate the result with adjacent dependencies. Do not change several settings at once: the purpose of this step is to establish whether this area is healthy, misconfigured, unavailable, or simply not involved in the failure.
- Evidence for this area is explicit, reproducible and consistent with the intended design.
Exit codes
Examine this diagnostic area: exit codes. Use the narrowest reproducible test, compare the observed state with the documented or expected state, record timestamps and exact errors, and correlate the result with adjacent dependencies. Do not change several settings at once: the purpose of this step is to establish whether this area is healthy, misconfigured, unavailable, or simply not involved in the failure.
- Evidence for this area is explicit, reproducible and consistent with the intended design.
OOM events
Examine this diagnostic area: OOM events. Use the narrowest reproducible test, compare the observed state with the documented or expected state, record timestamps and exact errors, and correlate the result with adjacent dependencies. Do not change several settings at once: the purpose of this step is to establish whether this area is healthy, misconfigured, unavailable, or simply not involved in the failure.
- Evidence for this area is explicit, reproducible and consistent with the intended design.
Dependencies
Examine this diagnostic area: dependencies. Use the narrowest reproducible test, compare the observed state with the documented or expected state, record timestamps and exact errors, and correlate the result with adjacent dependencies. Do not change several settings at once: the purpose of this step is to establish whether this area is healthy, misconfigured, unavailable, or simply not involved in the failure.
- Evidence for this area is explicit, reproducible and consistent with the intended design.
Controlling the restart policy
Examine this diagnostic area: controlling the restart policy. Use the narrowest reproducible test, compare the observed state with the documented or expected state, record timestamps and exact errors, and correlate the result with adjacent dependencies. Do not change several settings at once: the purpose of this step is to establish whether this area is healthy, misconfigured, unavailable, or simply not involved in the failure.
- Evidence for this area is explicit, reproducible and consistent with the intended design.
Relaunching deliberately
Examine this diagnostic area: relaunching deliberately. Use the narrowest reproducible test, compare the observed state with the documented or expected state, record timestamps and exact errors, and correlate the result with adjacent dependencies. Do not change several settings at once: the purpose of this step is to establish whether this area is healthy, misconfigured, unavailable, or simply not involved in the failure.
- Evidence for this area is explicit, reproducible and consistent with the intended design.
Validate the complete service
Repeat the original user, system or application workflow from the real source and verify the complete result, not only one command or one local check. Confirm that logs and monitoring show the expected behavior and that no temporary debug, bypass, test account, rule or maintenance setting remains enabled.
- The end-to-end service works or the remaining failure is isolated to a clearly identified component.
Technical commands from the original procedure
These technical blocks are preserved byte-for-byte from the historical procedure and kept in their original order. Review names, addresses, paths and parameters before use.
docker ps -a --no-truncdocker inspect <CONTAINER> --format '{{json .State}}'docker inspect <CONTAINER> --format '{{json .HostConfig.RestartPolicy}}'docker logs --timestamps --tail 200 <CONTAINER>docker logs --since 30m <CONTAINER>docker update --restart=no <CONTAINER>docker stop <CONTAINER>docker inspect <CONTAINER> --format 'OOM={{.State.OOMKilled}} Exit={{.State.ExitCode}} Error={{.State.Error}}'dmesg -T | grep -i -E 'oom|killed process' | tail -30df -hdocker inspect <CONTAINER> --format '{{json .Mounts}}'docker inspect <CONTAINER> --format '{{json .Config.Env}}'getent hosts <DEPENDENCY>nc -vz <DEPENDENCY> <PORT>docker start <CONTAINER>docker logs -f <CONTAINER>docker update --restart=unless-stopped <CONTAINER>docker inspect <CONTAINER> --format 'Restarts={{.RestartCount}} Status={{.State.Status}}'Validation
The procedure is validated when:
- The original symptom or change request has been tested end to end.
- The effective configuration matches the intended design and no unexplained error remains in the relevant logs.
- Temporary troubleshooting controls have been removed and monitoring remains normal.
- The result, evidence and any follow-up action are documented.
Rollback
- Restore the configuration, policy, binding, route, credential assignment or service state recorded in the baseline when the change does not meet its success criteria.
- Remove temporary rules, test objects and diagnostic settings that were introduced only for the procedure.
- After rollback, repeat the minimum health checks to confirm that the previous service level has been restored.
Troubleshooting / common errors
- A deleted file can still consume disk space while held open by a process.
- Systemd and container restart policies can hide the first meaningful failure; inspect the earliest relevant log entry.
- If the result changes between tests, compare source, destination, identity, time and policy context before changing additional settings.
- If a command succeeds but the application still fails, continue at the next protocol or application layer instead of widening access.
- If the expected evidence is missing, verify that logging, auditing and the test path actually cover the failing component.
- If the change does not improve the measured symptom, restore the previous state and reassess the working hypothesis.