Agent restart failed after concurrent update-core runs

[DavidePrincipi]

Steps to reproduce

• On an NS8 cluster worker node, start update-core automatically (scheduler)
• Within ~1 minute, start a second update-core manually on the same node
• Wait for update-core to send shutdown to all rootless module agents
• Observe agent.service restart failures; later, run cluster update-module and observe it aborts because module agents are missing

Expected behavior

Rootless module agents restart correctly after core update, and update-core is serialized (no concurrent runs).

Actual behavior

All rootless module agents stop and remain down until manual intervention (systemctl --user daemon-reload + restart). Cluster update-module aborts with Client module/ was not found.

Relevant excerpt:

Feb 13 03:29:40.109771 <node> agent@node[PID]: Signal "user defined signal 1" caught: shutdown started.
...
Feb 13 03:29:40.432107 <node> systemd[USER_PID]: agent.service: Failed to load environment files: No such file or directory
Feb 13 03:29:40.432127 <node> systemd[USER_PID]: agent.service: Failed to run 'start' task: No such file or directory
...
Feb 13 03:29:41.666918 <node> systemd[USER_PID]: agent.service: Start request repeated too quickly.
Feb 13 03:29:41.666937 <node> systemd[USER_PID]: agent.service: Failed with result 'resources'.

Feb 15 08:05:28 <node> agent@cluster[PID]: agent.tasks.exceptions.TaskSubmissionCheckFailed: Client "module/<module>" was not found
Feb 15 08:05:28 <node> agent@cluster[PID]: task/cluster/<task>: action "update-module" status is "aborted" (1) at step 50update

Components

NS8 core 3.17.1 (from logs: ghcr.io/nethserver/core:3.17.1).
Modules affected: all rootless module agents on the worker (not a single module).

See also

• Discussion https://mattermost.nethesis.it/nethesis/pl/uzuoty49ei8idnrof93u8ddgsa

---

Thanks to @nrauso

[DavidePrincipi]

In Testing, Core 3.20.0-dev.5

Test case 1

• Install Core 3.20.0-dev.5, an update to 3.20.0-dev.6 should be available
• Start core update from CLI with runagent apply-updates
• While the previous command downloads core images start the core update concurrently from the UI

The UI must return an "Action already running" error.

Test case 2

• Start two concurrent module updates in sequence with:

  api-cli run update-module --data '{"module_url":"ghcr.io/nethserver/traefik:4.2.1","instances":["traefik1"],"force":false}'
  

One of them will fail with an "Action already running" error.