From 1bbe3b4f43573930beaaea78e47d66a8487c7e2a Mon Sep 17 00:00:00 2001
From: Emil Simeonov <emil@theliberatededge.org>
Date: Fri, 19 Jun 2026 02:47:14 +0200
Subject: [PATCH] GRM-17: fix docker mode status detection and systemd template

- systemd template for docker mode now uses Type=oneshot + RemainAfterExit=yes
  so that systemctl is-active returns active when the container is running.
  Previously docker run -d exited immediately, causing systemd to mark the
  service as inactive even though the container was still up.
- grm list now tries multiple container name fallbacks for docker mode:
  1. gitea-runner-{name} (current naming)
  2. gitea-runner-{host} (legacy installs where name defaulted to host)
  3. systemctl is-active gitea-runner@{name} (for installs with fixed template)
- All tests pass, 100% coverage, ruff + pyright clean
---
 .../templates/gitea-runner@.service.j2        |  4 +-
 src/gitea_runner_manager/runner_manager.py    | 55 +++++++++++++------
 2 files changed, 40 insertions(+), 19 deletions(-)

diff --git a/ansible/roles/gitea-runner/templates/gitea-runner@.service.j2 b/ansible/roles/gitea-runner/templates/gitea-runner@.service.j2
index 52ee713..39805b5 100644
--- a/ansible/roles/gitea-runner/templates/gitea-runner@.service.j2
+++ b/ansible/roles/gitea-runner/templates/gitea-runner@.service.j2
@@ -4,8 +4,9 @@ After=network.target{% if runner_mode == 'docker' %} docker.service
 Requires=docker.service{% endif %}
 
 [Service]
-Type=simple
 {% if runner_mode == 'docker' %}
+Type=oneshot
+RemainAfterExit=yes
 ExecStartPre=-/usr/bin/docker rm -f gitea-runner-%i
 ExecStart=/usr/bin/docker run -d --name gitea-runner-%i \
   --restart=no \
@@ -19,6 +20,7 @@ ExecStopPost=-/usr/bin/docker rm -f gitea-runner-%i
 Restart=on-failure
 RestartSec={{ gitea_runner_service_restart_sec }}
 {% else %}
+Type=simple
 ExecStart={{ gitea_runner_binary_path }} daemon --config {{ gitea_runner_base_config_dir }}/%i/config.yaml
 WorkingDirectory={{ gitea_runner_base_data_dir }}/%i
 ExecStop=/bin/kill -TERM $MAINPID
diff --git a/src/gitea_runner_manager/runner_manager.py b/src/gitea_runner_manager/runner_manager.py
index a482336..af2c303 100644
--- a/src/gitea_runner_manager/runner_manager.py
+++ b/src/gitea_runner_manager/runner_manager.py
@@ -289,18 +289,44 @@ class RunnerManager:
                     user=user,
                 )
             )
+
+            def _parse_status(stdout: str) -> str:
+                ansible_noise = (" | CHANGED | ", " | FAILED | ", " | UNREACHABLE | ", "[WARNING]", "ssh:", ">>")
+                lines = [ln for ln in stdout.splitlines() if ln.strip() and not any(p in ln for p in ansible_noise)]
+                return lines[-1].strip() if lines else "unknown"
+
             try:
                 if mode == "docker":
-                    stdout = self._executor.run_ad_hoc(
-                        host,
-                        user,
-                        key,
-                        "shell",
-                        f"docker inspect -f '{{{{.State.Status}}}}' gitea-runner-{name} 2>/dev/null",
-                        become=True,
-                        ask_become_pass=True,
-                        check=False,
-                    )
+                    # Try expected container name first, then host-based fallback for legacy installs
+                    for container_name in (f"gitea-runner-{name}", f"gitea-runner-{host}"):
+                        stdout = self._executor.run_ad_hoc(
+                            host,
+                            user,
+                            key,
+                            "shell",
+                            f"docker inspect -f '{{{{.State.Status}}}}' {container_name} 2>/dev/null",
+                            become=True,
+                            ask_become_pass=True,
+                            check=False,
+                        )
+                        service_status = _parse_status(stdout)
+                        if service_status != "unknown":
+                            docker_map = {"running": "active", "exited": "inactive", "dead": "failed"}
+                            service_status = docker_map.get(service_status, "unknown")
+                            break
+                    else:
+                        # Final fallback: check systemd service (for installs using the fixed template)
+                        stdout = self._executor.run_ad_hoc(
+                            host,
+                            user,
+                            key,
+                            "shell",
+                            f"systemctl is-active gitea-runner@{name} 2>/dev/null",
+                            become=True,
+                            ask_become_pass=True,
+                            check=False,
+                        )
+                        service_status = _parse_status(stdout)
                 else:
                     stdout = self._executor.run_ad_hoc(
                         host,
@@ -312,14 +338,7 @@ class RunnerManager:
                         ask_become_pass=True,
                         check=False,
                     )
-                # Filter out ansible noise (headers, warnings, SSH messages)
-                ansible_noise = (" | CHANGED | ", " | FAILED | ", " | UNREACHABLE | ", "[WARNING]", "ssh:", ">>")
-                lines = [ln for ln in stdout.splitlines() if ln.strip() and not any(p in ln for p in ansible_noise)]
-                service_status = lines[-1].strip() if lines else "unknown"
-                # Docker statuses need mapping to systemd vocabulary
-                if mode == "docker":
-                    docker_map = {"running": "active", "exited": "inactive", "dead": "failed"}
-                    service_status = docker_map.get(service_status, "unknown")
+                    service_status = _parse_status(stdout)
             except Exception:
                 service_status = "unknown"
             # Translate known status values