fix: update Semaphore port to 3000 and enhance database backup process

update_semaphore.yml

@@ -15,14 +15,12 @@
     semaphore_project: semaphore
     semaphore_container: semaphore
     semaphore_compose_filename: "docker-compose-semaphore.yml"
-    # Published port on the VM. NOT 3000 — that only exists in the container.
-    semaphore_port: 3008
+    semaphore_port: 3000
 
     semaphore_update_log: "{{ compose_remote_base }}/semaphore-update.log"
+    semaphore_update_delay: 20
 
   tasks:
-    # This job gets killed when the container is recreated, so the final
-    # verification lives in the log and is reported on the next run.
     - name: Show result of the previous update
       ansible.builtin.command:
         argv: [tail, -n, "12", "{{ semaphore_update_log }}"]
@@ -34,9 +32,6 @@
       ansible.builtin.debug:
         var: prev_log.stdout_lines
 
-    # Env vars override /etc/semaphore/config.json. A stray SEMAPHORE_DB_*
-    # points Semaphore at a different database; a stray
-    # SEMAPHORE_ACCESS_KEY_ENCRYPTION makes every credential undecryptable.
     - name: Refuse compose that overrides DB or encryption config
       ansible.builtin.assert:
         that:
@@ -87,8 +82,6 @@
         dest: "{{ compose_remote_dir }}"
         remote_src: true
 
-    # Online snapshot via the sqlite backup API — consistent without stopping
-    # the server. Reads the DB path from the live config instead of guessing.
     - name: Back up Semaphore database
       ansible.builtin.command:
         argv:
@@ -102,7 +95,7 @@
             cfg = json.load(open('/etc/semaphore/config.json'))
             out = '/etc/semaphore/backups'
             os.makedirs(out, exist_ok=True)
-            ts = datetime.datetime.utcnow().strftime('%Y%m%d-%H%M%S')
+            ts = datetime.datetime.now(datetime.timezone.utc).strftime('%Y%m%d-%H%M%S')
             dst = f'{out}/database-{ts}.sqlite'
             a = sqlite3.connect('file:' + cfg['sqlite']['host'] + '?mode=ro', uri=True)
             b = sqlite3.connect(dst)
@@ -113,6 +106,10 @@
             print(dst, os.path.getsize(dst), 'bytes')
             for old in sorted(glob.glob(f'{out}/database-*.sqlite'), reverse=True)[10:]:
                 os.remove(old)
+                stale = f"{out}/config-{os.path.basename(old)[9:-7]}.json"
+                if os.path.exists(stale):
+                    os.remove(stale)
+                print('pruned', old)
       register: sem_db_backup
       changed_when: true
 
@@ -120,15 +117,6 @@
       ansible.builtin.debug:
         var: sem_db_backup.stdout_lines
 
-    - name: Pull latest Semaphore image
-      community.docker.docker_compose_v2:
-        project_name: "{{ semaphore_project }}"
-        project_src: "{{ compose_remote_dir }}"
-        files:
-          - "{{ semaphore_compose_filename }}"
-        pull: always
-
-
     - name: Recreate Semaphore in the background
       ansible.builtin.shell: |
         cat > "{{ compose_remote_base }}/semaphore-selfupdate.sh" <<'SH'
@@ -137,17 +125,27 @@
         exec 2>&1
         cd "{{ compose_remote_dir }}" || exit 1
         COMPOSE=(docker compose -p "{{ semaphore_project }}" -f "{{ semaphore_compose_filename }}")
+
+        # Let the calling job finish and its status get written to the DB
+        # before we potentially replace the container running it.
+        sleep {{ semaphore_update_delay }}
+
         echo "=== $(date -Is) update starting ==="
         "${COMPOSE[@]}" pull || { echo "FAIL: pull failed" >&2; exit 1; }
+
+        # No --force-recreate: compose replaces the container only when the
+        # image or the resolved config (including env_file) actually changed,
+        # so unchanged runs cause no downtime.
         "${COMPOSE[@]}" up -d --remove-orphans --wait --wait-timeout 300
+
         for _ in $(seq 1 60); do
           code=$(curl -s -o /dev/null -w '%{http_code}' \
             "http://127.0.0.1:{{ semaphore_port }}/api/ping" || true)
           if [ "$code" = "200" ]; then
             echo "image: $(docker inspect "{{ semaphore_container }}" \
               --format '{{ '{{' }}.Config.Image{{ '}}' }}')"
-            # --tail 40 misses the startup line when the container was not
-            # recreated; grep the whole log instead
+            # Grep the whole log: when the container was not replaced, the
+            # startup line is far outside the last few lines.
             docker logs "{{ semaphore_container }}" 2>&1 \
               | grep -m1 'SQLite @' || echo "WARN: no 'SQLite @' line"
             echo "OK: semaphore healthy at $(date -Is)"
@@ -169,6 +167,6 @@
     - name: What happens next
       ansible.builtin.debug:
         msg:
-          - "Update launched detached. If the image changed, this job dies here."
-          - "That is expected — the log is the success signal, not the task status."
-          - "tail -n 20 {{ semaphore_update_log }}"
+          - "Update starts in {{ semaphore_update_delay }}s, after this job ends."
+          - "If the image changed, the container is replaced — expected."
+          - "The log is the success signal: tail -n 20 {{ semaphore_update_log }}"