fix(observability): avoid false stopped Flynn status outside systemd

src/gateway/handlers/observability.test.ts

@@ -253,4 +253,67 @@ describe('ObservabilityCollector', () => {
     expect(typeof logs.lines[0]?.ts).toBe('number');
     expect(logs.lines[0]?.text).toContain('queue depth high');
   });
+
+  it('falls back to process-running Flynn status when not launched under systemd', async () => {
+    const originalInvocationId = process.env.INVOCATION_ID;
+    const originalJournalStream = process.env.JOURNAL_STREAM;
+    const originalSystemdExecPid = process.env.SYSTEMD_EXEC_PID;
+    delete process.env.INVOCATION_ID;
+    delete process.env.JOURNAL_STREAM;
+    delete process.env.SYSTEMD_EXEC_PID;
+
+    const runner = async (command: string, args: string[]) => {
+      const key = `${command} ${args.join(' ')}`;
+      if (key === 'systemctl show flynn.service --property=LoadState,ActiveState,SubState,Description,ExecMainPID,Result --no-pager') {
+        return {
+          stdout: 'LoadState=loaded\nActiveState=inactive\nSubState=dead\nDescription=Flynn daemon\nExecMainPID=0\nResult=success\n',
+          stderr: '',
+        };
+      }
+      if (key === 'systemctl --user show ollama.service --property=LoadState,ActiveState,SubState,UnitFileState,Description,ExecMainPID,Result --no-pager') {
+        return { stdout: 'LoadState=not-found\nActiveState=inactive\nSubState=dead\nDescription=Ollama\nExecMainPID=0\nResult=not-found\n', stderr: '' };
+      }
+      if (key === 'systemctl --user show llama-server.service --property=LoadState,ActiveState,SubState,UnitFileState,Description,ExecMainPID,Result --no-pager') {
+        return { stdout: 'LoadState=not-found\nActiveState=inactive\nSubState=dead\nDescription=llama.cpp\nExecMainPID=0\nResult=not-found\n', stderr: '' };
+      }
+      if (key === 'docker compose -f docker-compose.yml config --profiles') {
+        return { stdout: '', stderr: '' };
+      }
+      if (key === 'docker compose -f docker-compose.yml config --services') {
+        return { stdout: 'flynn\n', stderr: '' };
+      }
+      if (key === 'docker compose -f docker-compose.yml ps --all --format json') {
+        return { stdout: '[]', stderr: '' };
+      }
+      throw new Error(`Unexpected command: ${key}`);
+    };
+
+    try {
+      const collector = new ObservabilityCollector({
+        config: createConfig(),
+        runner,
+      });
+      await collector.forceSample();
+      const sources = await collector.listSources();
+      const flynn = sources.find((entry) => entry.id === 'systemd:flynn');
+      expect(flynn?.status).toBe('running');
+      expect(flynn?.logCapable).toBe(false);
+    } finally {
+      if (originalInvocationId === undefined) {
+        delete process.env.INVOCATION_ID;
+      } else {
+        process.env.INVOCATION_ID = originalInvocationId;
+      }
+      if (originalJournalStream === undefined) {
+        delete process.env.JOURNAL_STREAM;
+      } else {
+        process.env.JOURNAL_STREAM = originalJournalStream;
+      }
+      if (originalSystemdExecPid === undefined) {
+        delete process.env.SYSTEMD_EXEC_PID;
+      } else {
+        process.env.SYSTEMD_EXEC_PID = originalSystemdExecPid;
+      }
+    }
+  });
 });

src/gateway/handlers/observability.ts

@@ -130,6 +130,10 @@ interface SampleRecord {
   restartCount: number;
 }
 
+function isGatewayRunningUnderSystemd(): boolean {
+  return Boolean(process.env.INVOCATION_ID || process.env.JOURNAL_STREAM || process.env.SYSTEMD_EXEC_PID);
+}
+
 function defaultRunner(
   command: string,
   args: string[],
@@ -665,24 +669,28 @@ export class ObservabilityCollector {
     const snapshots: SourceSnapshot[] = [];
 
     const flynnMapped = mapSystemdStatus(flynnStatus.activeState, flynnStatus.error);
+    const fallbackToProcessRuntime = !isGatewayRunningUnderSystemd() && flynnMapped.status !== 'running';
+    const flynnStatusValue = fallbackToProcessRuntime ? 'running' : flynnMapped.status;
+    const flynnStateCode = fallbackToProcessRuntime ? STATE_RUNNING : flynnMapped.stateCode;
+    const flynnHealthCode = fallbackToProcessRuntime ? HEALTH_HEALTHY : flynnMapped.healthCode;
     snapshots.push({
       source: {
         id: 'systemd:flynn',
         name: 'Flynn daemon',
         kind: 'systemd_system',
         runtime: 'systemd_system',
-        status: flynnMapped.status,
+        status: flynnStatusValue,
         graphCapable: true,
-        logCapable: true,
+        logCapable: !fallbackToProcessRuntime,
         metadata: {
           unit: this.flynnSystemdUnit,
-          state: flynnStatus.activeState,
+          state: fallbackToProcessRuntime ? 'running' : flynnStatus.activeState,
-          statusText: flynnStatus.statusText,
+          statusText: fallbackToProcessRuntime ? 'running (gateway process)' : flynnStatus.statusText,
         },
       },
-      stateCode: flynnMapped.stateCode,
+      stateCode: flynnStateCode,
-      healthCode: flynnMapped.healthCode,
+      healthCode: flynnHealthCode,
-      hasError: Boolean(flynnStatus.error),
+      hasError: fallbackToProcessRuntime ? false : Boolean(flynnStatus.error),
       fingerprint: flynnStatus.pid ? `pid:${flynnStatus.pid}` : null,
     });