fix(instance): (systemd) fix when the instance is not running but the backend still says it's running

This commit is contained in:
2026-04-21 18:00:39 +08:00
parent c978e5c6b4
commit 3e3adf122c
2 changed files with 25 additions and 16 deletions
+8 -3
View File
@@ -635,13 +635,18 @@ func StartInstanceHandler(w http.ResponseWriter, r *http.Request) {
if is.watchdogConnected { if is.watchdogConnected {
if !watchdog.AddInstance(serviceName) { if !watchdog.AddInstance(serviceName) {
if err := IsInstanceRunning(instanceID); err != nil { // Check if instance is really running
SendErrorResponse(w, http.StatusInternalServerError, fmt.Sprintf("Failed to start instance %d: %v", instanceID, err))
postLog.Warning(fmt.Sprintf("[StartInstanceHandler] Instance %d may not be running: %v", instanceID, err))
return
}
postLog.Warning(fmt.Sprintf("[StartInstanceHandler] Failed to add watchdog instance %s", serviceName)) postLog.Warning(fmt.Sprintf("[StartInstanceHandler] Failed to add watchdog instance %s", serviceName))
SendSuccessResponse(w, "Instance started successfully but watchdog instance add failed", nil) SendSuccessResponse(w, "Instance started successfully but watchdog instance add failed", nil)
return return
} }
} }
if err := IsInstanceRunning(instanceID); err != nil { if err := IsInstanceRunning(instanceID); err != nil { // Check if instance is really running
SendErrorResponse(w, http.StatusInternalServerError, fmt.Sprintf("Failed to start instance %d: %v", instanceID, err)) SendErrorResponse(w, http.StatusInternalServerError, fmt.Sprintf("Failed to start instance %d: %v", instanceID, err))
postLog.Warning(fmt.Sprintf("[StartInstanceHandler] Instance %d may not be running: %v", instanceID, err)) postLog.Warning(fmt.Sprintf("[StartInstanceHandler] Instance %d may not be running: %v", instanceID, err))
return return
@@ -915,7 +920,7 @@ func GetInstanceStatusHandler(w http.ResponseWriter, r *http.Request) {
} }
SendSuccessResponse(w, "Instance status retrieved successfully", responseData) SendSuccessResponse(w, "Instance status retrieved successfully", responseData)
postLog.Info(fmt.Sprintf("[GetInstanceStatusHandler] Retrieved status for instance %d (name: %s), isRunning: %v", instanceID, instance.Name, responseData["isRunning"])) // postLog.Debug(fmt.Sprintf("[GetInstanceStatusHandler] Retrieved status for instance %d (name: %s), isRunning: %v", instanceID, instance.Name, responseData["isRunning"]))
} }
func GetInstanceInfoHandler(w http.ResponseWriter, r *http.Request) { func GetInstanceInfoHandler(w http.ResponseWriter, r *http.Request) {
@@ -1010,5 +1015,5 @@ func GetInstanceInfoHandler(w http.ResponseWriter, r *http.Request) {
} }
SendSuccessResponse(w, "Instance info retrieved successfully", responseData) SendSuccessResponse(w, "Instance info retrieved successfully", responseData)
postLog.Info(fmt.Sprintf("[GetInstanceInfoHandler] Retrieved info for instance %d (name: %s), userType: %s", instanceID, instance.Name, userType)) // postLog.Debug(fmt.Sprintf("[GetInstanceInfoHandler] Retrieved info for instance %d (name: %s), userType: %s", instanceID, instance.Name, userType))
} }
+16 -12
View File
@@ -427,33 +427,37 @@ func IsInstanceRunning(instanceID int) error {
} }
case "systemd": case "systemd":
cmd := exec.Command("systemctl", "is-active", serviceName) cmd := exec.Command("systemctl", "status", serviceName)
output, err := cmd.CombinedOutput() output, err := cmd.CombinedOutput()
status := strings.TrimSpace(string(output)) outputStr := string(output)
if err != nil { if err != nil {
exitError, ok := err.(*exec.ExitError) exitError, ok := err.(*exec.ExitError)
if ok { if ok {
exitCode := exitError.ExitCode() exitCode := exitError.ExitCode()
if exitCode == 1 && status == "inactive" { if exitCode == 1 {
postLog.Debug(fmt.Sprintf("[IsInstanceRunning] Systemd service %s is stopped", serviceName)) postLog.Debug(fmt.Sprintf("[IsInstanceRunning] Systemd service status command failed with exit code 1: %s, output: %s", serviceName, outputStr))
return fmt.Errorf("service %s is stopped", serviceName) return fmt.Errorf("service %s is not running", serviceName)
} }
if exitCode == 3 && status == "inactive" { if exitCode == 2 {
postLog.Debug(fmt.Sprintf("[IsInstanceRunning] Systemd service %s is inactive", serviceName)) postLog.Debug(fmt.Sprintf("[IsInstanceRunning] Systemd service status command failed with exit code 2: %s, output: %s", serviceName, outputStr))
return fmt.Errorf("service %s is inactive", serviceName) return fmt.Errorf("service %s is not running", serviceName)
} }
if exitCode == 4 && status == "unknown" { if exitCode == 3 {
postLog.Debug(fmt.Sprintf("[IsInstanceRunning] Systemd service %s does not exist", serviceName)) postLog.Debug(fmt.Sprintf("[IsInstanceRunning] Systemd service %s is not running: %s, output: %s", serviceName, outputStr))
return fmt.Errorf("service %s is not running", serviceName)
}
if exitCode == 4 {
postLog.Debug(fmt.Sprintf("[IsInstanceRunning] Systemd service %s does not exist: %s, output: %s", serviceName, outputStr))
return fmt.Errorf("service %s does not exist", serviceName) return fmt.Errorf("service %s does not exist", serviceName)
} }
} }
postLog.Error(fmt.Sprintf("[IsInstanceRunning] Failed to check systemd service status: %s, output: %s", err, output)) postLog.Error(fmt.Sprintf("[IsInstanceRunning] Failed to check systemd service status: %s, output: %s", err, outputStr))
return err return err
} }
if status == "active" { if strings.Contains(outputStr, "active (running)") {
postLog.Info(fmt.Sprintf("[IsInstanceRunning] Systemd service %s is running", serviceName)) postLog.Info(fmt.Sprintf("[IsInstanceRunning] Systemd service %s is running", serviceName))
return nil return nil
} else { } else {