자식 프로세스 비정상종료시 처리

client/client.go

@@ -272,8 +272,17 @@ func NewClient() (HoustonClient, error) {
 			case exited := <-exitChan:
 				var newprocs []*procmeta
 				for _, proc := range hc.childProcs {
-					if proc.cmd == exited && proc.state != protos.ProcessState_Stopped {
+					if proc.cmd == exited {
-						proc.state = protos.ProcessState_Stopped
+						if proc.state == protos.ProcessState_Running {
+							go func(cmd *exec.Cmd) {
+								if err := cmd.Process.Signal(syscall.SIGTERM); err != nil {
+									cmd.Process.Signal(os.Kill)
+								}
+								cmd.Wait()
+								cmd.Process.Release()
+								logger.Println("abnormal termination of process :", cmd.Args)
+							}(proc.cmd)
+						}
 					} else {
 						newprocs = append(newprocs, proc)
 					}
@@ -383,6 +392,7 @@ func (hc *houstonClient) Start() {
 
 		for _, proc := range hc.childProcs {
 			proc.cmd.Wait()
+			proc.cmd.Process.Release()
 		}
 	}()
 

client/operation.go

@@ -277,9 +277,10 @@ func (hc *houstonClient) launch(meta *procmeta) error {
 					return
 				}
 				logfile.Write(bt.buf[:bt.size])
+				logfile.Sync()
 				meta.buffers.push(bt.buf)
 				thisFileSize += bt.size
-				if thisFileSize > 1024*1024 {
+				if thisFileSize > 10*1024*1024 {
 					switchToNextFile()
 				}
 			}
@@ -414,9 +415,9 @@ func (hc *houstonClient) stopChildProcess(req *shared.StopProcessRequest, op pro
 
 	if len(killing) > 0 {
 		for _, proc := range killing {
+			proc.state = protos.ProcessState_Stopping
 			if err := proc.cmd.Process.Signal(syscall.SIGTERM); err != nil {
 				proc.cmd.Process.Signal(os.Kill)
-				proc.state = protos.ProcessState_Stopping
 			}
 		}
 
@@ -424,6 +425,7 @@ func (hc *houstonClient) stopChildProcess(req *shared.StopProcessRequest, op pro
 
 		for _, proc := range killing {
 			proc.cmd.Wait()
+			proc.cmd.Process.Release()
 		}
 
 		hc.childProcs = remains
@@ -473,6 +475,7 @@ func (hc *houstonClient) restartChildProcess(req *shared.RestartProcessRequest,
 
 	for _, proc := range restarts {
 		proc.cmd.Wait()
+		proc.cmd.Process.Release()
 		proc.state = protos.ProcessState_Stopped
 	}
 	op.Refresh(context.Background(), hc.makeOperationQueryRequest())

houston.sh

@@ -0,0 +1,4 @@
+#!/bin/sh
+
+nohup /home/opdev/houston -client -logfile &
+