logmon: recover from shutting down call locally

client/allocrunner/taskrunner/logmon_hook.go

@@ -6,13 +6,15 @@ import (
 	"fmt"
 	"path/filepath"
 	"runtime"
+	"time"
 
 	hclog "github.com/hashicorp/go-hclog"
 	plugin "github.com/hashicorp/go-plugin"
 	"github.com/hashicorp/nomad/client/allocrunner/interfaces"
 	"github.com/hashicorp/nomad/client/logmon"
 	"github.com/hashicorp/nomad/helper/uuid"
 	"github.com/hashicorp/nomad/nomad/structs"
+	bstructs "github.com/hashicorp/nomad/plugins/base/structs"
 	pstructs "github.com/hashicorp/nomad/plugins/shared/structs"
 )
 
@@ -95,7 +97,37 @@ func reattachConfigFromHookData(data map[string]string) (*plugin.ReattachConfig,
 func (h *logmonHook) Prestart(ctx context.Context,
 	req *interfaces.TaskPrestartRequest, resp *interfaces.TaskPrestartResponse) error {
 
-	// Attempt to reattach to logmon
+	tries := 0
+	for {
+		err := h.prestartOneLoop(ctx, req)
+		if err == bstructs.ErrPluginShutdown {
+			h.logger.Warn("logmon shutdown while making request", "error", err)
+
+			if tries > 3 {
+				return err
+			}
+
+			// retry after killing process and ensure we start a new logmon process
+			tries++
+			h.logmonPluginClient.Kill()
+			time.Sleep(1 * time.Second)
+			continue
+		} else if err != nil {
+			return err
+		}
+
+		rCfg := pstructs.ReattachConfigFromGoPlugin(h.logmonPluginClient.ReattachConfig())
+		jsonCfg, err := json.Marshal(rCfg)
+		if err != nil {
+			return err
+		}
+		resp.State = map[string]string{logmonReattachKey: string(jsonCfg)}
+		return nil
+	}
+}
+
+func (h *logmonHook) prestartOneLoop(ctx context.Context, req *interfaces.TaskPrestartRequest) error {
+	// attach to a running logmon if state indicates one
 	if h.logmonPluginClient == nil {
 		reattachConfig, err := reattachConfigFromHookData(req.PreviousState)
 		if err != nil {
@@ -105,12 +137,13 @@ func (h *logmonHook) Prestart(ctx context.Context,
 		if reattachConfig != nil {
 			if err := h.launchLogMon(reattachConfig); err != nil {
 				h.logger.Warn("failed to reattach to logmon process", "error", err)
+				// if we failed to launch logmon, try again below
 			}
 		}
 
 	}
 
-	// We did not reattach to a plugin and one is still not running.
+	// create a new client in initial starts, failed reattachment, or if we detect exits
 	if h.logmonPluginClient == nil || h.logmonPluginClient.Exited() {
 		if err := h.launchLogMon(nil); err != nil {
 			// Retry errors launching logmon as logmon may have crashed on start and
@@ -134,12 +167,6 @@ func (h *logmonHook) Prestart(ctx context.Context,
 		return err
 	}
 
-	rCfg := pstructs.ReattachConfigFromGoPlugin(h.logmonPluginClient.ReattachConfig())
-	jsonCfg, err := json.Marshal(rCfg)
-	if err != nil {
-		return err
-	}
-	resp.State = map[string]string{logmonReattachKey: string(jsonCfg)}
 	return nil
 }
 

client/allocrunner/taskrunner/logmon_hook_unix_test.go

@@ -10,6 +10,7 @@ import (
 	"os"
 	"syscall"
 	"testing"
+	"time"
 
 	"github.com/hashicorp/nomad/client/allocrunner/interfaces"
 	"github.com/hashicorp/nomad/helper/testlog"
@@ -86,3 +87,76 @@ func TestTaskRunner_LogmonHook_StartCrashStop(t *testing.T) {
 	// Running stop should shutdown logmon
 	require.NoError(t, hook.Stop(context.Background(), nil, nil))
 }
+
+// TestTaskRunner_LogmonHook_ShutdownMidStart simulates logmon crashing while the
+// Nomad client is calling Start() and asserts that we recover and spawn a new logmon.
+func TestTaskRunner_LogmonHook_ShutdownMidStart(t *testing.T) {
+	t.Parallel()
+
+	alloc := mock.BatchAlloc()
+	task := alloc.Job.TaskGroups[0].Tasks[0]
+
+	dir, err := ioutil.TempDir("", "nomadtest")
+	require.NoError(t, err)
+	defer func() {
+		require.NoError(t, os.RemoveAll(dir))
+	}()
+
+	hookConf := newLogMonHookConfig(task.Name, dir)
+	hook := newLogMonHook(hookConf, testlog.HCLogger(t))
+
+	req := interfaces.TaskPrestartRequest{
+		Task: task,
+	}
+	resp := interfaces.TaskPrestartResponse{}
+
+	// First start
+	require.NoError(t, hook.Prestart(context.Background(), &req, &resp))
+	defer hook.Stop(context.Background(), nil, nil)
+
+	origState := resp.State
+	origHookData := resp.State[logmonReattachKey]
+	require.NotEmpty(t, origHookData)
+
+	// Pluck PID out of reattach synthesize a crash
+	reattach := struct {
+		Pid int
+	}{}
+	require.NoError(t, json.Unmarshal([]byte(origHookData), &reattach))
+	pid := reattach.Pid
+	require.NotZero(t, pid)
+
+	proc, _ := os.FindProcess(pid)
+
+	// Assert logmon is running
+	require.NoError(t, proc.Signal(syscall.Signal(0)))
+
+	// SIGSTOP would freeze process without it being considered
+	// exited; so this causes process to be non-exited at beginning of call
+	// then we kill process while Start call is running
+	require.NoError(t, proc.Signal(syscall.SIGSTOP))
+	// sleep for the signal to take effect
+	time.Sleep(1 * time.Second)
+
+	go func() {
+		time.Sleep(2 * time.Second)
+
+		proc.Signal(syscall.SIGCONT)
+		proc.Signal(os.Kill)
+	}()
+
+	req.PreviousState = map[string]string{
+		logmonReattachKey: origHookData,
+	}
+
+	initLogmon, initClient := hook.logmon, hook.logmonPluginClient
+
+	resp = interfaces.TaskPrestartResponse{}
+	err = hook.Prestart(context.Background(), &req, &resp)
+	require.NoError(t, err)
+	require.NotEqual(t, origState, resp.State)
+
+	// assert that we got a new client and logmon
+	require.True(t, initLogmon != hook.logmon)
+	require.True(t, initClient != hook.logmonPluginClient)
+}

client/logmon/client.go

@@ -4,10 +4,14 @@ import (
 	"context"
 
 	"github.com/hashicorp/nomad/client/logmon/proto"
+	"github.com/hashicorp/nomad/helper/pluginutils/grpcutils"
 )
 
 type logmonClient struct {
 	client proto.LogMonClient
+
+	// doneCtx is closed when the plugin exits
+	doneCtx context.Context
 }
 
 func (c *logmonClient) Start(cfg *LogConfig) error {
@@ -21,11 +25,11 @@ func (c *logmonClient) Start(cfg *LogConfig) error {
 		StderrFifo:     cfg.StderrFifo,
 	}
 	_, err := c.client.Start(context.Background(), req)
-	return err
+	return grpcutils.HandleGrpcErr(err, c.doneCtx)
 }
 
 func (c *logmonClient) Stop() error {
 	req := &proto.StopRequest{}
 	_, err := c.client.Stop(context.Background(), req)
-	return err
+	return grpcutils.HandleGrpcErr(err, c.doneCtx)
 }

client/logmon/plugin.go

@@ -73,5 +73,8 @@ func (p *Plugin) GRPCServer(broker *plugin.GRPCBroker, s *grpc.Server) error {
 }
 
 func (p *Plugin) GRPCClient(ctx context.Context, broker *plugin.GRPCBroker, c *grpc.ClientConn) (interface{}, error) {
-	return &logmonClient{client: proto.NewLogMonClient(c)}, nil
+	return &logmonClient{
+		doneCtx: ctx,
+		client:  proto.NewLogMonClient(c),
+	}, nil
 }