Merge pull request #52 from alexef/flags

Add -g flag for grouping metrics together

args.go

@@ -20,16 +20,17 @@ type binArgs struct {
 	Cmd         string   // this is not a command line flag, but rather parsed results
 	CmdArgs     []string // this is not a command line flag, also parsed results
 	LockDir     string   `short:"d" long:"lock-dir" default:"/var/lock" description:"the directory where lock files will be placed"`
-	AllEvents   bool     `short:"e" long:"event" default:"false" description:"emit a start and end datadog event"`
-	FailEvent   bool     `short:"E" long:"event-fail" default:"false" description:"only emit an event on failure"`
-	LogFail     bool     `short:"F" long:"log-fail" default:"false" description:"when a command fails, log its full output (stdout/stderr) to the log directory using the UUID as the filename"`
+	AllEvents   bool     `short:"e" long:"event" description:"emit a start and end datadog event"`
+	FailEvent   bool     `short:"E" long:"event-fail" description:"only emit an event on failure"`
+	LogFail     bool     `short:"F" long:"log-fail" description:"when a command fails, log its full output (stdout/stderr) to the log directory using the UUID as the filename"`
 	EventGroup  string   `short:"G" long:"event-group" value-name:"<group>" description:"emit a cronner_group:<group> tag with Datadog events, does not get sent with statsd metrics"`
-	Lock        bool     `short:"k" long:"lock" default:"false" description:"lock based on label so that multiple commands with the same label can not run concurrently"`
+	Group       string   `short:"g" long:"group" value-name:"<group>" description:"emit a cronner_group:<group> tag with statsd metrics"`
+	Lock        bool     `short:"k" long:"lock" description:"lock based on label so that multiple commands with the same label can not run concurrently"`
 	Label       string   `short:"l" long:"label" description:"name for cron job to be used in statsd emissions and DogStatsd events. alphanumeric only; cronner will lowercase it"`
 	LogPath     string   `long:"log-path" default:"/var/log/cronner" description:"where to place the log files for command output (path for -F/--log-fail output)"`
 	LogLevel    string   `short:"L" long:"log-level" default:"error" description:"set the level at which to log at [none|error|info|debug]"`
 	Namespace   string   `short:"N" long:"namespace" default:"cronner" description:"namespace for statsd emissions, value is prepended to metric name by statsd client"`
-	Sensitive   bool     `short:"s" long:"sensitive" default:"false" description:"specify whether command output may contain sensitive details, this only avoids it being printed to stderr"`
+	Sensitive   bool     `short:"s" long:"sensitive" description:"specify whether command output may contain sensitive details, this only avoids it being printed to stderr"`
 	Version     bool     `short:"V" long:"version" description:"print the version string and exit"`
 	WarnAfter   uint64   `short:"w" long:"warn-after" default:"0" value-name:"N" description:"emit a warning event every N seconds if the job hasn't finished, set to 0 to disable"`
 	WaitSeconds uint64   `short:"W" long:"wait-secs" default:"0" description:"how long to wait for the file lock for"`

args_test.go

@@ -100,6 +100,7 @@ func (t *TestSuite) Test_binArgs_parse(c *C) {
 	c.Check(args.FailEvent, Equals, false)
 	c.Check(args.LogFail, Equals, false)
 	c.Check(args.EventGroup, Equals, "")
+	c.Check(args.Group, Equals, "")
 	c.Check(args.Lock, Equals, false)
 	c.Check(args.LogPath, Equals, "/var/log/cronner")
 	c.Check(args.LogLevel, Equals, "error")
@@ -120,6 +121,7 @@ func (t *TestSuite) Test_binArgs_parse(c *C) {
 		"-E",
 		"-F",
 		"-G", "test_group",
+		"-g", "metric_group",
 		"-k",
 		"-l", "test",
 		"-L", "info",
@@ -143,6 +145,7 @@ func (t *TestSuite) Test_binArgs_parse(c *C) {
 	c.Check(args.FailEvent, Equals, true)
 	c.Check(args.LogFail, Equals, true)
 	c.Check(args.EventGroup, Equals, "test_group")
+	c.Check(args.Group, Equals, "metric_group")
 	c.Check(args.Lock, Equals, true)
 	c.Check(args.Label, Equals, "test")
 	c.Check(args.LogLevel, Equals, "info")
@@ -165,6 +168,7 @@ func (t *TestSuite) Test_binArgs_parse(c *C) {
 		"--event-fail",
 		"--log-fail",
 		"--event-group", "test_group",
+		"--group", "metric_group",
 		"--lock",
 		"--label", "test",
 		"--log-path", "/var/log/testcronner",
@@ -186,6 +190,7 @@ func (t *TestSuite) Test_binArgs_parse(c *C) {
 	c.Check(args.FailEvent, Equals, true)
 	c.Check(args.LogFail, Equals, true)
 	c.Check(args.EventGroup, Equals, "test_group")
+	c.Check(args.Group, Equals, "metric_group")
 	c.Check(args.Lock, Equals, true)
 	c.Check(args.Label, Equals, "test")
 	c.Check(args.LogPath, Equals, "/var/log/testcronner")
@@ -206,6 +211,7 @@ func (t *TestSuite) Test_binArgs_parse(c *C) {
 		Arg0,
 		"--lock-dir=/var/testlock",
 		"--event-group=test_group",
+		"--group=metric_group",
 		"--label=test",
 		"--log-path=/var/log/testcronner",
 		"--log-level=info",
@@ -222,6 +228,7 @@ func (t *TestSuite) Test_binArgs_parse(c *C) {
 	c.Check(len(output), Equals, 0)
 	c.Check(args.LockDir, Equals, "/var/testlock")
 	c.Check(args.EventGroup, Equals, "test_group")
+	c.Check(args.Group, Equals, "metric_group")
 	c.Check(args.Label, Equals, "test")
 	c.Check(args.LogPath, Equals, "/var/log/testcronner")
 	c.Check(args.LogLevel, Equals, "info")

runner.go

@@ -191,8 +191,14 @@ func handleCommand(hndlr *cmdHandler) (int, []byte, float64, error) {
 	}
 
 	// emit the metric for how long it took us and return code
-	hndlr.gs.Timing(fmt.Sprintf("%v.time", hndlr.opts.Label), wallRtMs, nil)
-	hndlr.gs.Gauge(fmt.Sprintf("%v.exit_code", hndlr.opts.Label), float64(ret), nil)
+	tags := []string{}
+
+	if len(hndlr.opts.Group) > 0 {
+		tags = append(tags, fmt.Sprintf("cronner_group:%s", hndlr.opts.Group))
+	}
+
+	hndlr.gs.Timing(fmt.Sprintf("%v.time", hndlr.opts.Label), wallRtMs, tags)
+	hndlr.gs.Gauge(fmt.Sprintf("%v.exit_code", hndlr.opts.Label), float64(ret), tags)
 
 	out := b.Bytes()
 
@@ -278,7 +284,7 @@ func emitEvent(title, body, label, alertType string, hndlr *cmdHandler) {
 	tags := []string{"source_type:cronner", fmt.Sprintf("cronner_label_name:%v", label)}
 
 	if len(hndlr.opts.EventGroup) > 0 {
-		tags = append(tags, fmt.Sprintf("cronner_group:%v", hndlr.opts.EventGroup))
+		tags = append(tags, fmt.Sprintf("cronner_group:%s", hndlr.opts.EventGroup))
 	}
 
 	hndlr.gs.Event(title, body, fields, tags)

runner_test.go

@@ -245,6 +245,52 @@ func (t *TestSuite) Test_handleCommand(c *C) {
 		fmt.Sprintf(`_e{55,77}:Cron testCmd succeeded in %.5f seconds on brainbox01|UUID: %v\nexit code: 0\noutput: somevalue\n|k:%v|s:cronner|t:success|#source_type:cronner,cronner_label_name:testCmd,cronner_group:testgroup`, runTime/1000, t.h.uuid, t.h.uuid),
 	)
 
+	//
+	// Test that DD metrics contain the cronner_group tag
+	//
+
+	// Reset variables used
+	r = nil
+	err = nil
+	runTime = 0
+	match = nil
+
+	t.h.cmd = exec.Command("/bin/echo", "somevalue")
+	t.h.opts.Group = "metricgroup"
+	t.h.opts.EventGroup = ""
+
+	_, r, runTime, err = handleCommand(t.h)
+	c.Assert(err, IsNil)
+
+	stat, ok = <-t.out
+	c.Assert(ok, Equals, true)
+	c.Check(
+		string(stat),
+		Equals,
+		fmt.Sprintf(`_e{35,44}:Cron testCmd starting on brainbox01|UUID: %v\n|k:%v|s:cronner|t:info|#source_type:cronner,cronner_label_name:testCmd`, t.h.uuid, t.h.uuid),
+	)
+
+	stat, ok = <-t.out
+	c.Assert(ok, Equals, true)
+	timeStatTagRegex := regexp.MustCompile("^cronner.testCmd.time:([0-9\\.]+)\\|ms\\|#cronner_group:([a-z]+)$")
+	match = timeStatTagRegex.FindAllStringSubmatch(string(stat), -1)
+	c.Assert(len(match), Equals, 1)
+	c.Assert(len(match[0]), Equals, 3)
+	c.Check(strconv.FormatFloat(runTime, 'f', -1, 64), Equals, match[0][1])
+	c.Check("metricgroup", Equals, match[0][2])
+
+	stat, ok = <-t.out
+	c.Assert(ok, Equals, true)
+	c.Check(string(stat), Equals, "cronner.testCmd.exit_code:0|g|#cronner_group:metricgroup")
+
+	stat, ok = <-t.out
+	c.Assert(ok, Equals, true)
+	c.Check(
+		string(stat),
+		Equals,
+		fmt.Sprintf(`_e{55,77}:Cron testCmd succeeded in %.5f seconds on brainbox01|UUID: %v\nexit code: 0\noutput: somevalue\n|k:%v|s:cronner|t:success|#source_type:cronner,cronner_label_name:testCmd`, runTime/1000, t.h.uuid, t.h.uuid),
+	)
+
 	//
 	// Test that no output is given
 	//
@@ -257,6 +303,7 @@ func (t *TestSuite) Test_handleCommand(c *C) {
 
 	t.h.cmd = exec.Command("/bin/echo", "something")
 	t.h.opts.EventGroup = ""
+	t.h.opts.Group = ""
 
 	t.h.opts.LogFail = false
 	t.h.opts.Lock = true