Merge pull request #424 from hashicorp/f-alloc-status-cli

alloc-status displays individual task state

command/alloc_status.go

@@ -2,7 +2,11 @@ package command
 
 import (
 	"fmt"
+	"sort"
 	"strings"
+	"time"
+
+	"github.com/hashicorp/nomad/api"
 )
 
 type AllocStatusCommand struct {
@@ -13,14 +17,21 @@ func (c *AllocStatusCommand) Help() string {
 	helpText := `
 Usage: nomad alloc-status [options] <allocation>
 
-  Display information about existing allocations. This command can
-  be used to inspect the current status of all allocation,
-  including its running status, metadata, and verbose failure
-  messages reported by internal subsystems.
+  Display information about existing allocations and its tasks. This command can
+  be used to inspect the current status of all allocation, including its running
+  status, metadata, and verbose failure messages reported by internal
+  subsystems.
 
 General Options:
 
-  ` + generalOptionsUsage()
+  ` + generalOptionsUsage() + `
+
+Alloc Status Options:
+
+  -short
+    Display short output. Shows only the most recent task event.
+`
+
 	return strings.TrimSpace(helpText)
 }
 
@@ -29,8 +40,12 @@ func (c *AllocStatusCommand) Synopsis() string {
 }
 
 func (c *AllocStatusCommand) Run(args []string) int {
+	var short bool
+
 	flags := c.Meta.FlagSet("alloc-status", FlagSetClient)
 	flags.Usage = func() { c.Ui.Output(c.Help()) }
+	flags.BoolVar(&short, "short", false, "")
+
 	if err := flags.Parse(args); err != nil {
 		return 1
 	}
@@ -65,7 +80,6 @@ func (c *AllocStatusCommand) Run(args []string) int {
 		fmt.Sprintf("NodeID|%s", alloc.NodeID),
 		fmt.Sprintf("JobID|%s", alloc.JobID),
 		fmt.Sprintf("ClientStatus|%s", alloc.ClientStatus),
-		fmt.Sprintf("ClientDescription|%s", alloc.ClientDescription),
 		fmt.Sprintf("NodesEvaluated|%d", alloc.Metrics.NodesEvaluated),
 		fmt.Sprintf("NodesFiltered|%d", alloc.Metrics.NodesFiltered),
 		fmt.Sprintf("NodesExhausted|%d", alloc.Metrics.NodesExhausted),
@@ -74,9 +88,108 @@ func (c *AllocStatusCommand) Run(args []string) int {
 	}
 	c.Ui.Output(formatKV(basic))
 
+	// Print the state of each task.
+	if short {
+		c.shortTaskStatus(alloc)
+	} else {
+		c.taskStatus(alloc)
+	}
+
 	// Format the detailed status
 	c.Ui.Output("\n==> Status")
 	dumpAllocStatus(c.Ui, alloc)
 
 	return 0
 }
+
+// shortTaskStatus prints out the current state of each task.
+func (c *AllocStatusCommand) shortTaskStatus(alloc *api.Allocation) {
+	tasks := make([]string, 0, len(alloc.TaskStates)+1)
+	tasks = append(tasks, "Name|State|LastEvent|Time")
+	for task := range c.sortedTaskStateIterator(alloc.TaskStates) {
+		fmt.Println(task)
+		state := alloc.TaskStates[task]
+		lastState := state.State
+		var lastEvent, lastTime string
+
+		l := len(state.Events)
+		if l != 0 {
+			last := state.Events[l-1]
+			lastEvent = last.Type
+			lastTime = c.formatUnixNanoTime(last.Time)
+		}
+
+		tasks = append(tasks, fmt.Sprintf("%s|%s|%s|%s",
+			task, lastState, lastEvent, lastTime))
+	}
+
+	c.Ui.Output("\n==> Tasks")
+	c.Ui.Output(formatList(tasks))
+}
+
+// taskStatus prints out the most recent events for each task.
+func (c *AllocStatusCommand) taskStatus(alloc *api.Allocation) {
+	for task := range c.sortedTaskStateIterator(alloc.TaskStates) {
+		state := alloc.TaskStates[task]
+		events := make([]string, len(state.Events)+1)
+		events[0] = "Time|Type|Description"
+
+		size := len(state.Events)
+		for i, event := range state.Events {
+			formatedTime := c.formatUnixNanoTime(event.Time)
+
+			// Build up the description based on the event type.
+			var desc string
+			switch event.Type {
+			case api.TaskDriverFailure:
+				desc = event.DriverError
+			case api.TaskKilled:
+				desc = event.KillError
+			case api.TaskTerminated:
+				var parts []string
+				parts = append(parts, fmt.Sprintf("Exit Code: %d", event.ExitCode))
+
+				if event.Signal != 0 {
+					parts = append(parts, fmt.Sprintf("Signal: %d", event.Signal))
+				}
+
+				if event.Message != "" {
+					parts = append(parts, fmt.Sprintf("Exit Message: %q", event.Message))
+				}
+				desc = strings.Join(parts, ", ")
+			}
+
+			// Reverse order so we are sorted by time
+			events[size-i] = fmt.Sprintf("%s|%s|%s", formatedTime, event.Type, desc)
+		}
+
+		c.Ui.Output(fmt.Sprintf("\n==> Task %q is %q\nRecent Events:", task, state.State))
+		c.Ui.Output(formatList(events))
+	}
+}
+
+// formatUnixNanoTime is a helper for formating time for output.
+func (c *AllocStatusCommand) formatUnixNanoTime(nano int64) string {
+	t := time.Unix(0, nano)
+	return t.Format("15:04:05 01/02/06")
+}
+
+// sortedTaskStateIterator is a helper that takes the task state map and returns a
+// channel that returns the keys in a sorted order.
+func (c *AllocStatusCommand) sortedTaskStateIterator(m map[string]*api.TaskState) <-chan string {
+	output := make(chan string, len(m))
+	keys := make([]string, len(m))
+	i := 0
+	for k := range m {
+		keys[i] = k
+		i++
+	}
+	sort.Strings(keys)
+
+	for _, key := range keys {
+		output <- key
+	}
+
+	close(output)
+	return output
+}

website/source/docs/commands/alloc-status.html.md.erb

@@ -3,14 +3,15 @@ layout: "docs"
 page_title: "Commands: alloc-status"
 sidebar_current: "docs-commands-alloc-status"
 description: >
-  Display status and metadata about existing allocations
+  Display status and metadata about existing allocations and their tasks.
 ---
 
 # Command: alloc-status
 
-The `alloc-status` command displays status information and metadata about
-an existing allocation. It can be useful while debugging to reveal the
-underlying reasons for scheduling decisions or failures.
+The `alloc-status` command displays status information and metadata about an
+existing allocation and its tasks. It can be useful while debugging to reveal
+the underlying reasons for scheduling decisions or failures, as well as the
+current state of its tasks.
 
 ## Usage
 
@@ -24,25 +25,71 @@ and detailed information for it will be dumped.
 ## General Options
 
 <%= general_options_usage %>
+#
+## Status Options
+
+* `-short`: Display short output. Shows only the most recent task event.
 
 ## Examples
 
+Short status of an alloc:
+
 ```
-nomad alloc-status 9f3276d6-c873-c0a3-81ae-247e8c665cbe
-ID                = 9f3276d6-c873-c0a3-81ae-247e8c665cbe
-EvalID            = dc186cc2-a9b2-218e-cc00-eea3d4eaccf4
+$ nomad alloc-status --short a7365fe4-8b9f-4284-612d-a101fb41e773
+ID                = a7365fe4-8b9f-4284-612d-a101fb41e773
+EvalID            = 44c2d9ed-6377-ca3d-14a8-b2e6327230ce
 Name              = example.cache[0]
-NodeID            = <none>
+NodeID            = e55859b1-4330-f00b-da49-8a292432ead3
 JobID             = example
-ClientStatus      = failed
-ClientDescription = <none>
+ClientStatus      = running
 NodesEvaluated    = 1
-NodesFiltered     = 1
+NodesFiltered     = 0
 NodesExhausted    = 0
-AllocationTime    = 15.242µs
+AllocationTime    = 911.026µs
 CoalescedFailures = 0
+redis
+web
+
+==> Tasks
+Name   State    LastEvent  Time
+redis  running  Started    02:29:40 11/17/15
+web    running  Started    02:30:41 11/17/15
 
 ==> Status
-Allocation "9f3276d6-c873-c0a3-81ae-247e8c665cbe" status "failed" (1/1 nodes filtered)
-  * Constraint "$attr.kernel.name = linux" filtered 1 nodes
+Allocation "a7365fe4-8b9f-4284-612d-a101fb41e773" status "running" (0/1 nodes filtered)
+  * Score "e55859b1-4330-f00b-da49-8a292432ead3.binpack" = 10.334026
+```
+
+Full status of an alloc, which shows one of the tasks dying and then being restarted:
+
+```
+$ nomad alloc-status a7365fe4-8b9f-4284-612d-a101fb41e773
+ID                = a7365fe4-8b9f-4284-612d-a101fb41e773
+EvalID            = 44c2d9ed-6377-ca3d-14a8-b2e6327230ce
+Name              = example.cache[0]
+NodeID            = e55859b1-4330-f00b-da49-8a292432ead3
+JobID             = example
+ClientStatus      = running
+NodesEvaluated    = 1
+NodesFiltered     = 0
+NodesExhausted    = 0
+AllocationTime    = 911.026µs
+CoalescedFailures = 0
+
+==> Task "redis" is "running"
+Recent Events:
+Time               Type     Description
+02:29:40 11/17/15  Started  <none>
+
+==> Task "web" is "running"
+Recent Events:
+Time               Type        Description
+02:30:41 11/17/15  Started     <none>
+02:30:02 11/17/15  Terminated  Exit Code: 137, Exit Message: "Docker container exited with non-zero exit code: 137"
+02:29:40 11/17/15  Started     <none>
+
+==> Status
+Allocation "a7365fe4-8b9f-4284-612d-a101fb41e773" status "running" (0/1 nodes filtered)
+  * Score "e55859b1-4330-f00b-da49-8a292432ead3.binpack" = 10.334026
+
 ```