e2e: added tests for check restart behavior

e2e/consul/check_restart.go

@@ -0,0 +1,106 @@
+package consul
+
+import (
+	"fmt"
+	"os"
+	"reflect"
+	"regexp"
+	"strings"
+	"time"
+
+	e2e "github.com/hashicorp/nomad/e2e/e2eutil"
+	"github.com/hashicorp/nomad/e2e/framework"
+	"github.com/hashicorp/nomad/helper/uuid"
+)
+
+const ns = ""
+
+type CheckRestartE2ETest struct {
+	framework.TC
+	jobIds []string
+}
+
+func (tc *CheckRestartE2ETest) BeforeAll(f *framework.F) {
+	e2e.WaitForLeader(f.T(), tc.Nomad())
+	e2e.WaitForNodesReady(f.T(), tc.Nomad(), 1)
+}
+
+func (tc *CheckRestartE2ETest) AfterEach(f *framework.F) {
+	if os.Getenv("NOMAD_TEST_SKIPCLEANUP") == "1" {
+		return
+	}
+
+	for _, id := range tc.jobIds {
+		_, err := e2e.Command("nomad", "job", "stop", "-purge", id)
+		f.Assert().NoError(err)
+	}
+	tc.jobIds = []string{}
+	_, err := e2e.Command("nomad", "system", "gc")
+	f.Assert().NoError(err)
+}
+
+// TestGroupCheckRestart runs a job with a group service that will never
+// become healthy. Both tasks should be restarted up to the 'restart' limit.
+func (tc *CheckRestartE2ETest) TestGroupCheckRestart(f *framework.F) {
+
+	jobID := "test-group-check-restart-" + uuid.Generate()[0:8]
+	f.NoError(e2e.Register(jobID, "consul/input/checks_group_restart.nomad"))
+	tc.jobIds = append(tc.jobIds, jobID)
+
+	var allocID string
+
+	f.NoError(
+		e2e.WaitForAllocStatusComparison(
+			func() ([]string, error) { return e2e.AllocStatuses(jobID, ns) },
+			func(got []string) bool { return reflect.DeepEqual(got, []string{"failed"}) },
+			&e2e.WaitConfig{Interval: time.Second * 10, Retries: 30},
+		))
+
+	expected := "Exceeded allowed attempts 2 in interval 5m0s and mode is \"fail\""
+
+	out, err := e2e.Command("nomad", "alloc", "status", allocID)
+	f.NoError(err, "could not get allocation status")
+	f.Contains(out, expected,
+		fmt.Errorf("expected '%s', got\n%v", expected, out))
+
+	re := regexp.MustCompile(`Total Restarts += (.*)\n`)
+	match := re.FindAllStringSubmatch(out, -1)
+	for _, m := range match {
+		f.Equal("2", strings.TrimSpace(m[1]),
+			fmt.Errorf("expected exactly 2 restarts for both tasks, got:\n%v", out))
+	}
+}
+
+// TestTaskCheckRestart runs a job with a task service that will never become
+// healthy. Only the failed task should be restarted up to the 'restart'
+// limit.
+func (tc *CheckRestartE2ETest) TestTaskCheckRestart(f *framework.F) {
+
+	jobID := "test-task-check-restart-" + uuid.Generate()[0:8]
+	f.NoError(e2e.Register(jobID, "consul/input/checks_task_restart.nomad"))
+	tc.jobIds = append(tc.jobIds, jobID)
+
+	var allocID string
+
+	f.NoError(
+		e2e.WaitForAllocStatusComparison(
+			func() ([]string, error) { return e2e.AllocStatuses(jobID, ns) },
+			func(got []string) bool { return reflect.DeepEqual(got, []string{"failed"}) },
+			&e2e.WaitConfig{Interval: time.Second * 10, Retries: 30},
+		))
+
+	expected := "Exceeded allowed attempts 2 in interval 5m0s and mode is \"fail\""
+
+	out, err := e2e.Command("nomad", "alloc", "status", allocID)
+	f.NoError(err, "could not get allocation status")
+	f.Contains(out, expected,
+		fmt.Errorf("expected '%s', got\n%v", expected, out))
+
+	re := regexp.MustCompile(`Total Restarts += (.*)\n`)
+	match := re.FindAllStringSubmatch(out, -1)
+	f.Equal("2", strings.TrimSpace(match[0][1]),
+		fmt.Errorf("expected exactly 2 restarts for failed task, got:\n%v", out))
+
+	f.Equal("0", strings.TrimSpace(match[1][1]),
+		fmt.Errorf("expected exactly no restarts for healthy task, got:\n%v", out))
+}

e2e/consul/consul.go

@@ -35,6 +35,7 @@ func init() {
 		Cases: []framework.TestCase{
 			new(ConsulE2ETest),
 			new(ScriptChecksE2ETest),
+			new(CheckRestartE2ETest),
 		},
 	})
 }

e2e/consul/input/checks_group_restart.nomad

@@ -0,0 +1,64 @@
+job "group_check_restart" {
+  datacenters = ["dc1"]
+  type        = "service"
+
+  constraint {
+    attribute = "${attr.kernel.name}"
+    value     = "linux"
+  }
+
+  group "group_check_restart" {
+    network {
+      mode = "bridge"
+    }
+
+    restart {
+      attempts = 2
+      delay    = "1s"
+      interval = "5m"
+      mode     = "fail"
+    }
+
+    service {
+      name = "group-service-1"
+      port = "9003"
+
+      # this check should always time out and so the service
+      # should not be marked healthy, resulting in the tasks
+      # getting restarted
+      check {
+        name     = "always-dead"
+        type     = "script"
+        task     = "fail"
+        interval = "2s"
+        timeout  = "1s"
+        command  = "sleep"
+        args     = ["10"]
+
+        check_restart {
+          limit           = 2
+          grace           = "5s"
+          ignore_warnings = false
+        }
+      }
+    }
+
+    task "fail" {
+      driver = "raw_exec"
+
+      config {
+        command = "bash"
+        args    = ["-c", "sleep 15000"]
+      }
+    }
+
+    task "ok" {
+      driver = "raw_exec"
+
+      config {
+        command = "bash"
+        args    = ["-c", "sleep 15000"]
+      }
+    }
+  }
+}

e2e/consul/input/checks_task_restart.nomad

@@ -0,0 +1,63 @@
+job "task_check" {
+  datacenters = ["dc1"]
+  type        = "service"
+
+  constraint {
+    attribute = "${attr.kernel.name}"
+    value     = "linux"
+  }
+
+  group "task_check" {
+    count = 1
+
+    restart {
+      attempts = 2
+      delay    = "1s"
+      interval = "5m"
+      mode     = "fail"
+    }
+
+    task "fail" {
+
+      service {
+        name = "task-service-1"
+
+        # this check should always time out and so the service
+        # should not be marked healthy
+        check {
+          name     = "always-dead"
+          type     = "script"
+          interval = "2s"
+          timeout  = "1s"
+          command  = "sleep"
+          args     = ["10"]
+
+          check_restart {
+            limit           = 2
+            grace           = "5s"
+            ignore_warnings = false
+          }
+
+        }
+      }
+
+      driver = "raw_exec"
+
+      config {
+        command = "bash"
+        args    = ["-c", "sleep 15000"]
+      }
+    }
+
+
+    task "ok" {
+      driver = "raw_exec"
+
+      config {
+        command = "bash"
+        args    = ["-c", "sleep 15000"]
+      }
+    }
+
+  }
+}