agent: support passing ?reason= for custom notes field values on maintenance checks

command/agent/agent.go

@@ -1027,7 +1027,7 @@ func serviceMaintCheckID(serviceID string) string {
 
 // EnableServiceMaintenance will register a false health check against the given
 // service ID with critical status. This will exclude the service from queries.
-func (a *Agent) EnableServiceMaintenance(serviceID string) error {
+func (a *Agent) EnableServiceMaintenance(serviceID, reason string) error {
 	service, ok := a.state.Services()[serviceID]
 	if !ok {
 		return fmt.Errorf("No service registered with ID %q", serviceID)
@@ -1039,12 +1039,17 @@ func (a *Agent) EnableServiceMaintenance(serviceID string) error {
 		return nil
 	}
 
+	// Use default notes if no reason provided
+	if reason == "" {
+		reason = "Maintenance mode is enabled for this service"
+	}
+
 	// Create and register the critical health check
 	check := &structs.HealthCheck{
 		Node:        a.config.NodeName,
 		CheckID:     checkID,
 		Name:        "Service Maintenance Mode",
-		Notes:       "Maintenance mode is enabled for this service",
+		Notes:       reason,
 		ServiceID:   service.ID,
 		ServiceName: service.Service,
 		Status:      structs.HealthCritical,
@@ -1076,18 +1081,23 @@ func (a *Agent) DisableServiceMaintenance(serviceID string) error {
 }
 
 // EnableNodeMaintenance places a node into maintenance mode.
-func (a *Agent) EnableNodeMaintenance() {
+func (a *Agent) EnableNodeMaintenance(reason string) {
 	// Ensure node maintenance is not already enabled
 	if _, ok := a.state.Checks()[nodeMaintCheckID]; ok {
 		return
 	}
 
+	// Use a default notes value
+	if reason == "" {
+		reason = "Maintenance mode is enabled for this node"
+	}
+
 	// Create and register the node maintenance check
 	check := &structs.HealthCheck{
 		Node:    a.config.NodeName,
 		CheckID: nodeMaintCheckID,
 		Name:    "Node Maintenance Mode",
-		Notes:   "Maintenance mode is enabled for this node",
+		Notes:   reason,
 		Status:  structs.HealthCritical,
 	}
 	a.AddCheck(check, nil, true)

command/agent/agent_endpoint.go

@@ -220,7 +220,8 @@ func (s *HTTPServer) AgentServiceMaintenance(resp http.ResponseWriter, req *http
 	}
 
 	if enable {
-		if err = s.agent.EnableServiceMaintenance(serviceID); err != nil {
+		reason := params.Get("reason")
+		if err = s.agent.EnableServiceMaintenance(serviceID, reason); err != nil {
 			resp.WriteHeader(404)
 			resp.Write([]byte(err.Error()))
 		}
@@ -257,7 +258,7 @@ func (s *HTTPServer) AgentNodeMaintenance(resp http.ResponseWriter, req *http.Re
 	}
 
 	if enable {
-		s.agent.EnableNodeMaintenance()
+		s.agent.EnableNodeMaintenance(params.Get("reason"))
 	} else {
 		s.agent.DisableNodeMaintenance()
 	}

command/agent/agent_endpoint_test.go

@@ -566,7 +566,7 @@ func TestHTTPAgent_EnableServiceMaintenance(t *testing.T) {
 	}
 
 	// Force the service into maintenance mode
-	req, _ := http.NewRequest("PUT", "/v1/agent/service/maintenance/test?enable=true", nil)
+	req, _ := http.NewRequest("PUT", "/v1/agent/service/maintenance/test?enable=true&reason=broken", nil)
 	resp := httptest.NewRecorder()
 	if _, err := srv.AgentServiceMaintenance(resp, req); err != nil {
 		t.Fatalf("err: %s", err)
@@ -577,9 +577,15 @@ func TestHTTPAgent_EnableServiceMaintenance(t *testing.T) {
 
 	// Ensure the maintenance check was registered
 	checkID := serviceMaintCheckID("test")
-	if _, ok := srv.agent.state.Checks()[checkID]; !ok {
+	check, ok := srv.agent.state.Checks()[checkID]
+	if !ok {
 		t.Fatalf("should have registered maintenance check")
 	}
+
+	// Ensure the reason was set in notes
+	if check.Notes != "broken" {
+		t.Fatalf("bad: %#v", check)
+	}
 }
 
 func TestHTTPAgent_DisableServiceMaintenance(t *testing.T) {
@@ -598,7 +604,7 @@ func TestHTTPAgent_DisableServiceMaintenance(t *testing.T) {
 	}
 
 	// Force the service into maintenance mode
-	if err := srv.agent.EnableServiceMaintenance("test"); err != nil {
+	if err := srv.agent.EnableServiceMaintenance("test", ""); err != nil {
 		t.Fatalf("err: %s", err)
 	}
 
@@ -653,7 +659,8 @@ func TestHTTPAgent_EnableNodeMaintenance(t *testing.T) {
 	defer srv.agent.Shutdown()
 
 	// Force the node into maintenance mode
-	req, _ := http.NewRequest("PUT", "/v1/agent/self/maintenance?enable=true", nil)
+	req, _ := http.NewRequest(
+		"PUT", "/v1/agent/self/maintenance?enable=true&reason=broken", nil)
 	resp := httptest.NewRecorder()
 	if _, err := srv.AgentNodeMaintenance(resp, req); err != nil {
 		t.Fatalf("err: %s", err)
@@ -663,9 +670,15 @@ func TestHTTPAgent_EnableNodeMaintenance(t *testing.T) {
 	}
 
 	// Ensure the maintenance check was registered
-	if _, ok := srv.agent.state.Checks()[nodeMaintCheckID]; !ok {
+	check, ok := srv.agent.state.Checks()[nodeMaintCheckID]
+	if !ok {
 		t.Fatalf("should have registered maintenance check")
 	}
+
+	// Ensure the reason was set in notes
+	if check.Notes != "broken" {
+		t.Fatalf("bad: %#v", check)
+	}
 }
 
 func TestHTTPAgent_DisableNodeMaintenance(t *testing.T) {
@@ -675,7 +688,7 @@ func TestHTTPAgent_DisableNodeMaintenance(t *testing.T) {
 	defer srv.agent.Shutdown()
 
 	// Force the node into maintenance mode
-	srv.agent.EnableNodeMaintenance()
+	srv.agent.EnableNodeMaintenance("")
 
 	// Leave maintenance mode
 	req, _ := http.NewRequest("PUT", "/v1/agent/self/maintenance?enable=false", nil)

command/agent/agent_test.go

@@ -916,16 +916,22 @@ func TestAgent_ServiceMaintenanceMode(t *testing.T) {
 	}
 
 	// Enter maintenance mode for the service
-	if err := agent.EnableServiceMaintenance("redis"); err != nil {
+	if err := agent.EnableServiceMaintenance("redis", "broken"); err != nil {
 		t.Fatalf("err: %s", err)
 	}
 
 	// Make sure the critical health check was added
 	checkID := serviceMaintCheckID("redis")
-	if _, ok := agent.state.Checks()[checkID]; !ok {
+	check, ok := agent.state.Checks()[checkID]
+	if !ok {
 		t.Fatalf("should have registered critical maintenance check")
 	}
 
+	// Ensure the reason was set in notes
+	if check.Notes != "broken" {
+		t.Fatalf("bad: %#v", check)
+	}
+
 	// Leave maintenance mode
 	if err := agent.DisableServiceMaintenance("redis"); err != nil {
 		t.Fatalf("err: %s", err)
@@ -935,6 +941,20 @@ func TestAgent_ServiceMaintenanceMode(t *testing.T) {
 	if _, ok := agent.state.Checks()[checkID]; ok {
 		t.Fatalf("should have deregistered maintenance check")
 	}
+
+	// Enter service maintenance mode without providing a reason
+	if err := agent.EnableServiceMaintenance("redis", ""); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	// Ensure the check was registered with the default notes
+	check, ok = agent.state.Checks()[checkID]
+	if !ok {
+		t.Fatalf("should have registered critical check")
+	}
+	if check.Notes == "" {
+		t.Fatalf("bad: %#v", check)
+	}
 }
 
 func TestAgent_NodeMaintenanceMode(t *testing.T) {
@@ -944,13 +964,19 @@ func TestAgent_NodeMaintenanceMode(t *testing.T) {
 	defer agent.Shutdown()
 
 	// Enter maintenance mode for the node
-	agent.EnableNodeMaintenance()
+	agent.EnableNodeMaintenance("broken")
 
 	// Make sure the critical health check was added
-	if _, ok := agent.state.Checks()[nodeMaintCheckID]; !ok {
+	check, ok := agent.state.Checks()[nodeMaintCheckID]
+	if !ok {
 		t.Fatalf("should have registered critical node check")
 	}
 
+	// Ensure the reason was set in notes
+	if check.Notes != "broken" {
+		t.Fatalf("bad: %#v", check)
+	}
+
 	// Leave maintenance mode
 	agent.DisableNodeMaintenance()
 
@@ -958,4 +984,16 @@ func TestAgent_NodeMaintenanceMode(t *testing.T) {
 	if _, ok := agent.state.Checks()[nodeMaintCheckID]; ok {
 		t.Fatalf("should have deregistered critical node check")
 	}
+
+	// Enter maintenance mode without passing a reason
+	agent.EnableNodeMaintenance("")
+
+	// Make sure the check was registered with the default note
+	check, ok = agent.state.Checks()[nodeMaintCheckID]
+	if !ok {
+		t.Fatalf("should have registered critical node check")
+	}
+	if check.Notes == "" {
+		t.Fatalf("bad: %#v", check)
+	}
 }