func (s *S) TestMachinesList(c *check.C) {
iaas.RegisterIaasProvider("test-iaas", newTestIaaS)
_, err := iaas.CreateMachineForIaaS("test-iaas", map[string]string{"id": "myid1"})
defer (&iaas.Machine{Id: "myid1"}).Destroy()
c.Assert(err, check.IsNil)
_, err = iaas.CreateMachineForIaaS("test-iaas", map[string]string{"id": "myid2"})
defer (&iaas.Machine{Id: "myid2"}).Destroy()
c.Assert(err, check.IsNil)
recorder := httptest.NewRecorder()
request, err := http.NewRequest("GET", "/iaas/machines", nil)
c.Assert(err, check.IsNil)
request.Header.Set("Authorization", "bearer "+s.admintoken.GetValue())
m := RunServer(true)
m.ServeHTTP(recorder, request)
c.Assert(recorder.Code, check.Equals, http.StatusOK)
var machines []iaas.Machine
err = json.NewDecoder(recorder.Body).Decode(&machines)
c.Assert(err, check.IsNil)
c.Assert(machines[0].Id, check.Equals, "myid1")
c.Assert(machines[0].Address, check.Equals, "myid1.somewhere.com")
c.Assert(machines[0].CreationParams, check.DeepEquals, map[string]string{
"id": "myid1",
"iaas": "test-iaas",
"iaas-id": "myid1",
})
c.Assert(machines[1].Id, check.Equals, "myid2")
c.Assert(machines[1].Address, check.Equals, "myid2.somewhere.com")
c.Assert(machines[1].CreationParams, check.DeepEquals, map[string]string{
"id": "myid2",
"iaas": "test-iaas",
"iaas-id": "myid2",
})
}
func (s *S) TestHealerHealNodeWaitAndRegisterError(c *check.C) {
iaas.RegisterIaasProvider("my-healer-iaas", iaasTesting.NewHealerIaaSConstructor("addr1", nil))
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaas.RegisterIaasProvider("my-healer-iaas", iaasTesting.NewHealerIaaSConstructor("addr2", nil))
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", 2)
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p := provisiontest.ProvisionerInstance
err = p.AddNode(provision.AddNodeOptions{
Address: "http://addr1:1",
Metadata: map[string]string{"iaas": "my-healer-iaas"},
})
c.Assert(err, check.IsNil)
p.PrepareFailure("AddNode", fmt.Errorf("add node error"))
healer := newNodeHealer(nodeHealerArgs{
WaitTimeNewMachine: time.Second,
})
healer.Shutdown()
nodes, err := p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
created, err := healer.healNode(nodes[0])
c.Assert(err, check.ErrorMatches, ".*error registering new node: add node error.*")
c.Assert(created, check.IsNil)
nodes, err = p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
c.Assert(nodes[0].Status(), check.Equals, "enabled")
}
func (s *S) TestHealerHandleErrorThrottled(c *check.C) {
factory, iaasInst := iaasTesting.NewHealerIaaSConstructorWithInst("addr1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "addr2"
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", 2)
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p := provisiontest.ProvisionerInstance
err = p.AddNode(provision.AddNodeOptions{
Address: "http://addr1:1",
Metadata: map[string]string{"iaas": "my-healer-iaas"},
})
c.Assert(err, check.IsNil)
node, err := p.GetNode("http://addr1:1")
c.Assert(err, check.IsNil)
healer := newNodeHealer(nodeHealerArgs{
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Minute,
})
healer.Shutdown()
healer.started = time.Now().Add(-3 * time.Second)
conf := healerConfig()
err = conf.SaveBase(NodeHealerConfig{Enabled: boolPtr(true), MaxUnresponsiveTime: intPtr(1)})
c.Assert(err, check.IsNil)
err = healer.UpdateNodeData(node, []provision.NodeCheckResult{})
c.Assert(err, check.IsNil)
time.Sleep(1200 * time.Millisecond)
nodes, err := p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr1")
for i := 0; i < 3; i++ {
var evt *event.Event
evt, err = event.NewInternal(&event.Opts{
Target: event.Target{Type: "node", Value: nodes[0].Address()},
InternalKind: "healer",
Allowed: event.Allowed(permission.PermPoolReadEvents),
})
c.Assert(err, check.IsNil)
err = evt.Done(nil)
c.Assert(err, check.IsNil)
}
err = healer.tryHealingNode(nodes[0], "myreason", nil)
c.Assert(err, check.ErrorMatches, "Error trying to insert node healing event, healing aborted: event throttled, limit for healer on node \".*?\" is 3 every 5m0s")
nodes, err = p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
}
func (s *S) TestHealerHealNodeWaitAndRegisterError(c *check.C) {
defer func() {
machines, _ := iaas.ListMachines()
for _, m := range machines {
m.Destroy()
}
}()
iaas.RegisterIaasProvider("my-healer-iaas", newHealerIaaSConstructor("127.0.0.1", nil))
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaas.RegisterIaasProvider("my-healer-iaas", newHealerIaaSConstructor("localhost", nil))
node1, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
node2, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
node2.PrepareFailure("ping-failure", "/_ping")
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", urlPort(node2.URL()))
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
cluster, err := cluster.New(nil, &cluster.MapStorage{},
cluster.Node{Address: node1.URL()},
)
c.Assert(err, check.IsNil)
node1.PrepareFailure("pingErr", "/_ping")
cluster.StartActiveMonitoring(100 * time.Millisecond)
time.Sleep(300 * time.Millisecond)
cluster.StopActiveMonitoring()
var p dockerProvisioner
err = p.Initialize()
c.Assert(err, check.IsNil)
p.cluster = cluster
healer := nodeHealer{
locks: make(map[string]*sync.Mutex),
provisioner: &p,
disabledTime: 0,
failuresBeforeHealing: 1,
waitTimeNewMachine: 1 * time.Second,
}
nodes, err := p.getCluster().UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(urlPort(nodes[0].Address), check.Equals, urlPort(node1.URL()))
c.Assert(urlToHost(nodes[0].Address), check.Equals, "127.0.0.1")
c.Assert(nodes[0].FailureCount() > 0, check.Equals, true)
nodes[0].Metadata["iaas"] = "my-healer-iaas"
created, err := healer.healNode(&nodes[0])
c.Assert(err, check.ErrorMatches, ".*error registering new node.*")
c.Assert(created.Address, check.Equals, "")
c.Assert(nodes[0].FailureCount(), check.Equals, 0)
nodes, err = p.getCluster().UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(urlPort(nodes[0].Address), check.Equals, urlPort(node1.URL()))
c.Assert(urlToHost(nodes[0].Address), check.Equals, "127.0.0.1")
}
func (s *S) TestMachinesDestroy(c *check.C) {
iaas.RegisterIaasProvider("test-iaas", newTestIaaS)
_, err := iaas.CreateMachineForIaaS("test-iaas", map[string]string{"id": "myid1"})
c.Assert(err, check.IsNil)
recorder := httptest.NewRecorder()
request, err := http.NewRequest("DELETE", "/iaas/machines/myid1", nil)
c.Assert(err, check.IsNil)
request.Header.Set("Authorization", "bearer "+s.admintoken.GetValue())
m := RunServer(true)
m.ServeHTTP(recorder, request)
c.Assert(recorder.Code, check.Equals, http.StatusOK)
}
func (s *S) TestHealerHealNodeDestroyError(c *check.C) {
factory, iaasInst := iaasTesting.NewHealerIaaSConstructorWithInst("addr1")
iaasInst.DelErr = fmt.Errorf("my destroy error")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "addr2"
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", 2)
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p := provisiontest.ProvisionerInstance
err = p.AddNode(provision.AddNodeOptions{
Address: "http://addr1:1",
Metadata: map[string]string{"iaas": "my-healer-iaas"},
})
c.Assert(err, check.IsNil)
healer := newNodeHealer(nodeHealerArgs{
WaitTimeNewMachine: time.Minute,
})
healer.Shutdown()
nodes, err := p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr1")
buf := bytes.Buffer{}
log.SetLogger(log.NewWriterLogger(&buf, false))
defer log.SetLogger(nil)
created, err := healer.healNode(nodes[0])
c.Assert(err, check.IsNil)
c.Assert(created.Address, check.Equals, "http://addr2:2")
c.Assert(buf.String(), check.Matches, "(?s).*my destroy error.*")
nodes, err = p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr2:2")
machines, err = iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr2")
}
func (a *autoScaleConfig) addNode(event *autoScaleEvent, modelNodes []*cluster.Node) (*cluster.Node, error) {
metadata, err := chooseMetadataFromNodes(modelNodes)
if err != nil {
return nil, err
}
_, hasIaas := metadata["iaas"]
if !hasIaas {
return nil, fmt.Errorf("no IaaS information in nodes metadata: %#v", metadata)
}
machine, err := iaas.CreateMachineForIaaS(metadata["iaas"], metadata)
if err != nil {
return nil, fmt.Errorf("unable to create machine: %s", err.Error())
}
newAddr := machine.FormatNodeAddress()
event.logMsg("new machine created: %s - Waiting for docker to start...", newAddr)
createdNode := cluster.Node{
Address: newAddr,
Metadata: metadata,
CreationStatus: cluster.NodeCreationStatusPending,
}
err = a.provisioner.Cluster().Register(createdNode)
if err != nil {
machine.Destroy()
return nil, fmt.Errorf("error registering new node %s: %s", newAddr, err.Error())
}
q, err := queue.Queue()
if err == nil {
jobParams := monsterqueue.JobParams{
"endpoint": createdNode.Address,
"machine": machine.Id,
"metadata": createdNode.Metadata,
}
var job monsterqueue.Job
job, err = q.EnqueueWait(bs.QueueTaskName, jobParams, a.WaitTimeNewMachine)
if err == nil {
_, err = job.Result()
}
}
if err != nil {
machine.Destroy()
a.provisioner.Cluster().Unregister(newAddr)
return nil, fmt.Errorf("error running bs task: %s", err)
}
event.logMsg("new machine created: %s - started!", newAddr)
return &createdNode, nil
}
func (s *S) TestHealerHealNode(c *check.C) {
factory, iaasInst := iaasTesting.NewHealerIaaSConstructorWithInst("addr1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "addr2"
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", 2)
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p := provisiontest.ProvisionerInstance
err = p.AddNode(provision.AddNodeOptions{
Address: "http://addr1:1",
Metadata: map[string]string{"iaas": "my-healer-iaas"},
})
c.Assert(err, check.IsNil)
healer := newNodeHealer(nodeHealerArgs{
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Minute,
})
healer.Shutdown()
nodes, err := p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr1")
created, err := healer.healNode(nodes[0])
c.Assert(err, check.IsNil)
c.Assert(created.Address, check.Equals, "http://addr2:2")
nodes, err = p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr2:2")
machines, err = iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr2")
}
func (s *HandlersSuite) TestRemoveNodeHandlerRemoveIaaS(c *check.C) {
iaas.RegisterIaasProvider("my-xxx-iaas", newTestIaaS)
machine, err := iaas.CreateMachineForIaaS("my-xxx-iaas", map[string]string{})
c.Assert(err, check.IsNil)
mainDockerProvisioner.cluster, err = cluster.New(nil, &cluster.MapStorage{})
c.Assert(err, check.IsNil)
_, err = mainDockerProvisioner.getCluster().Register(fmt.Sprintf("http://%s:2375", machine.Address), nil)
c.Assert(err, check.IsNil)
b := bytes.NewBufferString(fmt.Sprintf(`{"address": "http://%s:2375", "remove_iaas": "true"}`, machine.Address))
req, err := http.NewRequest("POST", "/node/remove", b)
c.Assert(err, check.IsNil)
rec := httptest.NewRecorder()
err = removeNodeHandler(rec, req, nil)
c.Assert(err, check.IsNil)
nodes, err := mainDockerProvisioner.getCluster().Nodes()
c.Assert(len(nodes), check.Equals, 0)
_, err = iaas.FindMachineById(machine.Id)
c.Assert(err, check.Equals, mgo.ErrNotFound)
}
func (s *S) TestHealerHealNodeWaitAndRegisterError(c *check.C) {
iaas.RegisterIaasProvider("my-healer-iaas", dockertest.NewHealerIaaSConstructor("127.0.0.1", nil))
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaas.RegisterIaasProvider("my-healer-iaas", dockertest.NewHealerIaaSConstructor("localhost", nil))
node1, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
node2, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
node2.PrepareFailure("ping-failure", "/_ping")
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", dockertest.URLPort(node2.URL()))
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p, err := s.newFakeDockerProvisioner(node1.URL())
c.Assert(err, check.IsNil)
defer p.Destroy()
node1.PrepareFailure("pingErr", "/_ping")
p.Cluster().StartActiveMonitoring(100 * time.Millisecond)
time.Sleep(300 * time.Millisecond)
p.Cluster().StopActiveMonitoring()
healer := NewNodeHealer(NodeHealerArgs{
Provisioner: p,
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Second,
})
nodes, err := p.Cluster().UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(dockertest.URLPort(nodes[0].Address), check.Equals, dockertest.URLPort(node1.URL()))
c.Assert(net.URLToHost(nodes[0].Address), check.Equals, "127.0.0.1")
c.Assert(nodes[0].FailureCount() > 0, check.Equals, true)
nodes[0].Metadata["iaas"] = "my-healer-iaas"
created, err := healer.healNode(&nodes[0])
c.Assert(err, check.ErrorMatches, ".*timeout waiting for result.*")
c.Assert(created.Address, check.Equals, "")
c.Assert(nodes[0].FailureCount(), check.Equals, 0)
nodes, err = p.Cluster().Nodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(dockertest.URLPort(nodes[0].Address), check.Equals, dockertest.URLPort(node1.URL()))
c.Assert(net.URLToHost(nodes[0].Address), check.Equals, "127.0.0.1")
}
func (s *S) TestMachinesDestroy(c *check.C) {
iaas.RegisterIaasProvider("test-iaas", newTestIaaS)
_, err := iaas.CreateMachineForIaaS("test-iaas", map[string]string{"id": "myid1"})
c.Assert(err, check.IsNil)
recorder := httptest.NewRecorder()
request, err := http.NewRequest("DELETE", "/iaas/machines/myid1", nil)
c.Assert(err, check.IsNil)
request.Header.Set("Authorization", "bearer "+s.token.GetValue())
m := RunServer(true)
m.ServeHTTP(recorder, request)
c.Assert(recorder.Code, check.Equals, http.StatusOK)
c.Assert(eventtest.EventDesc{
Target: event.Target{Type: event.TargetTypeIaas, Value: "test-iaas"},
Owner: s.token.GetUserName(),
Kind: "machine.delete",
StartCustomData: []map[string]interface{}{
{"name": ":machine_id", "value": "myid1"},
},
}, eventtest.HasEvent)
}
func (s *HandlersSuite) TestRemoveNodeHandlerWithoutRemoveIaaS(c *gocheck.C) {
iaas.RegisterIaasProvider("some-iaas", TestIaaS{})
machine, err := iaas.CreateMachineForIaaS("some-iaas", map[string]string{})
c.Assert(err, gocheck.IsNil)
dCluster, err = cluster.New(nil, &cluster.MapStorage{})
c.Assert(err, gocheck.IsNil)
err = dCluster.Register(fmt.Sprintf("http://%s:4243", machine.Address), nil)
c.Assert(err, gocheck.IsNil)
b := bytes.NewBufferString(fmt.Sprintf(`{"address": "http://%s:4243", "remove_iaas": "false"}`, machine.Address))
req, err := http.NewRequest("POST", "/node/remove", b)
c.Assert(err, gocheck.IsNil)
rec := httptest.NewRecorder()
err = removeNodeHandler(rec, req, nil)
c.Assert(err, gocheck.IsNil)
nodes, err := dCluster.Nodes()
c.Assert(len(nodes), gocheck.Equals, 0)
dbM, err := iaas.FindMachineById(machine.Id)
c.Assert(err, gocheck.IsNil)
c.Assert(dbM.Id, gocheck.Equals, machine.Id)
}
func (h *nodeHealer) healNode(node *cluster.Node) (cluster.Node, error) {
emptyNode := cluster.Node{}
failingAddr := node.Address
nodeMetadata := node.CleanMetadata()
failingHost := urlToHost(failingAddr)
failures := node.FailureCount()
machine, err := iaas.CreateMachineForIaaS(nodeMetadata["iaas"], nodeMetadata)
if err != nil {
node.ResetFailures()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error creating new machine: %s", failures, failingHost, err.Error())
}
err = h.provisioner.getCluster().Unregister(failingAddr)
if err != nil {
machine.Destroy()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error unregistering old node: %s", failures, failingHost, err.Error())
}
newAddr := machine.FormatNodeAddress()
log.Debugf("New machine created during healing process: %s - Waiting for docker to start...", newAddr)
createdNode, err := h.provisioner.getCluster().WaitAndRegister(newAddr, nodeMetadata, h.waitTimeNewMachine)
if err != nil {
node.ResetFailures()
h.provisioner.getCluster().Register(failingAddr, nodeMetadata)
machine.Destroy()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error registering new node: %s", failures, failingHost, err.Error())
}
var buf bytes.Buffer
err = h.provisioner.moveContainers(failingHost, "", &buf)
if err != nil {
log.Errorf("Unable to move containers, skipping containers healing %q -> %q: %s: %s", failingHost, machine.Address, err.Error(), buf.String())
}
failingMachine, err := iaas.FindMachineByIdOrAddress(node.Metadata["iaas-id"], failingHost)
if err != nil {
return createdNode, fmt.Errorf("Unable to find failing machine %s in IaaS: %s", failingHost, err.Error())
}
err = failingMachine.Destroy()
if err != nil {
return createdNode, fmt.Errorf("Unable to destroy machine %s from IaaS: %s", failingHost, err.Error())
}
log.Debugf("Done auto-healing node %q, node %q created in its place.", failingHost, machine.Address)
return createdNode, nil
}
func addNodeForParams(params map[string]string, isRegister bool) (map[string]string, error) {
response := make(map[string]string)
var address string
if isRegister {
address, _ = params["address"]
delete(params, "address")
} else {
iaasName, _ := params["iaas"]
desc, err := iaas.Describe(iaasName)
if err != nil {
return response, err
}
response["description"] = desc
var m *iaas.Machine
if iaasName != "" {
m, err = iaas.CreateMachineForIaaS(iaasName, params)
} else {
m, err = iaas.CreateMachine(params)
}
if err != nil {
return response, err
}
nodeAddress, err := m.FormatNodeAddress()
if err != nil {
return response, err
}
params["iaas"] = m.Iaas
address = nodeAddress
}
err := validateNodeAddress(address)
if err != nil {
return response, err
}
err = dockerCluster().Register(address, params)
if err != nil {
return response, err
}
return response, err
}
func (s *S) TestHealerHealNodeCreateMachineError(c *check.C) {
factory, iaasInst := dockertest.NewHealerIaaSConstructorWithInst("127.0.0.1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "localhost"
iaasInst.Err = fmt.Errorf("my create machine error")
node1, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
p, err := s.newFakeDockerProvisioner(node1.URL())
c.Assert(err, check.IsNil)
defer p.Destroy()
node1.PrepareFailure("pingErr", "/_ping")
p.Cluster().StartActiveMonitoring(100 * time.Millisecond)
time.Sleep(300 * time.Millisecond)
p.Cluster().StopActiveMonitoring()
healer := NewNodeHealer(NodeHealerArgs{
Provisioner: p,
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Minute,
})
nodes, err := p.Cluster().UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(dockertest.URLPort(nodes[0].Address), check.Equals, dockertest.URLPort(node1.URL()))
c.Assert(net.URLToHost(nodes[0].Address), check.Equals, "127.0.0.1")
c.Assert(nodes[0].FailureCount() > 0, check.Equals, true)
nodes[0].Metadata["iaas"] = "my-healer-iaas"
created, err := healer.healNode(&nodes[0])
c.Assert(err, check.ErrorMatches, ".*my create machine error.*")
c.Assert(created.Address, check.Equals, "")
c.Assert(nodes[0].FailureCount(), check.Equals, 0)
nodes, err = p.Cluster().UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(dockertest.URLPort(nodes[0].Address), check.Equals, dockertest.URLPort(node1.URL()))
c.Assert(net.URLToHost(nodes[0].Address), check.Equals, "127.0.0.1")
}
func (a *autoScaleConfig) addNode(evt *event.Event, modelNodes []*cluster.Node) (*cluster.Node, error) {
metadata, err := chooseMetadataFromNodes(modelNodes)
if err != nil {
return nil, err
}
_, hasIaas := metadata["iaas"]
if !hasIaas {
return nil, errors.Errorf("no IaaS information in nodes metadata: %#v", metadata)
}
machine, err := iaas.CreateMachineForIaaS(metadata["iaas"], metadata)
if err != nil {
return nil, errors.Wrap(err, "unable to create machine")
}
newAddr := machine.FormatNodeAddress()
evt.Logf("new machine created: %s - Waiting for docker to start...", newAddr)
createOpts := provision.AddNodeOptions{
Address: newAddr,
Metadata: metadata,
WaitTO: a.WaitTimeNewMachine,
CaCert: machine.CaCert,
ClientCert: machine.ClientCert,
ClientKey: machine.ClientKey,
}
err = a.provisioner.AddNode(createOpts)
if err != nil {
machine.Destroy()
a.provisioner.Cluster().Unregister(newAddr)
return nil, errors.Wrapf(err, "error adding new node %s", newAddr)
}
createdNode, err := a.provisioner.Cluster().GetNode(newAddr)
if err != nil {
machine.Destroy()
a.provisioner.Cluster().Unregister(newAddr)
return nil, errors.Wrapf(err, "error retrieving new node %s", newAddr)
}
evt.Logf("new machine created: %s - started!", newAddr)
return &createdNode, nil
}
func (s *S) TestTryHealingNodeDoubleCheck(c *check.C) {
factory, iaasInst := iaasTesting.NewHealerIaaSConstructorWithInst("addr1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "addr2"
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", 2)
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p := provisiontest.ProvisionerInstance
err = p.AddNode(provision.AddNodeOptions{
Address: "http://addr1:1",
Metadata: map[string]string{"iaas": "my-healer-iaas"},
})
c.Assert(err, check.IsNil)
healer := newNodeHealer(nodeHealerArgs{
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Minute,
})
healer.started = time.Now().Add(-3 * time.Second)
nodes, err := p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
healErr := healer.tryHealingNode(nodes[0], "something", nil)
c.Assert(healErr, check.IsNil)
nodes, err = p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr1")
c.Assert(eventtest.EventDesc{
IsEmpty: true,
}, eventtest.HasEvent)
}
func (s *S) TestRemoveNodeHandlerWithRemoveIaaS(c *check.C) {
iaas.RegisterIaasProvider("some-iaas", newTestIaaS)
machine, err := iaas.CreateMachineForIaaS("some-iaas", map[string]string{"id": "m1"})
c.Assert(err, check.IsNil)
err = s.provisioner.AddNode(provision.AddNodeOptions{
Address: fmt.Sprintf("http://%s:2375", machine.Address),
})
c.Assert(err, check.IsNil)
u := fmt.Sprintf("/node/http://%s:2375?remove-iaas=true", machine.Address)
req, err := http.NewRequest("DELETE", u, nil)
c.Assert(err, check.IsNil)
req.Header.Set("Authorization", "bearer "+s.token.GetValue())
rec := httptest.NewRecorder()
server := RunServer(true)
server.ServeHTTP(rec, req)
c.Assert(rec.Body.String(), check.Equals, "rebalancing...remove done!")
c.Assert(rec.Code, check.Equals, http.StatusOK)
nodes, err := s.provisioner.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 0)
_, err = iaas.FindMachineById(machine.Id)
c.Assert(err, check.Equals, mgo.ErrNotFound)
}
func (a *autoScaleConfig) addNode(modelNodes []*cluster.Node) (*cluster.Node, error) {
metadata, err := chooseMetadataFromNodes(modelNodes)
if err != nil {
return nil, err
}
_, hasIaas := metadata["iaas"]
if !hasIaas {
return nil, fmt.Errorf("no IaaS information in nodes metadata: %#v", metadata)
}
machine, err := iaas.CreateMachineForIaaS(metadata["iaas"], metadata)
if err != nil {
return nil, fmt.Errorf("unable to create machine: %s", err.Error())
}
newAddr := machine.FormatNodeAddress()
a.logDebug("new machine created: %s - Waiting for docker to start...", newAddr)
createdNode, err := a.provisioner.getCluster().WaitAndRegister(newAddr, metadata, a.waitTimeNewMachine)
if err != nil {
machine.Destroy()
return nil, fmt.Errorf("error registering new node %s: %s", newAddr, err.Error())
}
a.logDebug("new machine created: %s - started!", newAddr)
return &createdNode, nil
}
func (s *S) TestHealerHealNodeCreateMachineError(c *check.C) {
factory, iaasInst := iaasTesting.NewHealerIaaSConstructorWithInst("addr1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "addr2"
iaasInst.Err = fmt.Errorf("my create machine error")
p := provisiontest.ProvisionerInstance
err = p.AddNode(provision.AddNodeOptions{
Address: "http://addr1:1",
Metadata: map[string]string{"iaas": "my-healer-iaas"},
})
c.Assert(err, check.IsNil)
healer := newNodeHealer(nodeHealerArgs{
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Minute,
})
healer.Shutdown()
nodes, err := p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
fakeNode := nodes[0].(*provisiontest.FakeNode)
fakeNode.SetHealth(1, false)
c.Assert(fakeNode.FailureCount() > 0, check.Equals, true)
created, err := healer.healNode(nodes[0])
c.Assert(err, check.ErrorMatches, ".*my create machine error.*")
c.Assert(created, check.IsNil)
c.Assert(fakeNode.FailureCount(), check.Equals, 0)
nodes, err = p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
c.Assert(nodes[0].Status(), check.Equals, "enabled")
}
func (s *S) TestHealerHandleErrorFailureEvent(c *check.C) {
factory, iaasInst := iaasTesting.NewHealerIaaSConstructorWithInst("addr1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "addr2"
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", 2)
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p := provisiontest.ProvisionerInstance
err = p.AddNode(provision.AddNodeOptions{
Address: "http://addr1:1",
Metadata: map[string]string{"iaas": "my-healer-iaas"},
})
c.Assert(err, check.IsNil)
node, err := p.GetNode("http://addr1:1")
c.Assert(err, check.IsNil)
healer := newNodeHealer(nodeHealerArgs{
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Minute,
})
healer.Shutdown()
healer.started = time.Now().Add(-3 * time.Second)
conf := healerConfig()
err = conf.SaveBase(NodeHealerConfig{Enabled: boolPtr(true), MaxUnresponsiveTime: intPtr(1)})
c.Assert(err, check.IsNil)
err = healer.UpdateNodeData(node, []provision.NodeCheckResult{})
c.Assert(err, check.IsNil)
time.Sleep(1200 * time.Millisecond)
nodes, err := p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr1")
p.PrepareFailure("AddNode", fmt.Errorf("error registering new node"))
nodes[0].(*provisiontest.FakeNode).SetHealth(2, true)
waitTime := healer.HandleError(nodes[0].(provision.NodeHealthChecker))
c.Assert(waitTime, check.Equals, time.Duration(0))
nodes, err = p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
machines, err = iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr1")
c.Assert(eventtest.EventDesc{
Target: event.Target{Type: "node", Value: "http://addr1:1"},
Kind: "healer",
StartCustomData: map[string]interface{}{
"reason": "2 consecutive failures",
"node._id": "http://addr1:1",
},
ErrorMatches: `Can't auto-heal after 2 failures for node addr1: error registering new node: error registering new node`,
}, eventtest.HasEvent)
}
func (s *S) TestCheckActiveHealing(c *check.C) {
conf := healerConfig()
err := conf.SaveBase(NodeHealerConfig{Enabled: boolPtr(true), MaxUnresponsiveTime: intPtr(1)})
c.Assert(err, check.IsNil)
factory, iaasInst := iaasTesting.NewHealerIaaSConstructorWithInst("addr1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err = iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "addr2"
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", 2)
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p := provisiontest.ProvisionerInstance
err = p.AddNode(provision.AddNodeOptions{
Address: "http://addr1:1",
Metadata: map[string]string{"iaas": "my-healer-iaas"},
})
c.Assert(err, check.IsNil)
node, err := p.GetNode("http://addr1:1")
c.Assert(err, check.IsNil)
healer := newNodeHealer(nodeHealerArgs{
WaitTimeNewMachine: time.Minute,
})
healer.Shutdown()
healer.started = time.Now().Add(-3 * time.Second)
err = healer.UpdateNodeData(node, []provision.NodeCheckResult{})
c.Assert(err, check.IsNil)
time.Sleep(1200 * time.Millisecond)
nodes, err := p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr1")
healer.runActiveHealing()
nodes, err = p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr2:2")
machines, err = iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr2")
c.Assert(eventtest.EventDesc{
Target: event.Target{Type: "node", Value: "http://addr1:1"},
Kind: "healer",
StartCustomData: map[string]interface{}{
"reason": bson.M{"$regex": `last update \d+\.\d*?s ago, last success \d+\.\d*?s ago`},
"lastcheck.time": bson.M{"$exists": true},
"node._id": "http://addr1:1",
},
EndCustomData: map[string]interface{}{
"_id": "http://addr2:2",
},
}, eventtest.HasEvent)
}
func (h *Healer) HandleError(node cluster.Node) time.Duration {
defaultWait := 1 * time.Minute
failures := node.FailureCount()
if failures < 5 {
return defaultWait
}
failingAddr := node.Address
failingHost := urlToHost(failingAddr)
containers, err := listContainersByHost(failingHost)
if err != nil {
log.Errorf("Error in cluster healer, trying to list containers: %s", err.Error())
return defaultWait
}
// Empty host let's just try again in the future
if len(containers) == 0 {
return defaultWait
}
iaasName, hasIaas := node.Metadata["iaas"]
if !hasIaas {
log.Errorf("Can't auto-heal after %d failures for node %s: no IaaS information.", failures, failingHost)
return defaultWait
}
machine, err := iaas.CreateMachineForIaaS(iaasName, node.Metadata)
if err != nil {
log.Errorf("Can't auto-heal after %d failures for node %s: error creating new machine: %s", failures, failingHost, err.Error())
return defaultWait
}
newAddr, err := machine.FormatNodeAddress()
if err != nil {
log.Errorf("Can't auto-heal after %d failures for node %s: error formatting address: %s", failures, failingHost, err.Error())
machine.Destroy()
return defaultWait
}
cluster := dockerCluster()
err = cluster.Unregister(failingAddr)
if err != nil {
log.Errorf("Can't auto-heal after %d failures for node %s: error unregistering old node: %s", failures, failingHost, err.Error())
return defaultWait
}
err = cluster.WaitAndRegister(newAddr, node.Metadata, 2*time.Minute)
if err != nil {
log.Errorf("Can't auto-heal after %d failures for node %s: error registering new node: %s", failures, failingHost, err.Error())
machine.Destroy()
return defaultWait
}
var buf bytes.Buffer
encoder := json.NewEncoder(&buf)
err = moveContainers(failingHost, machine.Address, encoder)
if err != nil {
log.Errorf("Unable to move containers from: %s to: %s - %s", failingHost, machine.Address, err.Error())
return 0
}
failingMachine, err := iaas.FindMachineByAddress(failingHost)
if err != nil {
log.Errorf("Unable to find failing machine %s in IaaS", failingHost)
return 0
}
err = failingMachine.Destroy()
if err != nil {
log.Errorf("Unable to find destroy machine %s from IaaS", failingHost)
}
return 0
}
func (s *S) TestTryHealingNodeConcurrent(c *check.C) {
defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(10))
factory, iaasInst := iaasTesting.NewHealerIaaSConstructorWithInst("addr1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "addr2"
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", 2)
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p := provisiontest.ProvisionerInstance
err = p.AddNode(provision.AddNodeOptions{
Address: "http://addr1:1",
Metadata: map[string]string{"iaas": "my-healer-iaas"},
})
c.Assert(err, check.IsNil)
node, err := p.GetNode("http://addr1:1")
c.Assert(err, check.IsNil)
healer := newNodeHealer(nodeHealerArgs{
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Minute,
})
healer.started = time.Now().Add(-3 * time.Second)
conf := healerConfig()
err = conf.SaveBase(NodeHealerConfig{Enabled: boolPtr(true), MaxUnresponsiveTime: intPtr(1)})
c.Assert(err, check.IsNil)
err = healer.UpdateNodeData(node, []provision.NodeCheckResult{})
c.Assert(err, check.IsNil)
time.Sleep(1200 * time.Millisecond)
nodes, err := p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr1:1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr1")
wg := sync.WaitGroup{}
for i := 0; i < 100; i++ {
wg.Add(1)
go func() {
defer wg.Done()
healErr := healer.tryHealingNode(nodes[0], "something", nil)
c.Assert(healErr, check.IsNil)
}()
}
wg.Wait()
nodes, err = p.ListNodes(nil)
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(nodes[0].Address(), check.Equals, "http://addr2:2")
machines, err = iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "addr2")
c.Assert(eventtest.EventDesc{
Target: event.Target{Type: "node", Value: "http://addr1:1"},
Kind: "healer",
StartCustomData: map[string]interface{}{
"reason": "something",
"node._id": "http://addr1:1",
},
EndCustomData: map[string]interface{}{
"_id": "http://addr2:2",
},
}, eventtest.HasEvent)
}
func (h *NodeHealer) healNode(node provision.Node) (*provision.NodeSpec, error) {
failingAddr := node.Address()
// Copy metadata to ensure underlying data structure is not modified.
newNodeMetadata := map[string]string{}
for k, v := range node.Metadata() {
newNodeMetadata[k] = v
}
failingHost := net.URLToHost(failingAddr)
healthNode, isHealthNode := node.(provision.NodeHealthChecker)
failures := 0
if isHealthNode {
failures = healthNode.FailureCount()
}
machine, err := iaas.CreateMachineForIaaS(newNodeMetadata["iaas"], newNodeMetadata)
if err != nil {
if isHealthNode {
healthNode.ResetFailures()
}
return nil, errors.Wrapf(err, "Can't auto-heal after %d failures for node %s: error creating new machine", failures, failingHost)
}
err = node.Provisioner().UpdateNode(provision.UpdateNodeOptions{
Address: failingAddr,
Disable: true,
})
if err != nil {
machine.Destroy()
return nil, errors.Wrapf(err, "Can't auto-heal after %d failures for node %s: error unregistering old node", failures, failingHost)
}
newAddr := machine.FormatNodeAddress()
log.Debugf("New machine created during healing process: %s - Waiting for docker to start...", newAddr)
createOpts := provision.AddNodeOptions{
Address: newAddr,
Metadata: newNodeMetadata,
WaitTO: h.waitTimeNewMachine,
CaCert: machine.CaCert,
ClientCert: machine.ClientCert,
ClientKey: machine.ClientKey,
}
err = node.Provisioner().AddNode(createOpts)
if err != nil {
if isHealthNode {
healthNode.ResetFailures()
}
node.Provisioner().UpdateNode(provision.UpdateNodeOptions{Address: failingAddr, Enable: true})
machine.Destroy()
return nil, errors.Wrapf(err, "Can't auto-heal after %d failures for node %s: error registering new node", failures, failingHost)
}
nodeSpec := provision.NodeToSpec(node)
nodeSpec.Address = newAddr
nodeSpec.Metadata = newNodeMetadata
var buf bytes.Buffer
err = node.Provisioner().RemoveNode(provision.RemoveNodeOptions{
Address: failingAddr,
Rebalance: true,
Writer: &buf,
})
if err != nil {
log.Errorf("Unable to move containers, skipping containers healing %q -> %q: %s: %s", failingHost, machine.Address, err, buf.String())
}
failingMachine, err := iaas.FindMachineByIdOrAddress(node.Metadata()["iaas-id"], failingHost)
if err != nil {
return &nodeSpec, errors.Wrapf(err, "Unable to find failing machine %s in IaaS", failingHost)
}
err = failingMachine.Destroy()
if err != nil {
return &nodeSpec, errors.Wrapf(err, "Unable to destroy machine %s from IaaS", failingHost)
}
log.Debugf("Done auto-healing node %q, node %q created in its place.", failingHost, machine.Address)
return &nodeSpec, nil
}
func (s *S) TestHealerHealNode(c *check.C) {
factory, iaasInst := dockertest.NewHealerIaaSConstructorWithInst("127.0.0.1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "localhost"
node1, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
node2, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", dockertest.URLPort(node2.URL()))
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p, err := s.newFakeDockerProvisioner(node1.URL())
c.Assert(err, check.IsNil)
defer p.Destroy()
app := provisiontest.NewFakeApp("myapp", "python", 0)
_, err = p.StartContainers(dockertest.StartContainersArgs{
Endpoint: node1.URL(),
App: app,
Amount: map[string]int{"web": 1},
Image: "tsuru/python",
PullImage: true,
})
c.Assert(err, check.IsNil)
healer := NewNodeHealer(NodeHealerArgs{
Provisioner: p,
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Minute,
})
nodes, err := p.Cluster().UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(dockertest.URLPort(nodes[0].Address), check.Equals, dockertest.URLPort(node1.URL()))
c.Assert(net.URLToHost(nodes[0].Address), check.Equals, "127.0.0.1")
containers := p.AllContainers()
c.Assert(err, check.IsNil)
c.Assert(containers, check.HasLen, 1)
c.Assert(containers[0].HostAddr, check.Equals, "127.0.0.1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "127.0.0.1")
nodes[0].Metadata["iaas"] = "my-healer-iaas"
created, err := healer.healNode(&nodes[0])
c.Assert(err, check.IsNil)
c.Assert(created.Address, check.Equals, fmt.Sprintf("http://localhost:%d", dockertest.URLPort(node2.URL())))
nodes, err = p.Cluster().UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(dockertest.URLPort(nodes[0].Address), check.Equals, dockertest.URLPort(node2.URL()))
c.Assert(net.URLToHost(nodes[0].Address), check.Equals, "localhost")
machines, err = iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "localhost")
err = tsurutest.WaitCondition(5*time.Second, func() bool {
containers := p.AllContainers()
return len(containers) == 1 && containers[0].HostAddr == "localhost"
})
c.Assert(err, check.IsNil)
}
func (s *S) TestHealerHandleError(c *check.C) {
factory, iaasInst := dockertest.NewHealerIaaSConstructorWithInst("127.0.0.1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.Addr = "localhost"
node1, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
node2, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", dockertest.URLPort(node2.URL()))
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
p, err := s.newFakeDockerProvisioner(node1.URL())
c.Assert(err, check.IsNil)
defer p.Destroy()
app := provisiontest.NewFakeApp("myapp", "python", 0)
_, err = p.StartContainers(dockertest.StartContainersArgs{
Endpoint: node1.URL(),
App: app,
Amount: map[string]int{"web": 1},
Image: "tsuru/python",
PullImage: true,
})
c.Assert(err, check.IsNil)
healer := NewNodeHealer(NodeHealerArgs{
Provisioner: p,
FailuresBeforeHealing: 1,
WaitTimeNewMachine: time.Minute,
})
nodes, err := p.Cluster().UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(dockertest.URLPort(nodes[0].Address), check.Equals, dockertest.URLPort(node1.URL()))
c.Assert(net.URLToHost(nodes[0].Address), check.Equals, "127.0.0.1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "127.0.0.1")
nodes[0].Metadata["iaas"] = "my-healer-iaas"
nodes[0].Metadata["Failures"] = "2"
waitTime := healer.HandleError(&nodes[0])
c.Assert(waitTime, check.Equals, time.Duration(0))
nodes, err = p.Cluster().Nodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(dockertest.URLPort(nodes[0].Address), check.Equals, dockertest.URLPort(node2.URL()))
c.Assert(net.URLToHost(nodes[0].Address), check.Equals, "localhost")
machines, err = iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "localhost")
healingColl, err := healingCollection()
c.Assert(err, check.IsNil)
defer healingColl.Close()
var events []HealingEvent
err = healingColl.Find(nil).All(&events)
c.Assert(err, check.IsNil)
c.Assert(events, check.HasLen, 1)
c.Assert(events[0].Action, check.Equals, "node-healing")
c.Assert(events[0].StartTime, check.Not(check.DeepEquals), time.Time{})
c.Assert(events[0].EndTime, check.Not(check.DeepEquals), time.Time{})
c.Assert(events[0].Error, check.Equals, "")
c.Assert(events[0].Successful, check.Equals, true)
c.Assert(events[0].FailingNode.Address, check.Equals, fmt.Sprintf("http://127.0.0.1:%d/", dockertest.URLPort(node1.URL())))
c.Assert(events[0].CreatedNode.Address, check.Equals, fmt.Sprintf("http://localhost:%d", dockertest.URLPort(node2.URL())))
}
func (s *S) TestHealerHandleError(c *check.C) {
rollback := startTestRepositoryServer()
defer rollback()
defer func() {
machines, _ := iaas.ListMachines()
for _, m := range machines {
m.Destroy()
}
}()
factory, iaasInst := newHealerIaaSConstructorWithInst("127.0.0.1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.addr = "localhost"
node1, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
node2, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", urlPort(node2.URL()))
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
cluster, err := cluster.New(nil, &cluster.MapStorage{},
cluster.Node{Address: node1.URL()},
)
c.Assert(err, check.IsNil)
var p dockerProvisioner
err = p.Initialize()
c.Assert(err, check.IsNil)
p.cluster = cluster
appInstance := provisiontest.NewFakeApp("myapp", "python", 0)
defer p.Destroy(appInstance)
p.Provision(appInstance)
imageId, err := appCurrentImageName(appInstance.GetName())
c.Assert(err, check.IsNil)
customData := map[string]interface{}{
"procfile": "web: python ./myapp",
}
err = saveImageCustomData(imageId, customData)
c.Assert(err, check.IsNil)
_, err = addContainersWithHost(&changeUnitsPipelineArgs{
toHost: "127.0.0.1",
toAdd: map[string]*containersToAdd{"web": {Quantity: 1}},
app: appInstance,
imageId: imageId,
provisioner: &p,
})
c.Assert(err, check.IsNil)
conn, err := db.Conn()
c.Assert(err, check.IsNil)
defer conn.Close()
appStruct := &app.App{
Name: appInstance.GetName(),
}
err = conn.Apps().Insert(appStruct)
c.Assert(err, check.IsNil)
defer conn.Apps().Remove(bson.M{"name": appStruct.Name})
healer := nodeHealer{
locks: make(map[string]*sync.Mutex),
provisioner: &p,
disabledTime: 0,
failuresBeforeHealing: 1,
waitTimeNewMachine: 1 * time.Second,
}
nodes, err := cluster.UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(urlPort(nodes[0].Address), check.Equals, urlPort(node1.URL()))
c.Assert(urlToHost(nodes[0].Address), check.Equals, "127.0.0.1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "127.0.0.1")
nodes[0].Metadata["iaas"] = "my-healer-iaas"
nodes[0].Metadata["Failures"] = "2"
waitTime := healer.HandleError(&nodes[0])
c.Assert(waitTime, check.Equals, time.Duration(0))
nodes, err = cluster.UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(urlPort(nodes[0].Address), check.Equals, urlPort(node2.URL()))
c.Assert(urlToHost(nodes[0].Address), check.Equals, "localhost")
machines, err = iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "localhost")
healingColl, err := healingCollection()
c.Assert(err, check.IsNil)
defer healingColl.Close()
var events []healingEvent
err = healingColl.Find(nil).All(&events)
c.Assert(err, check.IsNil)
c.Assert(events, check.HasLen, 1)
c.Assert(events[0].Action, check.Equals, "node-healing")
c.Assert(events[0].StartTime, check.Not(check.DeepEquals), time.Time{})
c.Assert(events[0].EndTime, check.Not(check.DeepEquals), time.Time{})
c.Assert(events[0].Error, check.Equals, "")
c.Assert(events[0].Successful, check.Equals, true)
c.Assert(events[0].FailingNode.Address, check.Equals, fmt.Sprintf("http://127.0.0.1:%d/", urlPort(node1.URL())))
c.Assert(events[0].CreatedNode.Address, check.Equals, fmt.Sprintf("http://localhost:%d", urlPort(node2.URL())))
}
func (s *S) TestHealerHealNode(c *check.C) {
rollback := startTestRepositoryServer()
defer rollback()
defer func() {
machines, _ := iaas.ListMachines()
for _, m := range machines {
m.Destroy()
}
}()
factory, iaasInst := newHealerIaaSConstructorWithInst("127.0.0.1")
iaas.RegisterIaasProvider("my-healer-iaas", factory)
_, err := iaas.CreateMachineForIaaS("my-healer-iaas", map[string]string{})
c.Assert(err, check.IsNil)
iaasInst.addr = "localhost"
node1, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
node2, err := testing.NewServer("127.0.0.1:0", nil, nil)
c.Assert(err, check.IsNil)
config.Set("iaas:node-protocol", "http")
config.Set("iaas:node-port", urlPort(node2.URL()))
defer config.Unset("iaas:node-protocol")
defer config.Unset("iaas:node-port")
cluster, err := cluster.New(nil, &cluster.MapStorage{},
cluster.Node{Address: node1.URL()},
)
c.Assert(err, check.IsNil)
appInstance := provisiontest.NewFakeApp("myapp", "python", 0)
var p dockerProvisioner
err = p.Initialize()
c.Assert(err, check.IsNil)
p.cluster = cluster
defer p.Destroy(appInstance)
p.Provision(appInstance)
imageId, err := appCurrentImageName(appInstance.GetName())
c.Assert(err, check.IsNil)
customData := map[string]interface{}{
"procfile": "web: python ./myapp",
}
err = saveImageCustomData(imageId, customData)
c.Assert(err, check.IsNil)
_, err = addContainersWithHost(&changeUnitsPipelineArgs{
toHost: "127.0.0.1",
toAdd: map[string]*containersToAdd{"web": {Quantity: 1}},
app: appInstance,
imageId: imageId,
provisioner: &p,
})
c.Assert(err, check.IsNil)
conn, err := db.Conn()
c.Assert(err, check.IsNil)
defer conn.Close()
appStruct := &app.App{
Name: appInstance.GetName(),
}
err = conn.Apps().Insert(appStruct)
c.Assert(err, check.IsNil)
defer conn.Apps().Remove(bson.M{"name": appStruct.Name})
healer := nodeHealer{
locks: make(map[string]*sync.Mutex),
provisioner: &p,
disabledTime: 0,
failuresBeforeHealing: 1,
waitTimeNewMachine: 1 * time.Second,
}
nodes, err := p.getCluster().UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(urlPort(nodes[0].Address), check.Equals, urlPort(node1.URL()))
c.Assert(urlToHost(nodes[0].Address), check.Equals, "127.0.0.1")
containers, err := p.listAllContainers()
c.Assert(err, check.IsNil)
c.Assert(containers, check.HasLen, 1)
c.Assert(containers[0].HostAddr, check.Equals, "127.0.0.1")
machines, err := iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "127.0.0.1")
nodes[0].Metadata["iaas"] = "my-healer-iaas"
created, err := healer.healNode(&nodes[0])
c.Assert(err, check.IsNil)
c.Assert(created.Address, check.Equals, fmt.Sprintf("http://localhost:%d", urlPort(node2.URL())))
nodes, err = cluster.UnfilteredNodes()
c.Assert(err, check.IsNil)
c.Assert(nodes, check.HasLen, 1)
c.Assert(urlPort(nodes[0].Address), check.Equals, urlPort(node2.URL()))
c.Assert(urlToHost(nodes[0].Address), check.Equals, "localhost")
machines, err = iaas.ListMachines()
c.Assert(err, check.IsNil)
c.Assert(machines, check.HasLen, 1)
c.Assert(machines[0].Address, check.Equals, "localhost")
done := make(chan bool)
go func() {
for range time.Tick(100 * time.Millisecond) {
containers, err := p.listAllContainers()
if err == nil && len(containers) == 1 && containers[0].HostAddr == "localhost" {
close(done)
return
}
}
}()
select {
case <-done:
case <-time.After(5 * time.Second):
c.Fatal("Timed out waiting for containers to move")
}
}
func (h *NodeHealer) healNode(node *cluster.Node) (cluster.Node, error) {
emptyNode := cluster.Node{}
failingAddr := node.Address
nodeMetadata := node.CleanMetadata()
failingHost := net.URLToHost(failingAddr)
failures := node.FailureCount()
machine, err := iaas.CreateMachineForIaaS(nodeMetadata["iaas"], nodeMetadata)
if err != nil {
node.ResetFailures()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error creating new machine: %s", failures, failingHost, err.Error())
}
err = h.provisioner.Cluster().Unregister(failingAddr)
if err != nil {
machine.Destroy()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error unregistering old node: %s", failures, failingHost, err.Error())
}
newAddr := machine.FormatNodeAddress()
log.Debugf("New machine created during healing process: %s - Waiting for docker to start...", newAddr)
createdNode := cluster.Node{
Address: newAddr,
Metadata: nodeMetadata,
CreationStatus: cluster.NodeCreationStatusPending,
}
err = h.provisioner.Cluster().Register(createdNode)
if err != nil {
node.ResetFailures()
h.provisioner.Cluster().Register(cluster.Node{Address: failingAddr, Metadata: nodeMetadata})
machine.Destroy()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error registering new node: %s", failures, failingHost, err.Error())
}
q, err := queue.Queue()
if err != nil {
return emptyNode, err
}
jobParams := monsterqueue.JobParams{
"endpoint": createdNode.Address,
"machine": machine.Id,
"metadata": createdNode.Metadata,
}
job, err := q.EnqueueWait(bs.QueueTaskName, jobParams, h.waitTimeNewMachine)
if err == nil {
_, err = job.Result()
}
if err != nil {
node.ResetFailures()
h.provisioner.Cluster().Register(cluster.Node{Address: failingAddr, Metadata: nodeMetadata})
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error waiting for the bs task: %s", failures, failingHost, err.Error())
}
var buf bytes.Buffer
err = h.provisioner.MoveContainers(failingHost, "", &buf)
if err != nil {
log.Errorf("Unable to move containers, skipping containers healing %q -> %q: %s: %s", failingHost, machine.Address, err.Error(), buf.String())
}
failingMachine, err := iaas.FindMachineByIdOrAddress(node.Metadata["iaas-id"], failingHost)
if err != nil {
return createdNode, fmt.Errorf("Unable to find failing machine %s in IaaS: %s", failingHost, err.Error())
}
err = failingMachine.Destroy()
if err != nil {
return createdNode, fmt.Errorf("Unable to destroy machine %s from IaaS: %s", failingHost, err.Error())
}
log.Debugf("Done auto-healing node %q, node %q created in its place.", failingHost, machine.Address)
return createdNode, nil
}