Nail/fission-src
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #69 from fission/poolmgr-gc

Browse Source 
Poolmgr: ensure orphaned resources are cleaned up
This commit is contained in:
Soam Vasani
2017-01-06 17:25:06 -08:00
committed by GitHub
parent 595e634617 37a00b6116
commit d5516686c8
4 changed files with 185 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
poolmgr/cleanup.go
+147
View File
@@ -0,0 +1,147 @@
/*
Copyright 2016 The Fission Authors.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
package poolmgr
import (
"log"
"time"
"k8s.io/client-go/1.5/kubernetes"
"k8s.io/client-go/1.5/pkg/api"
)
// cleanupOldPoolmgrResources looks for resources created by an old
// poolmgr instance and cleans them up.
func cleanupOldPoolmgrResources(client *kubernetes.Clientset, namespace string, instanceId string) {
go func() {
err := cleanup(client, namespace, instanceId)
if err != nil {
// TODO retry cleanup; logged and ignored for now
log.Printf("Failed to cleanup: %v", err)
}
}()
}
func cleanup(client *kubernetes.Clientset, namespace string, instanceId string) error {
// Deployments are used for idle pools and can be cleaned up
// immediately. (We should "adopt" these instead of creating
// a new pool.)
err := cleanupDeployments(client, namespace, instanceId)
if err != nil {
return err
}
// See K8s #33845 and related bugs: deleting a deployment
// through the API doesn't cause the associated ReplicaSet to
// be deleted. (Fixed recently, but we may be running a
// version before the fix.)
err = cleanupReplicaSets(client, namespace, instanceId)
if err != nil {
return err
}
// Pods might still be running user functions, so we give them
// a few minutes before terminating them. This time is the
// maximum function runtime, plus the time a router might
// still route to an old instance, i.e. router cache expiry
// time.
time.Sleep(6 * time.Minute)
err = cleanupPods(client, namespace, instanceId)
if err != nil {
return err
}
err = cleanupServices(client, namespace, instanceId)
if err != nil {
return err
}
return nil
}
func cleanupDeployments(client *kubernetes.Clientset, namespace string, instanceId string) error {
deploymentList, err := client.Extensions().Deployments(namespace).List(api.ListOptions{})
if err != nil {
return err
}
for _, dep := range deploymentList.Items {
id, ok := dep.ObjectMeta.Labels[POOLMGR_INSTANCEID_LABEL]
if ok && id != instanceId {
log.Printf("Cleaning up deployment %v", dep.ObjectMeta.Name)
err := client.Extensions().Deployments(namespace).Delete(dep.ObjectMeta.Name, nil)
logErr("cleaning up deployment", err)
// ignore err
}
}
return nil
}
func cleanupReplicaSets(client *kubernetes.Clientset, namespace string, instanceId string) error {
rsList, err := client.Extensions().ReplicaSets(namespace).List(api.ListOptions{})
if err != nil {
return err
}
for _, rs := range rsList.Items {
id, ok := rs.ObjectMeta.Labels[POOLMGR_INSTANCEID_LABEL]
if ok && id != instanceId {
log.Printf("Cleaning up replicaset %v", rs.ObjectMeta.Name)
err := client.Extensions().ReplicaSets(namespace).Delete(rs.ObjectMeta.Name, nil)
logErr("cleaning up replicaset", err)
}
}
return nil
}
func cleanupPods(client *kubernetes.Clientset, namespace string, instanceId string) error {
podList, err := client.Core().Pods(namespace).List(api.ListOptions{})
if err != nil {
return err
}
for _, pod := range podList.Items {
id, ok := pod.ObjectMeta.Labels[POOLMGR_INSTANCEID_LABEL]
if ok && id != instanceId {
log.Printf("Cleaning up pod %v", pod.ObjectMeta.Name)
err := client.Core().Pods(namespace).Delete(pod.ObjectMeta.Name, nil)
logErr("cleaning up pod", err)
// ignore err
}
}
return nil
}
func cleanupServices(client *kubernetes.Clientset, namespace string, instanceId string) error {
svcList, err := client.Core().Services(namespace).List(api.ListOptions{})
if err != nil {
return err
}
for _, svc := range svcList.Items {
id, ok := svc.ObjectMeta.Labels[POOLMGR_INSTANCEID_LABEL]
if ok && id != instanceId {
log.Printf("Cleaning up svc %v", svc.ObjectMeta.Name)
err := client.Core().Services(namespace).Delete(svc.ObjectMeta.Name, nil)
logErr("cleaning up svc", err)
// ignore err
}
}
return nil
}
func logErr(msg string, err error) {
if err != nil {
log.Printf("Error %v: %v", msg, err)
}
}
poolmgr/gp.go
+19 -10
View File
@@ -39,6 +39,8 @@ import (
"github.com/fission/fission"
)
const POOLMGR_INSTANCEID_LABEL string = "poolmgrInstanceId"
type (
GenericPool struct {
env *fission.Environment
@@ -52,7 +54,9 @@ type (
useSvc bool // create service
poolInstanceId string // small random string to uniquify pod names
kubernetesClient *kubernetes.Clientset
requestChannel chan *choosePodRequest
instanceId string
requestChannel chan *choosePodRequest
}
// serialize the choosing of pods so that choices don't conflict
@@ -72,7 +76,8 @@ func MakeGenericPool(
env *fission.Environment,
initialReplicas int32,
namespace string,
fsCache *functionServiceCache) (*GenericPool, error) {
fsCache *functionServiceCache,
instanceId string) (*GenericPool, error) {
log.Printf("Creating pool for environment %v", env.Metadata)
// TODO: in general we need to provide the user a way to configure pools. Initial
@@ -88,6 +93,7 @@ func MakeGenericPool(
idlePodReapTime: 3 * time.Minute, // TODO make this configurable
fsCache: fsCache,
poolInstanceId: uniuri.NewLen(8),
instanceId: instanceId,
useSvc: false,
}
@@ -200,11 +206,12 @@ func (gp *GenericPool) _choosePod(newLabels map[string]string) (*v1.Pod, error)
}
}
func labelsForMetadata(metadata *fission.Metadata) map[string]string {
func (gp *GenericPool) labelsForMetadata(metadata *fission.Metadata) map[string]string {
return map[string]string{
"functionName": metadata.Name,
"functionUid": metadata.Uid,
"unmanaged": "true", // this allows us to easily find pods not managed by the deployment
"functionName": metadata.Name,
"functionUid": metadata.Uid,
"unmanaged": "true", // this allows us to easily find pods not managed by the deployment
POOLMGR_INSTANCEID_LABEL: gp.instanceId,
}
}
@@ -212,7 +219,7 @@ func labelsForMetadata(metadata *fission.Metadata) map[string]string {
// (via fetcher), and calls the function-run container to load it, resulting in a
// specialized pod.
func (gp *GenericPool) specializePod(metadata *fission.Metadata) (*v1.Pod, error) {
newLabels := labelsForMetadata(metadata)
newLabels := gp.labelsForMetadata(metadata)
log.Printf("[%v] Choosing pod from pool", metadata)
pod, err := gp.choosePod(newLabels)
@@ -287,6 +294,7 @@ func (gp *GenericPool) createPool() error {
podLabels := map[string]string{
"pool": poolDeploymentName,
POOLMGR_INSTANCEID_LABEL: gp.instanceId,
}
sharedMountPath := "/userfunc"
@@ -294,8 +302,9 @@ func (gp *GenericPool) createPool() error {
ObjectMeta: v1.ObjectMeta{
Name: poolDeploymentName,
Labels: map[string]string{
"environmentName": gp.env.Metadata.Name,
"environmentUid": gp.env.Metadata.Uid,
"environmentName": gp.env.Metadata.Name,
"environmentUid": gp.env.Metadata.Uid,
POOLMGR_INSTANCEID_LABEL: gp.instanceId,
},
},
Spec: v1beta1.DeploymentSpec{
@@ -411,7 +420,7 @@ func (gp *GenericPool) GetFuncSvc(m *fission.Metadata) (*funcSvc, error) {
svcName += ("-" + m.Uid)
}
labels := labelsForMetadata(m)
labels := gp.labelsForMetadata(m)
svc, err := gp.createSvc(svcName, labels)
if err != nil {
return nil, err
poolmgr/gpm.go
+14 -4
View File
@@ -34,8 +34,8 @@ type (
controllerUrl string
controllerClient *client.Client
fsCache *functionServiceCache
requestChannel chan *request
instanceId string
requestChannel chan *request
}
request struct {
env *fission.Environment
@@ -47,7 +47,13 @@ type (
}
)
func MakeGenericPoolManager(controllerUrl string, kubernetesClient *kubernetes.Clientset, namespace string, fsCache *functionServiceCache) *GenericPoolManager {
func MakeGenericPoolManager(
controllerUrl string,
kubernetesClient *kubernetes.Clientset,
namespace string,
fsCache *functionServiceCache,
instanceId string) *GenericPoolManager {
gpm := &GenericPoolManager{
pools: make(map[fission.Environment]*GenericPool),
kubernetesClient: kubernetesClient,
@@ -55,6 +61,7 @@ func MakeGenericPoolManager(controllerUrl string, kubernetesClient *kubernetes.C
controllerUrl: controllerUrl,
controllerClient: client.MakeClient(controllerUrl),
fsCache: fsCache,
instanceId: instanceId,
requestChannel: make(chan *request),
}
go gpm.service()
@@ -70,7 +77,10 @@ func (gpm *GenericPoolManager) service() {
var err error
pool, ok := gpm.pools[*req.env]
if !ok {
pool, err = MakeGenericPool(gpm.controllerUrl, gpm.kubernetesClient, req.env, 3, gpm.namespace, gpm.fsCache)
pool, err = MakeGenericPool(
gpm.controllerUrl, gpm.kubernetesClient, req.env,
3, // TODO configurable/autoscalable
gpm.namespace, gpm.fsCache, gpm.instanceId)
if err != nil {
req.responseChannel <- &response{error: err}
continue
poolmgr/poolmgr.go
+5 -2
View File
@@ -20,6 +20,7 @@ import (
"log"
"strings"
"github.com/dchest/uniuri"
"k8s.io/client-go/1.5/kubernetes"
"k8s.io/client-go/1.5/rest"
@@ -56,11 +57,13 @@ func StartPoolmgr(controllerUrl string, namespace string, port int) error {
return err
}
instanceId := uniuri.NewLen(8)
cleanupOldPoolmgrResources(kubernetesClient, namespace, instanceId)
fsCache := MakeFunctionServiceCache()
gpm := MakeGenericPoolManager(controllerUrl, kubernetesClient, namespace, fsCache, instanceId)
gpm := MakeGenericPoolManager(controllerUrl, kubernetesClient, namespace, fsCache)
api := MakeAPI(gpm, controllerClient, fsCache)
go api.Serve(port)
return nil