Files
fission-src/executor/executor.go
T
Ta-Ching ChenandGitHub 8b0a201f69 Fix executor tries to create same name deployment (#1082)
The root cause of the issue was introduced by PR https://github.com/fission/fission/pull/1009/files .
To be short, even the CRD of environment was delete, it still takes time for executor (poolmgr) to destroy env pool. In our cases, the previous test creates an env and delete it when test finished, then the next one creates the same name env, but failed to create pool due to the deploy name conflict. So the executor selects the pod from the first created env pool. Then, executor starts to delete the env pool, and makes the pod state became Termination state. To fix this problem, a unique name of deployment will be returned after this PR to prevent the name conflict.
2019-01-30 22:14:14 +08:00

233 lines
6.6 KiB
Go

/*
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 executor
import (
"fmt"
"log"
"net/http"
"runtime/debug"
"strings"
"sync"
"time"
"github.com/dchest/uniuri"
"github.com/pkg/errors"
"github.com/prometheus/client_golang/prometheus/promhttp"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"github.com/fission/fission"
"github.com/fission/fission/crd"
"github.com/fission/fission/executor/fscache"
"github.com/fission/fission/executor/newdeploy"
"github.com/fission/fission/executor/poolmgr"
"github.com/fission/fission/executor/reaper"
)
type (
Executor struct {
gpm *poolmgr.GenericPoolManager
ndm *newdeploy.NewDeploy
fissionClient *crd.FissionClient
fsCache *fscache.FunctionServiceCache
requestChan chan *createFuncServiceRequest
fsCreateWg map[string]*sync.WaitGroup
}
createFuncServiceRequest struct {
funcMeta *metav1.ObjectMeta
respChan chan *createFuncServiceResponse
}
createFuncServiceResponse struct {
funcSvc *fscache.FuncSvc
err error
}
)
func MakeExecutor(gpm *poolmgr.GenericPoolManager, ndm *newdeploy.NewDeploy, fissionClient *crd.FissionClient, fsCache *fscache.FunctionServiceCache) *Executor {
executor := &Executor{
gpm: gpm,
ndm: ndm,
fissionClient: fissionClient,
fsCache: fsCache,
requestChan: make(chan *createFuncServiceRequest),
fsCreateWg: make(map[string]*sync.WaitGroup),
}
go executor.serveCreateFuncServices()
return executor
}
// All non-cached function service requests go through this goroutine
// serially. It parallelizes requests for different functions, and
// ensures that for a given function, only one request causes a pod to
// get specialized. In other words, it ensures that when there's an
// ongoing request for a certain function, all other requests wait for
// that request to complete.
func (executor *Executor) serveCreateFuncServices() {
for {
req := <-executor.requestChan
m := req.funcMeta
// Cache miss -- is this first one to request the func?
wg, found := executor.fsCreateWg[crd.CacheKey(m)]
if !found {
// create a waitgroup for other requests for
// the same function to wait on
wg := &sync.WaitGroup{}
wg.Add(1)
executor.fsCreateWg[crd.CacheKey(m)] = wg
// launch a goroutine for each request, to parallelize
// the specialization of different functions
go func() {
fsvc, err := executor.createServiceForFunction(m)
req.respChan <- &createFuncServiceResponse{
funcSvc: fsvc,
err: err,
}
delete(executor.fsCreateWg, crd.CacheKey(m))
wg.Done()
}()
} else {
// There's an existing request for this function, wait for it to finish
go func() {
log.Printf("Waiting for concurrent request for the same function: %v", m)
wg.Wait()
// get the function service from the cache
fsvc, err := executor.fsCache.GetByFunction(m)
// fsCache return error when the entry does not exist/expire.
// It normally happened if there are multiple requests are
// waiting for the same function and executor failed to cre-
// ate service for function.
err = errors.Wrap(err, fmt.Sprintf("Error getting service for function %v in namespace %v", m.Name, m.Namespace))
req.respChan <- &createFuncServiceResponse{
funcSvc: fsvc,
err: err,
}
}()
}
}
}
func (executor *Executor) getFunctionExecutorType(meta *metav1.ObjectMeta) (fission.ExecutorType, error) {
fn, err := executor.fissionClient.Functions(meta.Namespace).Get(meta.Name)
if err != nil {
return "", err
}
return fn.Spec.InvokeStrategy.ExecutionStrategy.ExecutorType, nil
}
func (executor *Executor) createServiceForFunction(meta *metav1.ObjectMeta) (*fscache.FuncSvc, error) {
log.Printf("[%v] No cached function service found, creating one", meta.Name)
executorType, err := executor.getFunctionExecutorType(meta)
if err != nil {
return nil, err
}
var fsvc *fscache.FuncSvc
var fsvcErr error
switch executorType {
case fission.ExecutorTypeNewdeploy:
fsvc, fsvcErr = executor.ndm.GetFuncSvc(meta)
default:
fsvc, fsvcErr = executor.gpm.GetFuncSvc(meta)
}
if fsvcErr != nil {
fsvcErr = errors.Wrap(fsvcErr, fmt.Sprintf("[%v] Error creating service for function", meta.Name))
log.Print(fsvcErr)
} else if fsvc != nil {
_, err = executor.fsCache.Add(*fsvc)
if err != nil {
return nil, err
}
}
executor.fsCache.IncreaseColdStarts(meta.Name, string(meta.UID))
return fsvc, fsvcErr
}
// isValidAddress invokes isValidService or isValidPod depending on the type of executor
func (executor *Executor) isValidAddress(fsvc *fscache.FuncSvc) bool {
if fsvc.Executor == fscache.NEWDEPLOY {
return executor.ndm.IsValid(fsvc)
} else {
return executor.gpm.IsValid(fsvc)
}
}
func dumpStackTrace() {
debug.PrintStack()
}
func serveMetric() {
// Expose the registered metrics via HTTP.
metricAddr := ":8080"
http.Handle("/metrics", promhttp.Handler())
log.Fatal(http.ListenAndServe(metricAddr, nil))
}
// StartExecutor Starts executor and the executor components such as Poolmgr,
// deploymgr and potential future executor types
func StartExecutor(fissionNamespace string, functionNamespace string, envBuilderNamespace string, port int) error {
// setup a signal handler for SIGTERM
fission.SetupStackTraceHandler()
fissionClient, kubernetesClient, _, err := crd.MakeFissionClient()
err = fissionClient.WaitForCRDs()
if err != nil {
log.Fatalf("Error waiting for CRDs: %v", err)
}
restClient := fissionClient.GetCrdClient()
if err != nil {
log.Printf("Failed to get kubernetes client: %v", err)
return err
}
fsCache := fscache.MakeFunctionServiceCache()
poolID := strings.ToLower(uniuri.NewLen(8))
reaper.CleanupOldExecutorObjects(kubernetesClient, poolID)
go reaper.CleanupRoleBindings(kubernetesClient, fissionClient, functionNamespace, envBuilderNamespace, time.Minute*30)
gpm := poolmgr.MakeGenericPoolManager(
fissionClient, kubernetesClient,
functionNamespace, poolID)
ndm := newdeploy.MakeNewDeploy(
fissionClient, kubernetesClient, restClient,
functionNamespace, poolID)
api := MakeExecutor(gpm, ndm, fissionClient, fsCache)
go api.Serve(port)
go serveMetric()
return nil
}