feat(manager): make check status from flink batch

This commit is contained in:
2024-12-07 01:09:25 +03:30
parent d1a715deea
commit 2c25323e62
12 changed files with 109 additions and 41 deletions

View File

@@ -41,11 +41,12 @@ func (crd Crd) SetJobStatus(jobUid types.UID, status string) error {
)
return err
}
patched, err := convertFromUnstructured(unstructuredJob)
_, err = convertFromUnstructured(unstructuredJob)
if err != nil {
lc.Logger.Error("[crd] [status] error in structure unstructured patched", zap.Error(err))
} else {
lc.Logger.Debug("[crd] [status] patched")
}
lc.Logger.Debug("[crd] [status] set status", zap.Any("statusUpdateObj", patched))
if err != nil {
lc.Logger.Error("[crd] [status] ", zap.Error(err))
return err

View File

@@ -1,33 +1,32 @@
package managed_job
import (
"errors"
"time"
"gitea.com/logicamp/lc"
"go.uber.org/zap"
)
func (job *ManagedJob) startCycle() {
ticker := time.NewTicker(5 * time.Second)
quit := make(chan struct{})
// func (job *ManagedJob) startCycle() {
// ticker := time.NewTicker(5 * time.Second)
// quit := make(chan struct{})
// load job state from db
job.loadState()
go func() {
for {
select {
case <-ticker.C:
job.cycle()
case <-quit:
ticker.Stop()
return
}
}
}()
}
// // load job state from db
// job.loadState()
// go func() {
// for {
// select {
// case <-ticker.C:
// job.cycle()
// case <-quit:
// ticker.Stop()
// return
// }
// }
// }()
// }
func (job *ManagedJob) cycle() {
func (job *ManagedJob) Cycle() {
lc.Logger.Debug("[managed-job] [new] check cycle", zap.String("jobKey", string(job.def.UID)))
// Init job
@@ -47,20 +46,15 @@ func (job *ManagedJob) cycle() {
job.crd.SetJobStatus(job.def.UID, string(job.state.Status))
// Check for set running or error state
if job.state.Status == JobStatusCreating || job.state.Status == JobStatusFailing {
/* if job.state.Status == JobStatusCreating || job.state.Status == JobStatusFailing {
err := job.checkStatus()
if errors.Is(err, ErrNoJobId) {
job.state = nil
}
return
}
} */
if job.state.Status == JobStatusRunning {
err := job.checkStatus()
if errors.Is(err, ErrNoJobId) {
job.state = nil
}
lc.Logger.Debug("savepoint interval", zap.Any("savepoint duration", job.def.Spec.SavepointInterval))
if (job.def.Spec.SavepointInterval.Duration != 0) && ((job.state.LastSavepointDate == nil) || time.Now().Add(-job.def.Spec.SavepointInterval.Duration).After(*job.state.LastSavepointDate)) {
if job.state.SavepointTriggerId == nil {
job.createSavepoint()

View File

@@ -24,7 +24,7 @@ func NewManagedJob(client *api.Client, db *buntdb.DB, def v1alpha1.FlinkJob, crd
db: db,
crd: crd,
}
job.startCycle()
//job.startCycle()
return job
}

View File

@@ -40,7 +40,13 @@ func (job *ManagedJob) run() error {
}
lc.Logger.Debug("[main] after run jar", zap.Any("run-jar-resp", runJarResp))
job.updateState(jobState{JobId: &runJarResp.JobId, Status: JobStatusCreating})
if job.state == nil {
job.state = &jobState{}
}
job.state.JobId = &runJarResp.JobId
job.state.Status = JobStatusCreating
job.updateState(*job.state)
//job.updateState(jobState{JobId: &runJarResp.JobId, Status: JobStatusCreating})
return err
}

View File

@@ -57,7 +57,15 @@ func (job *ManagedJob) removeSavepointTriggerId() {
job.updateState(*job.state)
}
func (job *ManagedJob) setStatus(status JobStatus) {
func (job *ManagedJob) SetStatus(status JobStatus) {
job.state.Status = status
job.updateState(*job.state)
}
func (job *ManagedJob) GetJobId() *string {
if job.state != nil && job.state.JobId != nil {
return job.state.JobId
} else {
return nil
}
}

View File

@@ -24,6 +24,6 @@ func (job *ManagedJob) checkStatus() error {
return err
}
//lc.Logger.Debug("[managed-job] [status]", zap.Any("status-resp", statusResp))
job.setStatus(JobStatus(statusResp.State))
job.SetStatus(JobStatus(statusResp.State))
return err
}

View File

@@ -0,0 +1,8 @@
package manager
import api "github.com/logi-camp/go-flink-client"
func (mgr *Manager) checkJobStatus(client *api.Client) error {
return nil
}

View File

@@ -5,39 +5,69 @@ import (
"flink-kube-operator/internal/managed_job"
"time"
"gitea.com/logicamp/lc"
api "github.com/logi-camp/go-flink-client"
"github.com/samber/lo"
"github.com/tidwall/buntdb"
"go.uber.org/zap"
"k8s.io/apimachinery/pkg/types"
)
var managedJobs = map[types.UID]managed_job.ManagedJob{}
type Manager struct {
client *api.Client
managedJobs map[types.UID]managed_job.ManagedJob
}
func Setup(client *api.Client, db *buntdb.DB, crdInstance *crd.Crd) {
func NewManager(client *api.Client, db *buntdb.DB, crdInstance *crd.Crd) Manager {
ticker := time.NewTicker(5 * time.Second)
quit := make(chan struct{})
mgr := Manager{
client: client,
managedJobs: map[types.UID]managed_job.ManagedJob{},
}
go func() {
for {
select {
case <-ticker.C:
cycle(client, db, crdInstance)
mgr.cycle(client, db, crdInstance)
case <-quit:
ticker.Stop()
return
}
}
}()
return mgr
}
func cycle(client *api.Client, db *buntdb.DB, crdInstance *crd.Crd) {
func (mgr *Manager) cycle(client *api.Client, db *buntdb.DB, crdInstance *crd.Crd) {
jobsOverviews, err := mgr.client.JobsOverview()
if err != nil {
lc.Logger.Error("[manager] [cycle] cannot check flink jobs status", zap.Error(err))
}
//lc.Logger.Debug("[manager] [cycle] overviews", zap.Any("overviews", jobsOverviews))
for _, uid := range crd.GetAllJobKeys() {
def := crd.GetJob(uid)
managedJob, ok := managedJobs[uid]
managedJob, ok := mgr.managedJobs[uid]
if ok {
managedJob.Update(def)
} else {
managedJob := managed_job.NewManagedJob(client, db, def, crdInstance)
managedJobs[uid] = *managedJob
managedJob = *managed_job.NewManagedJob(client, db, def, crdInstance)
//mgr.managedJobs[uid] = managedJob
}
jobOverview, ok := lo.Find(jobsOverviews.Jobs, func(job api.JobOverview) bool {
jobId := managedJob.GetJobId()
if jobId != nil {
return job.ID == *jobId
}
return false
})
if ok {
lc.Logger.Debug("[manager] read status from flink", zap.String("name", jobOverview.Name), zap.String("state", jobOverview.State))
managedJob.SetStatus(managed_job.JobStatus(jobOverview.State))
}
managedJob.Cycle()
mgr.managedJobs[uid] = managedJob
}
}