feat: remove buntdb dep
This commit is contained in:
@@ -1,6 +1,7 @@
|
||||
package managed_job
|
||||
|
||||
import (
|
||||
"flink-kube-operator/internal/crd/v1alpha1"
|
||||
"time"
|
||||
|
||||
"gitea.com/logicamp/lc"
|
||||
@@ -30,20 +31,30 @@ func (job *ManagedJob) Cycle() {
|
||||
lc.Logger.Debug("[managed-job] [new] check cycle", zap.String("jobKey", string(job.def.UID)))
|
||||
|
||||
// Init job
|
||||
if job.state == nil {
|
||||
if job.def.Status.JobStatus == "" {
|
||||
err := job.upload()
|
||||
if err != nil {
|
||||
job.setError("[upload-error] " + err.Error())
|
||||
job.crd.Patch(job.def.UID, map[string]interface{}{
|
||||
"status": map[string]interface{}{
|
||||
"error": "[upload-error] " + err.Error(),
|
||||
},
|
||||
})
|
||||
return
|
||||
}
|
||||
err = job.run()
|
||||
if err != nil {
|
||||
job.setError("[run-error] " + err.Error())
|
||||
job.crd.Patch(job.def.UID, map[string]interface{}{
|
||||
"status": map[string]interface{}{
|
||||
"error": "[run-error] " + err.Error(),
|
||||
},
|
||||
})
|
||||
return
|
||||
}
|
||||
return
|
||||
}
|
||||
job.crd.SetJobStatus(job.def.UID, string(job.state.Status))
|
||||
// job.crd.SetJobStatus(job.def.UID, v1alpha1.FlinkJobStatus{
|
||||
// JobStatus: job.def.Status.JobStatus,
|
||||
// })
|
||||
|
||||
// Check for set running or error state
|
||||
/* if job.state.Status == JobStatusCreating || job.state.Status == JobStatusFailing {
|
||||
@@ -54,9 +65,9 @@ func (job *ManagedJob) Cycle() {
|
||||
return
|
||||
} */
|
||||
|
||||
if job.state.Status == JobStatusRunning {
|
||||
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 {
|
||||
if job.def.Status.JobStatus == v1alpha1.JobStatusRunning {
|
||||
if (job.def.Spec.SavepointInterval.Duration != 0) && ((job.def.Status.LastSavepointDate == nil) || time.Now().Add(-job.def.Spec.SavepointInterval.Duration).After(*job.def.Status.LastSavepointDate)) {
|
||||
if job.def.Status.SavepointTriggerId == nil {
|
||||
job.createSavepoint()
|
||||
} else {
|
||||
job.trackSavepoint()
|
||||
@@ -64,10 +75,10 @@ func (job *ManagedJob) Cycle() {
|
||||
}
|
||||
return
|
||||
}
|
||||
if job.state.Status == JobStatusFailed {
|
||||
if job.def.Status.JobStatus == v1alpha1.JobStatusFailed && job.def.Status.LastSavepointPath != nil {
|
||||
job.restore()
|
||||
return
|
||||
}
|
||||
|
||||
lc.Logger.Warn("[managed-job] [cycle]", zap.String("unhanded job status", string(job.state.Status)))
|
||||
lc.Logger.Warn("[managed-job] [cycle]", zap.String("unhanded job status", string(job.def.Status.JobStatus)))
|
||||
}
|
||||
|
||||
@@ -5,23 +5,19 @@ import (
|
||||
"flink-kube-operator/internal/crd/v1alpha1"
|
||||
|
||||
api "github.com/logi-camp/go-flink-client"
|
||||
"github.com/tidwall/buntdb"
|
||||
)
|
||||
|
||||
type ManagedJob struct {
|
||||
def v1alpha1.FlinkJob
|
||||
client *api.Client
|
||||
jarId string
|
||||
db *buntdb.DB
|
||||
state *jobState
|
||||
crd *crd.Crd
|
||||
}
|
||||
|
||||
func NewManagedJob(client *api.Client, db *buntdb.DB, def v1alpha1.FlinkJob, crd *crd.Crd) *ManagedJob {
|
||||
func NewManagedJob(client *api.Client, def v1alpha1.FlinkJob, crd *crd.Crd) *ManagedJob {
|
||||
job := &ManagedJob{
|
||||
def: def,
|
||||
client: client,
|
||||
db: db,
|
||||
crd: crd,
|
||||
}
|
||||
//job.startCycle()
|
||||
|
||||
@@ -1,5 +1,5 @@
|
||||
package managed_job
|
||||
|
||||
func (job *ManagedJob) Stop() {
|
||||
job.client.StopJob(*job.state.JobId)
|
||||
job.client.StopJob(*job.def.Status.JobId)
|
||||
}
|
||||
|
||||
@@ -1,6 +1,8 @@
|
||||
package managed_job
|
||||
|
||||
import (
|
||||
"flink-kube-operator/internal/crd/v1alpha1"
|
||||
|
||||
"gitea.com/logicamp/lc"
|
||||
api "github.com/logi-camp/go-flink-client"
|
||||
"go.uber.org/zap"
|
||||
@@ -8,16 +10,16 @@ import (
|
||||
|
||||
// restore the job from savepoint and jarId in managedJob
|
||||
func (job *ManagedJob) restore() error {
|
||||
if job.state.LastSavepointPath == nil {
|
||||
lc.Logger.Error("[managed-job] [restore]", zap.Error(ErrNoSavepointPath))
|
||||
return ErrNoSavepointPath
|
||||
if job.def.Status.LastSavepointPath == nil {
|
||||
lc.Logger.Error("[managed-job] [restore]", zap.Error(v1alpha1.ErrNoSavepointPath))
|
||||
return v1alpha1.ErrNoSavepointPath
|
||||
}
|
||||
lc.Logger.Debug("[managed-job] [restore] restoring", zap.String("savepointPath", *job.state.LastSavepointPath))
|
||||
lc.Logger.Debug("[managed-job] [restore] restoring", zap.String("savepointPath", *job.def.Status.LastSavepointPath))
|
||||
runJarResp, err := job.client.RunJar(api.RunOpts{
|
||||
JarID: job.jarId,
|
||||
AllowNonRestoredState: true,
|
||||
EntryClass: job.def.Spec.EntryClass,
|
||||
SavepointPath: *job.state.LastSavepointPath,
|
||||
SavepointPath: *job.def.Status.LastSavepointPath,
|
||||
})
|
||||
if err != nil {
|
||||
lc.Logger.Error("[managed-job] [run]", zap.Error(err))
|
||||
@@ -25,10 +27,17 @@ func (job *ManagedJob) restore() error {
|
||||
}
|
||||
lc.Logger.Debug("[main] after run jar", zap.Any("run-jar-resp", runJarResp))
|
||||
|
||||
job.state.JobId = &runJarResp.JobId
|
||||
job.state.Status = JobStatusCreating
|
||||
job.state.Error = nil
|
||||
job.updateState(*job.state)
|
||||
// job.def.Status.JobId = &runJarResp.JobId
|
||||
// job.def.Status.JobStatus = v1alpha1.JobStatusCreating
|
||||
// job.def.Status.Error = nil
|
||||
job.crd.Patch(job.def.UID, map[string]interface{}{
|
||||
"status": map[string]interface{}{
|
||||
"jobId": &runJarResp.JobId,
|
||||
"jobStatus": v1alpha1.JobStatusCreating,
|
||||
"lifeCycleStatus": v1alpha1.LifeCycleStatusRestoring,
|
||||
"error": nil,
|
||||
},
|
||||
})
|
||||
|
||||
return err
|
||||
}
|
||||
|
||||
@@ -1,6 +1,7 @@
|
||||
package managed_job
|
||||
|
||||
import (
|
||||
"flink-kube-operator/internal/crd/v1alpha1"
|
||||
"flink-kube-operator/internal/jar"
|
||||
|
||||
"gitea.com/logicamp/lc"
|
||||
@@ -40,12 +41,23 @@ func (job *ManagedJob) run() error {
|
||||
}
|
||||
lc.Logger.Debug("[main] after run jar", zap.Any("run-jar-resp", runJarResp))
|
||||
|
||||
if job.state == nil {
|
||||
job.state = &jobState{}
|
||||
}
|
||||
job.state.JobId = &runJarResp.JobId
|
||||
job.state.Status = JobStatusCreating
|
||||
job.updateState(*job.state)
|
||||
// if job.state == nil {
|
||||
// job.state = &jobState{}
|
||||
// }
|
||||
// job.state.JobId = &runJarResp.JobId
|
||||
// job.state.Status = v1alpha1.JobStatusCreating
|
||||
// job.updateState(*job.state)
|
||||
// job.crd.SetJobStatus(job.def.UID, v1alpha1.FlinkJobStatus{
|
||||
// JobId: job.state.JobId,
|
||||
// })
|
||||
job.crd.Patch(job.def.UID, map[string]interface{}{
|
||||
"status": map[string]interface{}{
|
||||
"jobId": &runJarResp.JobId,
|
||||
"jobStatus": v1alpha1.JobStatusCreating,
|
||||
"lifeCycleStatus": v1alpha1.LifeCycleStatusInitializing,
|
||||
"error": nil,
|
||||
},
|
||||
})
|
||||
//job.updateState(jobState{JobId: &runJarResp.JobId, Status: JobStatusCreating})
|
||||
|
||||
return err
|
||||
|
||||
@@ -1,7 +1,9 @@
|
||||
package managed_job
|
||||
|
||||
import (
|
||||
"flink-kube-operator/internal/crd/v1alpha1"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"gitea.com/logicamp/lc"
|
||||
api "github.com/logi-camp/go-flink-client"
|
||||
@@ -9,39 +11,53 @@ import (
|
||||
)
|
||||
|
||||
func (job ManagedJob) createSavepoint() error {
|
||||
if job.state.JobId == nil {
|
||||
if job.def.Status.JobId == nil {
|
||||
lc.Logger.Debug("[managed-job] [savepoint] no job id")
|
||||
return ErrNoJobId
|
||||
return v1alpha1.ErrNoJobId
|
||||
}
|
||||
lc.Logger.Info("[managed-job] [savepoint] creating savepoint", zap.String("interval", job.def.Spec.SavepointInterval.String()))
|
||||
resp, err := job.client.SavePoints(*job.state.JobId, "/flink-data/savepoints-2/", false)
|
||||
resp, err := job.client.SavePoints(*job.def.Status.JobId, "/flink-data/savepoints-2/", false)
|
||||
if err != nil {
|
||||
lc.Logger.Error("[managed-job] [savepoint] error in creating savepoint", zap.Error(err))
|
||||
return err
|
||||
}
|
||||
lc.Logger.Debug("[managed-job] [savepoint]", zap.Any("savepoint-resp", resp))
|
||||
job.setSavepointTriggerId(resp.RequestID)
|
||||
|
||||
job.crd.Patch(job.def.UID, map[string]interface{}{
|
||||
"status": map[string]interface{}{
|
||||
"savepointTriggerId": resp.RequestID,
|
||||
},
|
||||
})
|
||||
return nil
|
||||
}
|
||||
|
||||
func (job ManagedJob) trackSavepoint() error {
|
||||
if job.state.JobId == nil {
|
||||
if job.def.Status.JobId == nil {
|
||||
lc.Logger.Debug("[managed-job] [savepoint] no job id")
|
||||
return ErrNoJobId
|
||||
return v1alpha1.ErrNoJobId
|
||||
}
|
||||
if job.state.SavepointTriggerId == nil {
|
||||
if job.def.Status.SavepointTriggerId == nil {
|
||||
lc.Logger.Debug("[managed-job] [savepoint] no job id")
|
||||
return ErrNoSavepointTriggerId
|
||||
return v1alpha1.ErrNoSavepointTriggerId
|
||||
}
|
||||
resp, err := job.client.TrackSavepoint(*job.state.JobId, *job.state.SavepointTriggerId)
|
||||
resp, err := job.client.TrackSavepoint(*job.def.Status.JobId, *job.def.Status.SavepointTriggerId)
|
||||
lc.Logger.Debug("[managed-job] [savepoint] track savepoint", zap.Any("status.Id", resp.Status.Id), zap.Any("failureCause.stacktrace", resp.Operation.FailureCause.StackTrace), zap.Error(err))
|
||||
if err != nil {
|
||||
if strings.IndexAny(err.Error(), "http status not 2xx: 404") == 0 {
|
||||
job.removeSavepointTriggerId()
|
||||
job.crd.Patch(job.def.UID, map[string]interface{}{
|
||||
"status": map[string]interface{}{
|
||||
"savepointTriggerId": nil,
|
||||
},
|
||||
})
|
||||
}
|
||||
}
|
||||
if resp.Status.Id == api.SavepointStatusInCompleted {
|
||||
job.setSavepointLocation(resp.Operation.Location)
|
||||
job.crd.Patch(job.def.UID, map[string]interface{}{
|
||||
"status": map[string]interface{}{
|
||||
"lastSavepointPath": resp.Operation.Location,
|
||||
"lastSavepointDate": time.Now(),
|
||||
},
|
||||
})
|
||||
}
|
||||
|
||||
return nil
|
||||
|
||||
@@ -1,70 +1,77 @@
|
||||
package managed_job
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"time"
|
||||
// import (
|
||||
// "encoding/json"
|
||||
// "flink-kube-operator/internal/crd/v1alpha1"
|
||||
// "time"
|
||||
|
||||
"github.com/tidwall/buntdb"
|
||||
)
|
||||
// "github.com/tidwall/buntdb"
|
||||
// )
|
||||
|
||||
// get state of job from local db
|
||||
func (job *ManagedJob) loadState() {
|
||||
job.db.View(
|
||||
func(tx *buntdb.Tx) error {
|
||||
if val, err := tx.Get(string(job.def.GetUID())); err != nil {
|
||||
return err
|
||||
} else {
|
||||
return json.Unmarshal([]byte(val), job.state)
|
||||
}
|
||||
})
|
||||
}
|
||||
// // get state of job from local db
|
||||
// func (job *ManagedJob) loadState() {
|
||||
// job.db.View(
|
||||
// func(tx *buntdb.Tx) error {
|
||||
// if val, err := tx.Get(string(job.def.GetUID())); err != nil {
|
||||
// return err
|
||||
// } else {
|
||||
// return json.Unmarshal([]byte(val), job.state)
|
||||
// }
|
||||
// })
|
||||
// }
|
||||
|
||||
// save state of job to local db
|
||||
func (job *ManagedJob) updateState(state jobState) {
|
||||
job.state = &state
|
||||
// // save state of job to local db
|
||||
// func (job *ManagedJob) updateState(state jobState) {
|
||||
// job.state = &state
|
||||
|
||||
value, _ := json.Marshal(job.state)
|
||||
job.db.Update(func(tx *buntdb.Tx) error {
|
||||
_, _, err := tx.Set(string(job.def.GetUID()), string(value), nil)
|
||||
return err
|
||||
})
|
||||
}
|
||||
// value, _ := json.Marshal(job.state)
|
||||
// job.db.Update(func(tx *buntdb.Tx) error {
|
||||
// _, _, err := tx.Set(string(job.def.GetUID()), string(value), nil)
|
||||
// return err
|
||||
// })
|
||||
// }
|
||||
|
||||
func (job *ManagedJob) setError(errMsg string) {
|
||||
if job.state == nil {
|
||||
job.state = &jobState{}
|
||||
}
|
||||
job.state.Error = &errMsg
|
||||
job.state.Status = JobStatusError
|
||||
job.updateState(*job.state)
|
||||
}
|
||||
// func (job *ManagedJob) setError(errMsg string) {
|
||||
// if job.state == nil {
|
||||
// job.state = &jobState{}
|
||||
// }
|
||||
// job.state.Error = &errMsg
|
||||
// job.state.Status = v1alpha1.JobStatusError
|
||||
// job.updateState(*job.state)
|
||||
// job.crd.SetJobStatus(job.def.UID, v1alpha1.FlinkJobStatus{
|
||||
// JobStatus: job.state.Status,
|
||||
// })
|
||||
// }
|
||||
|
||||
func (job *ManagedJob) setSavepointLocation(savepointId string) {
|
||||
job.state.LastSavepointPath = &savepointId
|
||||
job.state.SavepointTriggerId = nil
|
||||
n := time.Now()
|
||||
job.state.LastSavepointDate = &n
|
||||
job.updateState(*job.state)
|
||||
}
|
||||
// func (job *ManagedJob) setSavepointLocation(savepointId string) {
|
||||
// job.state.LastSavepointPath = &savepointId
|
||||
// job.state.SavepointTriggerId = nil
|
||||
// n := time.Now()
|
||||
// job.state.LastSavepointDate = &n
|
||||
// job.updateState(*job.state)
|
||||
// job.crd.SetJobStatus(job.def.UID, v1alpha1.FlinkJobStatus{
|
||||
// LastSavepointPath: job.state.LastSavepointPath,
|
||||
// })
|
||||
// }
|
||||
|
||||
func (job *ManagedJob) setSavepointTriggerId(savepointReqId string) {
|
||||
job.state.SavepointTriggerId = &savepointReqId
|
||||
job.updateState(*job.state)
|
||||
}
|
||||
// func (job *ManagedJob) setSavepointTriggerId(savepointReqId string) {
|
||||
// job.state.SavepointTriggerId = &savepointReqId
|
||||
// job.updateState(*job.state)
|
||||
// }
|
||||
|
||||
func (job *ManagedJob) removeSavepointTriggerId() {
|
||||
job.state.SavepointTriggerId = nil
|
||||
job.updateState(*job.state)
|
||||
}
|
||||
// func (job *ManagedJob) removeSavepointTriggerId() {
|
||||
// job.state.SavepointTriggerId = nil
|
||||
// job.updateState(*job.state)
|
||||
// }
|
||||
|
||||
func (job *ManagedJob) SetStatus(status JobStatus) {
|
||||
job.state.Status = status
|
||||
job.updateState(*job.state)
|
||||
}
|
||||
// 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
|
||||
if job.def.Status.JobId != nil {
|
||||
return job.def.Status.JobId
|
||||
} else {
|
||||
return nil
|
||||
}
|
||||
|
||||
@@ -1,6 +1,7 @@
|
||||
package managed_job
|
||||
|
||||
import (
|
||||
"flink-kube-operator/internal/crd/v1alpha1"
|
||||
"strings"
|
||||
|
||||
"gitea.com/logicamp/lc"
|
||||
@@ -8,22 +9,36 @@ import (
|
||||
)
|
||||
|
||||
func (job *ManagedJob) checkStatus() error {
|
||||
if job.state.JobId == nil {
|
||||
if job.def.Status.JobId == nil {
|
||||
lc.Logger.Debug("[managed-job] [status] no job id")
|
||||
return ErrNoJobId
|
||||
return v1alpha1.ErrNoJobId
|
||||
}
|
||||
statusResp, err := job.client.Job(*job.state.JobId)
|
||||
statusResp, err := job.client.Job(*job.def.Status.JobId)
|
||||
if err != nil {
|
||||
lc.Logger.Debug("[managed-job] [status] cannot fetch status", zap.Error(err))
|
||||
if strings.IndexAny(err.Error(), "http status not 2xx: 404") == 0 {
|
||||
job.updateState(jobState{
|
||||
JobId: job.state.JobId,
|
||||
Status: JobStatusNotFound,
|
||||
// job.updateState(jobState{
|
||||
// JobId: job.state.JobId,
|
||||
// Status: v1alpha1.JobStatusNotFound,
|
||||
// })
|
||||
job.crd.Patch(job.def.UID, map[string]interface{}{
|
||||
"status": map[string]interface{}{
|
||||
"jobId": &job.def.Status.JobId,
|
||||
"jobStatus": "",
|
||||
"lifeCycleStatus": v1alpha1.LifeCycleStatusFailed,
|
||||
"error": "Job not found",
|
||||
},
|
||||
})
|
||||
}
|
||||
return err
|
||||
}
|
||||
//lc.Logger.Debug("[managed-job] [status]", zap.Any("status-resp", statusResp))
|
||||
job.SetStatus(JobStatus(statusResp.State))
|
||||
job.crd.Patch(job.def.UID, map[string]interface{}{
|
||||
"status": map[string]interface{}{
|
||||
"jobId": &job.def.Status.JobId,
|
||||
"jobStatus": statusResp.State,
|
||||
"lifeCycleStatus": v1alpha1.LifeCycleStatusFailed,
|
||||
"error": "Job not found",
|
||||
},
|
||||
})
|
||||
return err
|
||||
}
|
||||
|
||||
@@ -1,40 +1,16 @@
|
||||
package managed_job
|
||||
|
||||
import (
|
||||
"errors"
|
||||
"flink-kube-operator/internal/crd/v1alpha1"
|
||||
"time"
|
||||
)
|
||||
|
||||
type JobStatus string
|
||||
|
||||
var (
|
||||
ErrNoJobId = errors.New("[managed-job] no job id")
|
||||
ErrNoSavepointTriggerId = errors.New("[managed-job] no savepoint trigger id")
|
||||
ErrNoSavepointPath = errors.New("[managed-job] no savepoint path")
|
||||
)
|
||||
|
||||
const (
|
||||
JobStatusInitializing JobStatus = "INITIALIZING"
|
||||
JobStatusRunning JobStatus = "RUNNING"
|
||||
JobStatusCreating JobStatus = "CREATING"
|
||||
JobStatusNotFound JobStatus = "NotFound"
|
||||
JobStatusError JobStatus = "ERROR"
|
||||
JobStatusReconciling JobStatus = "RECONCILING"
|
||||
JobStatusFailed JobStatus = "FAILED"
|
||||
JobStatusFailing JobStatus = "FAILING"
|
||||
JobStatusRestarting JobStatus = "RESTARTING"
|
||||
JobStatusFinished JobStatus = "FINISHED"
|
||||
JobStatusCanceled JobStatus = "CANCELED"
|
||||
JobStatusCancelling JobStatus = "CANCELLING"
|
||||
JobStatusSuspended JobStatus = "SUSPENDED"
|
||||
)
|
||||
|
||||
type jobState struct {
|
||||
Status JobStatus `json:"status"`
|
||||
Error *string `json:"error"`
|
||||
Info *string `json:"info"`
|
||||
JobId *string `json:"job_id"`
|
||||
LastSavepointPath *string `json:"last_savepoint_location"`
|
||||
SavepointTriggerId *string `json:"savepoint_trigger_id"`
|
||||
LastSavepointDate *time.Time `json:"last_savepoint_time"`
|
||||
Status v1alpha1.JobStatus `json:"status"`
|
||||
Error *string `json:"error"`
|
||||
Info *string `json:"info"`
|
||||
JobId *string `json:"job_id"`
|
||||
LastSavepointPath *string `json:"last_savepoint_location"`
|
||||
SavepointTriggerId *string `json:"savepoint_trigger_id"`
|
||||
LastSavepointDate *time.Time `json:"last_savepoint_time"`
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user