logicamp/flink-kube-operator
3
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(savepoint): missing savepoint path env and not handled savepoint errors

Browse Source
This commit is contained in:
Mohamad Khani
2024-12-13 14:57:55 +03:30
parent b2d2295d07
commit 89702d287a
3 changed files with 81 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
internal/managed_job/savepoint.go
View File

@@ -2,6 +2,7 @@ package managed_job
import (
"flink-kube-operator/internal/crd/v1alpha1"
"os"
"strings"
"time"
@@ -17,7 +18,7 @@ func (job ManagedJob) createSavepoint() error {
return v1alpha1.ErrNoJobId
}
pkg.Logger.Info("[managed-job] [savepoint] creating savepoint", zap.String("interval", job.def.Spec.SavepointInterval.String()))
resp, err := job.client.SavePoints(*job.def.Status.JobId, "/flink-data/savepoints-2/", false)
resp, err := job.client.SavePoints(*job.def.Status.JobId, os.Getenv("SAVEPOINT_PATH"), false)
if err != nil {
pkg.Logger.Error("[managed-job] [savepoint] error in creating savepoint", zap.Error(err))
return err
@@ -42,24 +43,38 @@ func (job ManagedJob) trackSavepoint() error {
return v1alpha1.ErrNoSavepointTriggerId
}
resp, err := job.client.TrackSavepoint(*job.def.Status.JobId, *job.def.Status.SavepointTriggerId)
pkg.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 {
pkg.Logger.Debug("[managed-job] [savepoint] track savepoint",
zap.Any("status.Id", resp.Status.Id),
zap.Any("failureCause.stacktrace", resp.Operation.FailureCause.StackTrace),
zap.Any("failureCause.class", resp.Operation.FailureCause.Class),
zap.Error(err),
)
if err != nil || resp.Operation.FailureCause.Class != "" {
if err != nil {
if strings.IndexAny(err.Error(), "http status not 2xx: 404") == 0 {
job.crd.Patch(job.def.UID, map[string]interface{}{
"status": map[string]interface{}{
"savepointTriggerId": nil,
},
})
}
} else {
job.crd.Patch(job.def.UID, map[string]interface{}{
"status": map[string]interface{}{
"savepointTriggerId": nil,
"error": resp.Operation.FailureCause.StackTrace,
},
})
}
} else {
if resp.Status.Id == api.SavepointStatusInCompleted {
job.crd.Patch(job.def.UID, map[string]interface{}{
"status": map[string]interface{}{
"lastSavepointPath": resp.Operation.Location,
"lastSavepointDate": time.Now().Format(time.RFC3339),
},
})
}
}
if resp.Status.Id == api.SavepointStatusInCompleted {
job.crd.Patch(job.def.UID, map[string]interface{}{
"status": map[string]interface{}{
"lastSavepointPath": resp.Operation.Location,
"lastSavepointDate": time.Now().Format(time.RFC3339),
},
})
}
return nil
}