worker/server: requeue unresponsive jobs

If a job is unresponsive the worker has most likely crashed or been shut
down and the in-progress job been lost.

Instead of failing these jobs, requeue them up to two times. Once a job is lost
a third time it fails. This avoids infinite loops.

This is implemented by extending FinishJob to RequeuOrFinish job. It takes a
max number of requeues as an argument, and if that is 0, it has the same
behavior as FinishJob used to have.

If the maximum number of requeues has not yet been reached, then the running
job is returned to pending state to be picked up again.
This commit is contained in:
Tom Gundersen 2022-03-18 21:39:32 +00:00 committed by Sanne Raymaekers
parent d02f666a4b
commit 626530818d
8 changed files with 216 additions and 61 deletions

View file

@ -75,7 +75,8 @@ type job struct {
FinishedAt time.Time `json:"finished_at,omitempty"`
ExpiresAt time.Time `json:"expires_at,omitempty"`
Canceled bool `json:"canceled,omitempty"`
Retries uint64 `json:"retries"`
Canceled bool `json:"canceled,omitempty"`
}
// Create a new fsJobQueue object for `dir`. This object must have exclusive
@ -111,7 +112,7 @@ func New(dir string) (*fsJobQueue, error) {
if !j.StartedAt.IsZero() && j.FinishedAt.IsZero() && !j.Canceled {
// Fail older running jobs which don't have a token stored
if j.Token == uuid.Nil {
err = q.FinishJob(j.Id, nil)
err = q.RequeueOrFinishJob(j.Id, 0, nil)
if err != nil {
return nil, fmt.Errorf("Error finishing job '%s' without a token: %v", j.Id, err)
}
@ -274,7 +275,7 @@ func (q *fsJobQueue) DequeueByID(ctx context.Context, id uuid.UUID) (uuid.UUID,
return j.Token, j.Dependencies, j.Type, j.Args, nil
}
func (q *fsJobQueue) FinishJob(id uuid.UUID, result interface{}) error {
func (q *fsJobQueue) RequeueOrFinishJob(id uuid.UUID, maxRetries uint64, result interface{}) error {
q.mu.Lock()
defer q.mu.Unlock()
@ -291,33 +292,57 @@ func (q *fsJobQueue) FinishJob(id uuid.UUID, result interface{}) error {
return jobqueue.ErrNotRunning
}
j.FinishedAt = time.Now()
j.Result, err = json.Marshal(result)
if err != nil {
return fmt.Errorf("error marshaling result: %v", err)
}
delete(q.heartbeats, j.Token)
delete(q.jobIdByToken, j.Token)
delete(q.heartbeats, j.Token)
// Write before notifying dependants, because it will be read again.
err = q.db.Write(id.String(), j)
if err != nil {
return fmt.Errorf("error writing job %s: %v", id, err)
}
if j.Retries >= maxRetries {
j.FinishedAt = time.Now()
for _, depid := range q.dependants[id] {
dep, err := q.readJob(depid)
j.Result, err = json.Marshal(result)
if err != nil {
return err
return fmt.Errorf("error marshaling result: %v", err)
}
err = q.maybeEnqueue(dep, false)
// Write before notifying dependants, because it will be read again.
err = q.db.Write(id.String(), j)
if err != nil {
return err
return fmt.Errorf("error writing job %s: %v", id, err)
}
for _, depid := range q.dependants[id] {
dep, err := q.readJob(depid)
if err != nil {
return err
}
err = q.maybeEnqueue(dep, false)
if err != nil {
return err
}
}
delete(q.dependants, id)
} else {
j.Token = uuid.Nil
j.StartedAt = time.Time{}
j.Retries += 1
// Write the job before updating in-memory state, so that the latter
// doesn't become corrupt when writing fails.
err = q.db.Write(j.Id.String(), j)
if err != nil {
return fmt.Errorf("cannot write job: %v", err)
}
// add the job to the list of pending ones
q.pending.PushBack(j.Id)
// notify all listeners in a non-blocking way
for c := range q.listeners {
select {
case c <- struct{}{}:
default:
}
}
}
delete(q.dependants, id)
return nil
}

View file

@ -33,6 +33,8 @@ func TestJobQueue(t *testing.T, makeJobQueue MakeJobQueue) {
t.Run("errors", wrap(testErrors))
t.Run("args", wrap(testArgs))
t.Run("cancel", wrap(testCancel))
t.Run("requeue", wrap(testRequeue))
t.Run("requeue-limit", wrap(testRequeueLimit))
t.Run("job-types", wrap(testJobTypes))
t.Run("dependencies", wrap(testDependencies))
t.Run("multiple-workers", wrap(testMultipleWorkers))
@ -61,7 +63,7 @@ func finishNextTestJob(t *testing.T, q jobqueue.JobQueue, jobType string, result
require.Equal(t, jobType, typ)
require.NotNil(t, args)
err = q.FinishJob(id, result)
err = q.RequeueOrFinishJob(id, 0, result)
require.NoError(t, err)
return id
@ -88,7 +90,7 @@ func testErrors(t *testing.T, q jobqueue.JobQueue) {
require.NoError(t, err)
require.Equal(t, id, idFromT)
err = q.FinishJob(id, nil)
err = q.RequeueOrFinishJob(id, 0, nil)
require.NoError(t, err)
// Make sure the token gets removed
@ -363,7 +365,7 @@ func testCancel(t *testing.T, q jobqueue.JobQueue) {
require.Equal(t, jobType, "clownfish")
require.True(t, canceled)
require.Nil(t, result)
err = q.FinishJob(id, &testResult{})
err = q.RequeueOrFinishJob(id, 0, &testResult{})
require.Error(t, err)
// Cancel a running job, which should not dequeue the canceled job from above
@ -383,7 +385,7 @@ func testCancel(t *testing.T, q jobqueue.JobQueue) {
require.Equal(t, jobType, "clownfish")
require.True(t, canceled)
require.Nil(t, result)
err = q.FinishJob(id, &testResult{})
err = q.RequeueOrFinishJob(id, 0, &testResult{})
require.Error(t, err)
// Cancel a finished job, which is a no-op
@ -396,7 +398,7 @@ func testCancel(t *testing.T, q jobqueue.JobQueue) {
require.Empty(t, deps)
require.Equal(t, "clownfish", typ)
require.Equal(t, json.RawMessage("null"), args)
err = q.FinishJob(id, &testResult{})
err = q.RequeueOrFinishJob(id, 0, &testResult{})
require.NoError(t, err)
err = q.CancelJob(id)
require.Error(t, err)
@ -409,6 +411,73 @@ func testCancel(t *testing.T, q jobqueue.JobQueue) {
require.NoError(t, err)
}
func testRequeue(t *testing.T, q jobqueue.JobQueue) {
// Requeue a non-existing job
err := q.RequeueOrFinishJob(uuid.New(), 1, nil)
require.Error(t, err)
// Requeue a pending job
id := pushTestJob(t, q, "clownfish", nil, nil, "")
require.NotEmpty(t, id)
err = q.RequeueOrFinishJob(id, 1, nil)
require.Error(t, err)
// Requeue a running job
r, tok1, deps, typ, args, err := q.Dequeue(context.Background(), []string{"clownfish"}, []string{""})
require.NoError(t, err)
require.Equal(t, id, r)
require.NotEmpty(t, tok1)
require.Empty(t, deps)
require.Equal(t, "clownfish", typ)
require.Equal(t, json.RawMessage("null"), args)
err = q.RequeueOrFinishJob(id, 1, nil)
require.NoError(t, err)
r, tok2, deps, typ, args, err := q.Dequeue(context.Background(), []string{"clownfish"}, []string{""})
require.NoError(t, err)
require.Equal(t, id, r)
require.NotEmpty(t, tok2)
require.NotEqual(t, tok1, tok2)
require.Empty(t, deps)
require.Equal(t, "clownfish", typ)
require.Equal(t, json.RawMessage("null"), args)
jobType, _, result, _, _, _, canceled, _, _, err := q.JobStatus(id)
require.NoError(t, err)
require.Equal(t, jobType, "clownfish")
require.False(t, canceled)
require.Nil(t, result)
err = q.RequeueOrFinishJob(id, 0, &testResult{})
require.NoError(t, err)
// Requeue a finished job
err = q.RequeueOrFinishJob(id, 1, nil)
require.Error(t, err)
}
func testRequeueLimit(t *testing.T, q jobqueue.JobQueue) {
// Start a job
id := pushTestJob(t, q, "clownfish", nil, nil, "")
require.NotEmpty(t, id)
_, _, _, _, _, err := q.Dequeue(context.Background(), []string{"clownfish"}, []string{""})
require.NoError(t, err)
// Requeue once
err = q.RequeueOrFinishJob(id, 1, nil)
require.NoError(t, err)
// Start again
_, _, _, _, _, err = q.Dequeue(context.Background(), []string{"clownfish"}, []string{""})
require.NoError(t, err)
_, _, result, _, _, finished, _, _, _, err := q.JobStatus(id)
require.NoError(t, err)
require.True(t, finished.IsZero())
require.Nil(t, result)
// Requeue a second time, this time finishing it
err = q.RequeueOrFinishJob(id, 1, &testResult{})
require.NoError(t, err)
_, _, result, _, _, finished, _, _, _, err = q.JobStatus(id)
require.NoError(t, err)
require.False(t, finished.IsZero())
require.NotNil(t, result)
}
func testHeartbeats(t *testing.T, q jobqueue.JobQueue) {
id := pushTestJob(t, q, "octopus", nil, nil, "")
// No heartbeats for queued job
@ -434,7 +503,7 @@ func testHeartbeats(t *testing.T, q jobqueue.JobQueue) {
require.NoError(t, err)
require.Equal(t, id2, id)
err = q.FinishJob(id, &testResult{})
err = q.RequeueOrFinishJob(id, 0, &testResult{})
require.NoError(t, err)
// No heartbeats for finished job
@ -456,7 +525,7 @@ func testDequeueByID(t *testing.T, q jobqueue.JobQueue) {
require.Equal(t, "octopus", typ)
require.NotNil(t, args)
err = q.FinishJob(one, nil)
err = q.RequeueOrFinishJob(one, 0, nil)
require.NoError(t, err)
require.Equal(t, two, finishNextTestJob(t, q, "octopus", testResult{}, nil))
@ -482,7 +551,7 @@ func testDequeueByID(t *testing.T, q jobqueue.JobQueue) {
_, _, _, _, err = q.DequeueByID(context.Background(), one)
require.Equal(t, jobqueue.ErrNotPending, err)
err = q.FinishJob(one, nil)
err = q.RequeueOrFinishJob(one, 0, nil)
require.NoError(t, err)
_, _, _, _, err = q.DequeueByID(context.Background(), one)