// updateSummaryWithJob creates or updates job summaries when new jobs are
// upserted or existing ones are updated
func (s *StateStore) updateSummaryWithJob(index uint64, job *structs.Job,
watcher watch.Items, txn *memdb.Txn) error {
existing, err := s.JobSummaryByID(job.ID)
if err != nil {
return fmt.Errorf("unable to retrieve summary for job: %v", err)
}
var hasSummaryChanged bool
if existing == nil {
existing = &structs.JobSummary{
JobID: job.ID,
Summary: make(map[string]structs.TaskGroupSummary),
CreateIndex: index,
}
hasSummaryChanged = true
}
for _, tg := range job.TaskGroups {
if _, ok := existing.Summary[tg.Name]; !ok {
newSummary := structs.TaskGroupSummary{
Complete: 0,
Failed: 0,
Running: 0,
Starting: 0,
}
existing.Summary[tg.Name] = newSummary
hasSummaryChanged = true
}
}
// The job summary has changed, so add to watcher and update the modify
// index.
if hasSummaryChanged {
existing.ModifyIndex = index
watcher.Add(watch.Item{Table: "job_summary"})
watcher.Add(watch.Item{JobSummary: job.ID})
// Update the indexes table for job summary
if err := txn.Insert("index", &IndexEntry{"job_summary", index}); err != nil {
return fmt.Errorf("index update failed: %v", err)
}
if err := txn.Insert("job_summary", *existing); err != nil {
return err
}
}
return nil
}
// nestedUpdateAllocFromClient is used to nest an update of an allocation with client status
func (s *StateStore) nestedUpdateAllocFromClient(txn *memdb.Txn, watcher watch.Items, index uint64, alloc *structs.Allocation) error {
// Look for existing alloc
existing, err := txn.First("allocs", "id", alloc.ID)
if err != nil {
return fmt.Errorf("alloc lookup failed: %v", err)
}
// Nothing to do if this does not exist
if existing == nil {
return nil
}
exist := existing.(*structs.Allocation)
// Trigger the watcher
watcher.Add(watch.Item{Alloc: alloc.ID})
watcher.Add(watch.Item{AllocEval: exist.EvalID})
watcher.Add(watch.Item{AllocJob: exist.JobID})
watcher.Add(watch.Item{AllocNode: exist.NodeID})
// Copy everything from the existing allocation
copyAlloc := new(structs.Allocation)
*copyAlloc = *exist
// Pull in anything the client is the authority on
copyAlloc.ClientStatus = alloc.ClientStatus
copyAlloc.ClientDescription = alloc.ClientDescription
copyAlloc.TaskStates = alloc.TaskStates
// Update the modify index
copyAlloc.ModifyIndex = index
if err := s.updateSummaryWithAlloc(index, copyAlloc, exist, watcher, txn); err != nil {
return fmt.Errorf("error updating job summary: %v", err)
}
// Update the allocation
if err := txn.Insert("allocs", copyAlloc); err != nil {
return fmt.Errorf("alloc insert failed: %v", err)
}
// Set the job's status
forceStatus := ""
if !copyAlloc.TerminalStatus() {
forceStatus = structs.JobStatusRunning
}
jobs := map[string]string{exist.JobID: forceStatus}
if err := s.setJobStatuses(index, watcher, txn, jobs, false); err != nil {
return fmt.Errorf("setting job status failed: %v", err)
}
return nil
}
// setJobStatus sets the status of the job by looking up associated evaluations
// and allocations. evalDelete should be set to true if setJobStatus is being
// called because an evaluation is being deleted (potentially because of garbage
// collection). If forceStatus is non-empty, the job's status will be set to the
// passed status.
func (s *StateStore) setJobStatus(index uint64, watcher watch.Items, txn *memdb.Txn,
job *structs.Job, evalDelete bool, forceStatus string) error {
// Capture the current status so we can check if there is a change
oldStatus := job.Status
newStatus := forceStatus
// If forceStatus is not set, compute the jobs status.
if forceStatus == "" {
var err error
newStatus, err = s.getJobStatus(txn, job, evalDelete)
if err != nil {
return err
}
}
// Fast-path if nothing has changed.
if oldStatus == newStatus {
return nil
}
// The job has changed, so add to watcher.
watcher.Add(watch.Item{Table: "jobs"})
watcher.Add(watch.Item{Job: job.ID})
// Copy and update the existing job
updated := job.Copy()
updated.Status = newStatus
updated.ModifyIndex = index
// Insert the job
if err := txn.Insert("jobs", updated); err != nil {
return fmt.Errorf("job insert failed: %v", err)
}
if err := txn.Insert("index", &IndexEntry{"jobs", index}); err != nil {
return fmt.Errorf("index update failed: %v", err)
}
return nil
}
// updateSummaryWithAlloc updates the job summary when allocations are updated
// or inserted
func (s *StateStore) updateSummaryWithAlloc(index uint64, alloc *structs.Allocation,
existingAlloc *structs.Allocation, watcher watch.Items, txn *memdb.Txn) error {
// We don't have to update the summary if the job is missing
if alloc.Job == nil {
return nil
}
summaryRaw, err := txn.First("job_summary", "id", alloc.JobID)
if err != nil {
return fmt.Errorf("unable to lookup job summary for job id %q: %v", err)
}
if summaryRaw == nil {
// Check if the job is de-registered
rawJob, err := txn.First("jobs", "id", alloc.JobID)
if err != nil {
return fmt.Errorf("unable to query job: %v", err)
}
// If the job is de-registered then we skip updating it's summary
if rawJob == nil {
return nil
}
return fmt.Errorf("job summary for job %q is not present", alloc.JobID)
}
summary := summaryRaw.(structs.JobSummary)
jobSummary := summary.Copy()
// Not updating the job summary because the allocation doesn't belong to the
// currently registered job
if jobSummary.CreateIndex != alloc.Job.CreateIndex {
return nil
}
tgSummary, ok := jobSummary.Summary[alloc.TaskGroup]
if !ok {
return fmt.Errorf("unable to find task group in the job summary: %v", alloc.TaskGroup)
}
var summaryChanged bool
if existingAlloc == nil {
switch alloc.DesiredStatus {
case structs.AllocDesiredStatusStop, structs.AllocDesiredStatusEvict:
s.logger.Printf("[ERR] state_store: new allocation inserted into state store with id: %v and state: %v",
alloc.ID, alloc.DesiredStatus)
}
switch alloc.ClientStatus {
case structs.AllocClientStatusPending:
tgSummary.Starting += 1
if tgSummary.Queued > 0 {
tgSummary.Queued -= 1
}
summaryChanged = true
case structs.AllocClientStatusRunning, structs.AllocClientStatusFailed,
structs.AllocClientStatusComplete:
s.logger.Printf("[ERR] state_store: new allocation inserted into state store with id: %v and state: %v",
alloc.ID, alloc.ClientStatus)
}
} else if existingAlloc.ClientStatus != alloc.ClientStatus {
// Incrementing the client of the bin of the current state
switch alloc.ClientStatus {
case structs.AllocClientStatusRunning:
tgSummary.Running += 1
case structs.AllocClientStatusFailed:
tgSummary.Failed += 1
case structs.AllocClientStatusPending:
tgSummary.Starting += 1
case structs.AllocClientStatusComplete:
tgSummary.Complete += 1
case structs.AllocClientStatusLost:
tgSummary.Lost += 1
}
// Decrementing the count of the bin of the last state
switch existingAlloc.ClientStatus {
case structs.AllocClientStatusRunning:
tgSummary.Running -= 1
case structs.AllocClientStatusPending:
tgSummary.Starting -= 1
case structs.AllocClientStatusLost:
tgSummary.Lost -= 1
case structs.AllocClientStatusFailed, structs.AllocClientStatusComplete:
default:
s.logger.Printf("[ERR] state_store: invalid old state of allocation with id: %v, and state: %v",
existingAlloc.ID, existingAlloc.ClientStatus)
}
summaryChanged = true
}
jobSummary.Summary[alloc.TaskGroup] = tgSummary
if summaryChanged {
jobSummary.ModifyIndex = index
watcher.Add(watch.Item{Table: "job_summary"})
watcher.Add(watch.Item{JobSummary: alloc.JobID})
// Update the indexes table for job summary
if err := txn.Insert("index", &IndexEntry{"job_summary", index}); err != nil {
return fmt.Errorf("index update failed: %v", err)
}
if err := txn.Insert("job_summary", *jobSummary); err != nil {
return fmt.Errorf("updating job summary failed: %v", err)
}
}
return nil
}
