// URL returns a url to builder step failure page.
func (f stepFailure) URL() string {
return client.StepURL(f.masterName, f.builderName, f.step.Name, f.build.Number)
}
// TODO: also check the build slaves to see if there are alerts for currently running builds that
// haven't shown up in CBE yet.
func (a *Analyzer) builderAlerts(masterName string, builderName string, b *messages.Builder) ([]messages.Alert, []error) {
if len(b.CachedBuilds) == 0 {
// TODO: Make an alert for this?
return nil, []error{errNoRecentBuilds}
}
recentBuildIDs := b.CachedBuilds
// Should be a *reverse* sort.
sort.Sort(buildNums(recentBuildIDs))
if len(recentBuildIDs) > a.MaxRecentBuilds {
recentBuildIDs = recentBuildIDs[:a.MaxRecentBuilds]
}
alerts, errs := []messages.Alert{}, []error{}
lastBuild, lastCompletedBuild, err := a.lastBuilds(masterName, builderName, recentBuildIDs)
if err != nil {
errs = append(errs, err)
return nil, errs
}
// Examining only the latest build is probably suboptimal since if it's still in progress it might
// not have hit a step that is going to fail and has failed repeatedly for the last few builds.
// AKA "Reliable failures". TODO: Identify "Reliable failures"
lastStep, lastUpdated, err := a.latestBuildStep(lastBuild)
if err != nil {
errs = append(errs, fmt.Errorf("Couldn't get latest build step for %s.%s: %v", masterName, builderName, err))
return alerts, errs
}
elapsed := a.Now().Sub(lastUpdated.Time())
links := []messages.Link{
{"Builder", client.BuilderURL(masterName, builderName)},
{"Last build", client.BuildURL(masterName, builderName, lastBuild.Number)},
{"Last build step", client.StepURL(masterName, builderName, lastStep, lastBuild.Number)},
}
switch b.State {
case messages.StateBuilding:
if elapsed > a.HungBuilderThresh && lastStep != StepCompletedRun {
alerts = append(alerts, messages.Alert{
Key: fmt.Sprintf("%s.%s.hung", masterName, builderName),
Title: fmt.Sprintf("%s.%s is hung in step %s.", masterName, builderName, lastStep),
Body: fmt.Sprintf("%s.%s has been building for %v (last step update %s), past the alerting threshold of %v", masterName, builderName, elapsed, lastUpdated.Time(), a.HungBuilderThresh),
Severity: hungBuilderSev,
Time: messages.TimeToEpochTime(a.Now()),
Links: links,
})
// Note, just because it's building doesn't mean it's in a good state. If the last N builds
// all failed (for some large N) then this might still be alertable.
}
case messages.StateOffline:
if elapsed > a.OfflineBuilderThresh {
alerts = append(alerts, messages.Alert{
Key: fmt.Sprintf("%s.%s.offline", masterName, builderName),
Title: fmt.Sprintf("%s.%s is offline.", masterName, builderName),
Body: fmt.Sprintf("%s.%s has been offline for %v (last step update %s %v), past the alerting threshold of %v", masterName, builderName, elapsed, lastUpdated.Time(), float64(lastUpdated), a.OfflineBuilderThresh),
Severity: offlineBuilderSev,
Time: messages.TimeToEpochTime(a.Now()),
Links: links,
})
}
case messages.StateIdle:
if b.PendingBuilds > a.IdleBuilderCountThresh {
alerts = append(alerts, messages.Alert{
Key: fmt.Sprintf("%s.%s.idle", masterName, builderName),
Title: fmt.Sprintf("%s.%s is idle with too many pending builds.", masterName, builderName),
Body: fmt.Sprintf("%s.%s is idle with %d pending builds, past the alerting threshold of %d", masterName, builderName, b.PendingBuilds, a.IdleBuilderCountThresh),
Severity: idleBuilderSev,
Time: messages.TimeToEpochTime(a.Now()),
Links: links,
})
}
default:
log.Errorf("Unknown %s.%s builder state: %s", masterName, builderName, b.State)
}
// Check for alerts on the most recent complete build
log.Infof("Checking %d most recent builds for alertable step failures: %s/%s", len(recentBuildIDs), masterName, builderName)
as, es := a.builderStepAlerts(masterName, builderName, []int64{lastCompletedBuild.Number})
if len(as) > 0 {
mostRecentComplete := 0
for i, id := range recentBuildIDs {
if id == lastCompletedBuild.Number {
mostRecentComplete = i
}
}
as, es = a.builderStepAlerts(masterName, builderName, recentBuildIDs[mostRecentComplete:])
alerts = append(alerts, as...)
errs = append(errs, es...)
}
return alerts, errs
}
