// Starts the job in the Buildkite Agent API. We'll retry on connection-related
// issues, but if a connection succeeds and we get an error response back from
// Buildkite, we won't bother retrying. For example, a "no such host" will
// retry, but a 422 from Buildkite won't.
func (r *JobRunner) startJob(startedAt time.Time) error {
r.Job.StartedAt = startedAt.UTC().Format(time.RFC3339Nano)
return retry.Do(func(s *retry.Stats) error {
_, err := r.APIClient.Jobs.Start(r.Job)
if err != nil {
if api.IsRetryableError(err) {
logger.Warn("%s (%s)", err, s)
} else {
logger.Warn("Buildkite rejected the call to start the job (%s)", err)
s.Break()
}
}
return err
}, &retry.Config{Maximum: 30, Interval: 5 * time.Second})
}
// Performs a ping, which returns what action the agent should take next.
func (a *AgentWorker) Ping() {
// Update the proc title
a.UpdateProcTitle("pinging")
ping, _, err := a.APIClient.Pings.Get()
if err != nil {
// If a ping fails, we don't really care, because it'll
// ping again after the interval.
logger.Warn("Failed to ping: %s", err)
return
}
// Should we switch endpoints?
if ping.Endpoint != "" && ping.Endpoint != a.Agent.Endpoint {
// Before switching to the new one, do a ping test to make sure it's
// valid. If it is, switch and carry on, otherwise ignore the switch
// for now.
newAPIClient := APIClient{Endpoint: ping.Endpoint, Token: a.Agent.AccessToken}.Create()
newPing, _, err := newAPIClient.Pings.Get()
if err != nil {
logger.Warn("Failed to ping the new endpoint %s - ignoring switch for now (%s)", ping.Endpoint, err)
} else {
// Replace the APIClient and process the new ping
a.APIClient = newAPIClient
a.Agent.Endpoint = ping.Endpoint
ping = newPing
}
}
// Is there a message that should be shown in the logs?
if ping.Message != "" {
logger.Info(ping.Message)
}
// Should the agent disconnect?
if ping.Action == "disconnect" {
a.Stop(false)
return
}
// If we don't have a job, there's nothing to do!
if ping.Job == nil {
// Update the proc title
a.UpdateProcTitle("idle")
return
}
// Update the proc title
a.UpdateProcTitle(fmt.Sprintf("job %s", strings.Split(ping.Job.ID, "-")[0]))
logger.Info("Assigned job %s. Accepting...", ping.Job.ID)
// Accept the job. We'll retry on connection related issues, but if
// Buildkite returns a 422 or 500 for example, we'll just bail out,
// re-ping, and try the whole process again.
var accepted *api.Job
retry.Do(func(s *retry.Stats) error {
accepted, _, err = a.APIClient.Jobs.Accept(ping.Job)
if err != nil {
if api.IsRetryableError(err) {
logger.Warn("%s (%s)", err, s)
} else {
logger.Warn("Buildkite rejected the call to accept the job (%s)", err)
s.Break()
}
}
return err
}, &retry.Config{Maximum: 30, Interval: 1 * time.Second})
// If `accepted` is nil, then the job was never accepted
if accepted == nil {
logger.Error("Failed to accept job")
return
}
// Now that the job has been accepted, we can start it.
a.jobRunner, err = JobRunner{
Endpoint: accepted.Endpoint,
Agent: a.Agent,
AgentConfiguration: a.AgentConfiguration,
Job: accepted,
}.Create()
// Was there an error creating the job runner?
if err != nil {
logger.Error("Failed to initialize job: %s", err)
return
}
// Start running the job
if err = a.jobRunner.Run(); err != nil {
logger.Error("Failed to run job: %s", err)
}
// No more job, no more runner.
a.jobRunner = nil
}