// Finishes the job in the Buildkite Agent API. This call will keep on retrying
// forever until it finally gets a successfull response from the API.
func (r *JobRunner) finishJob(finishedAt time.Time, exitStatus string, failedChunkCount int) error {
r.Job.FinishedAt = finishedAt.UTC().Format(time.RFC3339Nano)
r.Job.ExitStatus = exitStatus
r.Job.ChunksFailedCount = failedChunkCount
return retry.Do(func(s *retry.Stats) error {
response, err := r.APIClient.Jobs.Finish(r.Job)
if err != nil {
// If the API returns with a 422, that means that we
// succesfully tried to finish the job, but Buildkite
// rejected the finish for some reason. This can
// sometimes mean that Buildkite has cancelled the job
// before we get a chance to send the final API call
// (maybe this agent took too long to kill the
// process). In that case, we don't want to keep trying
// to finish the job forever so we'll just bail out and
// go find some more work to do.
if response != nil && response.StatusCode == 422 {
logger.Warn("Buildkite rejected the call to finish the job (%s)", err)
s.Break()
} else {
logger.Warn("%s (%s)", err, s)
}
}
return err
}, &retry.Config{Forever: true, Interval: 1 * time.Second})
}
func (d Download) Start() error {
return retry.Do(func(s *retry.Stats) error {
err := d.try()
if err != nil {
logger.Warn("Error trying to download %s (%s) %s", d.URL, err, s)
}
return err
}, &retry.Config{Maximum: d.Retries, Interval: 1 * time.Second})
}
func (a *ArtifactBatchCreator) Create() ([]*api.Artifact, error) {
length := len(a.Artifacts)
chunks := 30
// Split into the artifacts into chunks so we're not uploading a ton of
// files at once.
for i := 0; i < length; i += chunks {
j := i + chunks
if length < j {
j = length
}
// The artifacts that will be uploaded in this chunk
theseArtiacts := a.Artifacts[i:j]
// An ID is required so Buildkite can ensure this create
// operation is idompotent (if we try and upload the same ID
// twice, it'll just return the previous data and skip the
// upload)
batch := &api.ArtifactBatch{api.NewUUID(), theseArtiacts, a.UploadDestination}
logger.Info("Creating (%d-%d)/%d artifacts", i, j, length)
var creation *api.ArtifactBatchCreateResponse
var resp *api.Response
var err error
// Retry the batch upload a couple of times
err = retry.Do(func(s *retry.Stats) error {
creation, resp, err = a.APIClient.Artifacts.Create(a.JobID, batch)
if resp != nil && (resp.StatusCode == 401 || resp.StatusCode == 404 || resp.StatusCode == 500) {
s.Break()
}
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
// Did the batch creation eventually fail?
if err != nil {
return nil, err
}
// Save the id and instructions to each artifact
index := 0
for _, id := range creation.ArtifactIDs {
theseArtiacts[index].ID = id
theseArtiacts[index].UploadInstructions = creation.UploadInstructions
index += 1
}
}
return a.Artifacts, nil
}
func (r *JobRunner) onUploadHeaderTime(cursor int, total int, times map[string]string) {
retry.Do(func(s *retry.Stats) error {
_, err := r.APIClient.HeaderTimes.Save(r.Job.ID, &api.HeaderTimes{Times: times})
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 5 * time.Second})
}
// Connects the agent to the Buildkite Agent API, retrying up to 30 times if it
// fails.
func (a *AgentWorker) Connect() error {
return retry.Do(func(s *retry.Stats) error {
_, err := a.APIClient.Agents.Connect()
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
}
// Connects the agent to the Buildkite Agent API, retrying up to 30 times if it
// fails.
func (a *AgentWorker) Connect() error {
// Update the proc title
a.UpdateProcTitle("connecting")
return retry.Do(func(s *retry.Stats) error {
_, err := a.APIClient.Agents.Connect()
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
}
// Call when a chunk is ready for upload. It retry the chunk upload with an
// interval before giving up.
func (r *JobRunner) onUploadChunk(chunk *LogStreamerChunk) error {
return retry.Do(func(s *retry.Stats) error {
_, err := r.APIClient.Chunks.Upload(r.Job.ID, &api.Chunk{
Data: chunk.Data,
Sequence: chunk.Order,
})
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
}
// 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 heatbeat
func (a *AgentWorker) Heartbeat() error {
var beat *api.Heartbeat
var err error
// Retry the heartbeat a few times
err = retry.Do(func(s *retry.Stats) error {
beat, _, err = a.APIClient.Heartbeats.Beat()
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 5, Interval: 1 * time.Second})
if err != nil {
return err
}
logger.Debug("Heartbeat sent at %s and received at %s", beat.SentAt, beat.ReceivedAt)
return nil
}
func (b *Buildkite) WriteJobMetadata(metadata map[string]string) error {
client := b.agentClient
for k, v := range metadata {
metadatum := &buildkite.MetaData{
Key: k,
Value: v,
}
err := retry.Do(func(s *retry.Stats) error {
resp, err := client.MetaData.Set(b.jobId, metadatum)
if resp != nil && (resp.StatusCode == 401 || resp.StatusCode == 404) {
s.Break()
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
if err != nil {
return fmt.Errorf("error setting metadata %s: %s", k, err)
}
}
return nil
}
func (a *ArtifactBatchCreator) Create() ([]*api.Artifact, error) {
length := len(a.Artifacts)
chunks := 10
uploaded := []*api.Artifact{}
// Split into the artifacts into chunks so we're not uploading a ton of
// files at once.
for i := 0; i < length; i += chunks {
j := i + chunks
if length < j {
j = length
}
artifacts := a.Artifacts[i:j]
logger.Info("Creating (%d-%d)/%d artifacts", i, j, length)
var u []*api.Artifact
var err error
// Retry the batch upload a couple of times
err = retry.Do(func(s *retry.Stats) error {
u, _, err = a.APIClient.Artifacts.Create(a.JobID, artifacts)
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
if err != nil {
return nil, err
}
uploaded = append(uploaded, u...)
}
return uploaded, nil
}
// Takes the agent template and returns a registered agent. The registered
// agent includes the Access Token used to communicate with the Buildkite Agent
// API
func (r *AgentPool) RegisterAgent(agent *api.Agent) (*api.Agent, error) {
var registered *api.Agent
var err error
var resp *api.Response
register := func(s *retry.Stats) error {
registered, resp, err = r.APIClient.Agents.Register(agent)
if err != nil {
if resp != nil && resp.StatusCode == 401 {
logger.Warn("Buildkite rejected the registration (%s)", err)
s.Break()
} else {
logger.Warn("%s (%s)", err, s)
}
}
return err
}
err = retry.Do(register, &retry.Config{Maximum: 30, Interval: 1 * time.Second})
return registered, err
}
func (a *ArtifactUploader) upload(artifacts []*api.Artifact) error {
var uploader Uploader
// Determine what uploader to use
if a.Destination != "" {
if strings.HasPrefix(a.Destination, "s3://") {
uploader = new(S3Uploader)
} else {
return errors.New("Unknown upload destination: " + a.Destination)
}
} else {
uploader = new(FormUploader)
}
// Setup the uploader
err := uploader.Setup(a.Destination, a.APIClient.DebugHTTP)
if err != nil {
return err
}
// Set the URL's of the artifacts based on the uploader
for _, artifact := range artifacts {
artifact.URL = uploader.URL(artifact)
}
// Create the artifacts on Buildkite
batchCreator := ArtifactBatchCreator{
APIClient: a.APIClient,
JobID: a.JobID,
Artifacts: artifacts,
UploadDestination: a.Destination,
}
artifacts, err = batchCreator.Create()
if err != nil {
return err
}
// Prepare a concurrency pool to upload the artifacts
p := pool.New(pool.MaxConcurrencyLimit)
errors := []error{}
// Create a wait group so we can make sure the uploader waits for all
// the artifact states to upload before finishing
var stateUploaderWaitGroup sync.WaitGroup
stateUploaderWaitGroup.Add(1)
// A map to keep track of artifact states and how many we've uploaded
artifactsStates := make(map[string]string)
artifactStatesUploaded := 0
// Spin up a gourtine that'll uploading artifact statuses every few
// seconds in batches
go func() {
for artifactStatesUploaded < len(artifacts) {
statesToUpload := make(map[string]string)
// Grab all the states we need to upload, and remove
// them from the tracking map
for id, state := range artifactsStates {
statesToUpload[id] = state
delete(artifactsStates, id)
}
if len(statesToUpload) > 0 {
artifactStatesUploaded += len(statesToUpload)
for id, state := range statesToUpload {
logger.Debug("Artifact `%s` has state `%s`", id, state)
}
// Update the states of the artifacts in bulk.
err = retry.Do(func(s *retry.Stats) error {
_, err = a.APIClient.Artifacts.Update(a.JobID, statesToUpload)
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
if err != nil {
logger.Error("Error uploading artifact states: %s", err)
// Track the error that was raised
p.Lock()
errors = append(errors, err)
p.Unlock()
}
logger.Debug("Uploaded %d artfact states (%d/%d)", len(statesToUpload), artifactStatesUploaded, len(artifacts))
}
// Check again for states to upload in a few seconds
time.Sleep(1 * time.Second)
}
stateUploaderWaitGroup.Done()
}()
for _, artifact := range artifacts {
// Create new instance of the artifact for the goroutine
// See: http://golang.org/doc/effective_go.html#channels
artifact := artifact
p.Spawn(func() {
// Show a nice message that we're starting to upload the file
logger.Info("Uploading \"%s\" %d bytes", artifact.Path, artifact.FileSize)
// Upload the artifact and then set the state depending
// on whether or not it passed. We'll retry the upload
// a couple of times before giving up.
err = retry.Do(func(s *retry.Stats) error {
err := uploader.Upload(artifact)
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
var state string
// Did the upload eventually fail?
if err != nil {
logger.Error("Error uploading artifact \"%s\": %s", artifact.Path, err)
// Track the error that was raised
p.Lock()
errors = append(errors, err)
p.Unlock()
state = "error"
} else {
state = "finished"
}
artifactsStates[artifact.ID] = state
})
}
// Wait for the pool to finish
p.Wait()
// Wait for the statuses to finish uploading
stateUploaderWaitGroup.Wait()
if len(errors) > 0 {
logger.Fatal("There were errors with uploading some of the artifacts")
}
return nil
}
}
}
// Create the API client
client := agent.APIClient{
Endpoint: cfg.Endpoint,
Token: cfg.AgentAccessToken,
}.Create()
// Generate a UUID that will identifiy this pipeline change. We
// do this outside of the retry loop because we want this UUID
// to be the same for each attempt at updating the pipeline.
uuid := api.NewUUID()
// Retry the pipeline upload a few times before giving up
err = retry.Do(func(s *retry.Stats) error {
_, err = client.Pipelines.Upload(cfg.Job, &api.Pipeline{UUID: uuid, Data: input, FileName: filename, Replace: cfg.Replace})
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 5, Interval: 1 * time.Second})
if err != nil {
logger.Fatal("Failed to upload and process pipeline: %s", err)
}
logger.Info("Successfully uploaded and parsed pipeline config")
},
}
// Create the API client
client := agent.APIClient{
Endpoint: cfg.Endpoint,
Token: cfg.AgentAccessToken,
}.Create()
// Find the meta data value
var metaData *api.MetaData
var err error
var resp *api.Response
err = retry.Do(func(s *retry.Stats) error {
metaData, resp, err = client.MetaData.Get(cfg.Job, cfg.Key)
// Don't bother retrying if the response was one of these statuses
if resp != nil && (resp.StatusCode == 401 || resp.StatusCode == 404 || resp.StatusCode == 400) {
s.Break()
}
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 5 * time.Second})
if err != nil {
logger.Fatal("Failed to get meta-data: %s", err)
}
// Output the value to STDOUT
fmt.Print(metaData.Value)
},
}
// Create the API client
client := agent.APIClient{
Endpoint: cfg.Endpoint,
Token: cfg.AgentAccessToken,
}.Create()
// Create the meta data to set
metaData := &api.MetaData{
Key: cfg.Key,
Value: cfg.Value,
}
// Set the meta data
err := retry.Do(func(s *retry.Stats) error {
resp, err := client.MetaData.Set(cfg.Job, metaData)
if resp != nil && (resp.StatusCode == 401 || resp.StatusCode == 404) {
s.Break()
}
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
if err != nil {
logger.Fatal("Failed to set meta-data: %s", err)
}
},
}
func (a *ArtifactUploader) upload(artifacts []*api.Artifact) error {
var uploader Uploader
// Determine what uploader to use
if a.Destination != "" {
if strings.HasPrefix(a.Destination, "s3://") {
uploader = new(S3Uploader)
} else {
return errors.New("Unknown upload destination: " + a.Destination)
}
} else {
uploader = new(FormUploader)
}
// Setup the uploader
err := uploader.Setup(a.Destination)
if err != nil {
return err
}
// Set the URL's of the artifacts based on the uploader
for _, artifact := range artifacts {
artifact.URL = uploader.URL(artifact)
}
// Create the artifacts on Buildkite
batchCreator := ArtifactBatchCreator{
APIClient: a.APIClient,
JobID: a.JobID,
Artifacts: artifacts,
}
artifacts, err = batchCreator.Create()
if err != nil {
return err
}
p := pool.New(pool.MaxConcurrencyLimit)
errors := []error{}
for _, artifact := range artifacts {
// Create new instance of the artifact for the goroutine
// See: http://golang.org/doc/effective_go.html#channels
artifact := artifact
p.Spawn(func() {
// Show a nice message that we're starting to upload the file
logger.Info("Uploading \"%s\" %d bytes", artifact.Path, artifact.FileSize)
// Upload the artifact and then set the state depending
// on whether or not it passed. We'll retry the upload
// a couple of times before giving up.
err = retry.Do(func(s *retry.Stats) error {
err := uploader.Upload(artifact)
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
if err != nil {
artifact.State = "error"
logger.Error("Error uploading artifact \"%s\": %s", artifact.Path, err)
// Track the error that was raised
p.Lock()
errors = append(errors, err)
p.Unlock()
} else {
artifact.State = "finished"
}
// Update the state of the artifact on Buildkite, we
// retry this as well.
err = retry.Do(func(s *retry.Stats) error {
_, _, err = a.APIClient.Artifacts.Update(a.JobID, artifact)
if err != nil {
logger.Warn("%s (%s)", err, s)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
if err != nil {
logger.Error("Error marking artifact %s as uploaded: %s", artifact.Path, err)
// Track the error that was raised
p.Lock()
errors = append(errors, err)
p.Unlock()
}
})
}
p.Wait()
if len(errors) > 0 {
logger.Fatal("There were errors with uploading some of the artifacts")
}
return nil
}
// 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
}
// PostProcess sends the Artifact to Buildkite
func (p *PostProcessor) PostProcess(ui packer.Ui, artifact packer.Artifact) (packer.Artifact, bool, error) {
buildkiteEnabled := os.Getenv("BUILDKITE")
if buildkiteEnabled != "true" {
ui.Message("This Packer build is not run within Buildkite. Skipping the BuildKite post-processor.")
return artifact, true, nil
}
client := p.client.Create()
metadata := make([]buildkite.MetaData, 0, 2)
builderID := artifact.BuilderId()
// packer's artifact model doesn't support AMIs very well since
// both a region and an id need to be encoded into a single field.
// We handle it as a special case so we can make more useful metadata
// in the buildkite job.
// N.B. this doesn't support AMIs being created across multiple regions.
if builderID == "mitchellh.amazonebs" || builderID == "mitchellh.amazoninstance" || builderID == "mitchellh.amazonchroot" {
packedID := artifact.Id()
parts := strings.SplitN(packedID, ":", 2)
idKeyName := "artifact_ami_id"
regionKeyName := "artifact_ami_region"
// override key names with an optional prefixed version
if p.config.Prefix != "" {
idKeyName = fmt.Sprintf("%s_id", p.config.Prefix)
regionKeyName = fmt.Sprintf("%s_region", p.config.Prefix)
}
metadata = append(metadata, buildkite.MetaData{
Key: idKeyName,
Value: parts[1],
})
metadata = append(metadata, buildkite.MetaData{
Key: regionKeyName,
Value: parts[0],
})
} else {
id := artifact.Id()
idKeyName := "artifact_id"
// override key name with an optional prefixed version
if p.config.Prefix != "" {
idKeyName = fmt.Sprintf("%s_id", p.config.Prefix)
}
if id != "" {
metadata = append(metadata, buildkite.MetaData{
Key: idKeyName,
Value: id,
})
}
// TODO: Support uploading files as BuildKite artifacts
}
if len(metadata) > 0 {
ui.Message("Setting metadata in BuildKite:")
for _, item := range metadata {
ui.Message(fmt.Sprintf("- %s: %s", item.Key, item.Value))
err := retry.Do(func(s *retry.Stats) error {
resp, err := client.MetaData.Set(p.jobID, &item)
if resp != nil && (resp.StatusCode == 401 || resp.StatusCode == 404) {
s.Break()
}
if err != nil {
log.Println(err)
}
return err
}, &retry.Config{Maximum: 10, Interval: 1 * time.Second})
if err != nil {
return nil, false, err
}
}
}
return artifact, true, nil
}