"fmt"
"math/big"
"sync"
+ "sync/atomic"
+ "time"
"git.arvados.org/arvados.git/lib/cloud"
"git.arvados.org/arvados.git/sdk/go/arvados"
"github.com/aws/aws-sdk-go/aws"
+ "github.com/aws/aws-sdk-go/aws/awserr"
"github.com/aws/aws-sdk-go/aws/credentials"
+ "github.com/aws/aws-sdk-go/aws/credentials/ec2rolecreds"
+ "github.com/aws/aws-sdk-go/aws/ec2metadata"
+ "github.com/aws/aws-sdk-go/aws/request"
"github.com/aws/aws-sdk-go/aws/session"
"github.com/aws/aws-sdk-go/service/ec2"
"github.com/sirupsen/logrus"
// Driver is the ec2 implementation of the cloud.Driver interface.
var Driver = cloud.DriverFunc(newEC2InstanceSet)
+const (
+ throttleDelayMin = time.Second
+ throttleDelayMax = time.Minute
+)
+
type ec2InstanceSetConfig struct {
AccessKeyID string
SecretAccessKey string
}
type ec2InstanceSet struct {
- ec2config ec2InstanceSetConfig
- instanceSetID cloud.InstanceSetID
- logger logrus.FieldLogger
- client ec2Interface
- keysMtx sync.Mutex
- keys map[string]string
+ ec2config ec2InstanceSetConfig
+ instanceSetID cloud.InstanceSetID
+ logger logrus.FieldLogger
+ client ec2Interface
+ keysMtx sync.Mutex
+ keys map[string]string
+ throttleDelayCreate atomic.Value
+ throttleDelayInstances atomic.Value
}
func newEC2InstanceSet(config json.RawMessage, instanceSetID cloud.InstanceSetID, _ cloud.SharedResourceTags, logger logrus.FieldLogger) (prv cloud.InstanceSet, err error) {
if err != nil {
return nil, err
}
- awsConfig := aws.NewConfig().
- WithCredentials(credentials.NewStaticCredentials(
- instanceSet.ec2config.AccessKeyID,
- instanceSet.ec2config.SecretAccessKey,
- "")).
- WithRegion(instanceSet.ec2config.Region)
+
+ sess, err := session.NewSession()
+ if err != nil {
+ return nil, err
+ }
+ // First try any static credentials, fall back to an IAM instance profile/role
+ creds := credentials.NewChainCredentials(
+ []credentials.Provider{
+ &credentials.StaticProvider{Value: credentials.Value{AccessKeyID: instanceSet.ec2config.AccessKeyID, SecretAccessKey: instanceSet.ec2config.SecretAccessKey}},
+ &ec2rolecreds.EC2RoleProvider{Client: ec2metadata.New(sess)},
+ })
+
+ awsConfig := aws.NewConfig().WithCredentials(creds).WithRegion(instanceSet.ec2config.Region)
instanceSet.client = ec2.New(session.Must(session.NewSession(awsConfig)))
instanceSet.keys = make(map[string]string)
if instanceSet.ec2config.EBSVolumeType == "" {
sha1pkix := sha1.Sum([]byte(pkix))
md5fp = ""
sha1fp = ""
- for i := 0; i < len(md5pkix); i += 1 {
+ for i := 0; i < len(md5pkix); i++ {
md5fp += fmt.Sprintf(":%02x", md5pkix[i])
}
- for i := 0; i < len(sha1pkix); i += 1 {
+ for i := 0; i < len(sha1pkix); i++ {
sha1fp += fmt.Sprintf(":%02x", sha1pkix[i])
}
return md5fp[1:], sha1fp[1:], nil
}
rsv, err := instanceSet.client.RunInstances(&rii)
-
+ err = wrapError(err, &instanceSet.throttleDelayCreate)
if err != nil {
return nil, err
}
dii := &ec2.DescribeInstancesInput{Filters: filters}
for {
dio, err := instanceSet.client.DescribeInstances(dii)
+ err = wrapError(err, &instanceSet.throttleDelayInstances)
if err != nil {
return nil, err
}
}
}
-func (az *ec2InstanceSet) Stop() {
+func (instanceSet *ec2InstanceSet) Stop() {
}
type ec2Instance struct {
func (inst *ec2Instance) Address() string {
if inst.instance.PrivateIpAddress != nil {
return *inst.instance.PrivateIpAddress
- } else {
- return ""
}
+ return ""
}
func (inst *ec2Instance) RemoteUser() string {
func (inst *ec2Instance) VerifyHostKey(ssh.PublicKey, *ssh.Client) error {
return cloud.ErrNotImplemented
}
+
+type rateLimitError struct {
+ error
+ earliestRetry time.Time
+}
+
+func (err rateLimitError) EarliestRetry() time.Time {
+ return err.earliestRetry
+}
+
+var isCodeCapacity = map[string]bool{
+ "InsufficientInstanceCapacity": true,
+ "VcpuLimitExceeded": true,
+ "MaxSpotInstanceCountExceeded": true,
+}
+
+// isErrorCapacity returns whether the error is to be throttled based on its code.
+// Returns false if error is nil.
+func isErrorCapacity(err error) bool {
+ if aerr, ok := err.(awserr.Error); ok && aerr != nil {
+ if _, ok := isCodeCapacity[aerr.Code()]; ok {
+ return true
+ }
+ }
+ return false
+}
+
+type ec2QuotaError struct {
+ error
+}
+
+func (er *ec2QuotaError) IsQuotaError() bool {
+ return true
+}
+
+func wrapError(err error, throttleValue *atomic.Value) error {
+ if request.IsErrorThrottle(err) {
+ // Back off exponentially until an upstream call
+ // either succeeds or returns a non-throttle error.
+ d, _ := throttleValue.Load().(time.Duration)
+ d = d*3/2 + time.Second
+ if d < throttleDelayMin {
+ d = throttleDelayMin
+ } else if d > throttleDelayMax {
+ d = throttleDelayMax
+ }
+ throttleValue.Store(d)
+ return rateLimitError{error: err, earliestRetry: time.Now().Add(d)}
+ } else if isErrorCapacity(err) {
+ return &ec2QuotaError{err}
+ } else if err != nil {
+ throttleValue.Store(time.Duration(0))
+ return err
+ }
+ throttleValue.Store(time.Duration(0))
+ return nil
+}