X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/af6447c82f24b283b6b5f32b2b66dc6a47e24487..c89213f5a5e303050caaebe4f8fdf2980fc65605:/lib/cloud/ec2/ec2.go diff --git a/lib/cloud/ec2/ec2.go b/lib/cloud/ec2/ec2.go index c5565d4245..269a7d8def 100644 --- a/lib/cloud/ec2/ec2.go +++ b/lib/cloud/ec2/ec2.go @@ -13,30 +13,37 @@ import ( "encoding/json" "fmt" "math/big" - "strings" "sync" + "sync/atomic" + "time" - "git.curoverse.com/arvados.git/lib/cloud" - "git.curoverse.com/arvados.git/sdk/go/arvados" + "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" "golang.org/x/crypto/ssh" ) -const arvadosDispatchID = "arvados-dispatch-id" -const tagPrefix = "arvados-dispatch-tag-" - // 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 Region string - SecurityGroupIDs []string + SecurityGroupIDs arvados.StringSet SubnetID string AdminUsername string EBSVolumeType string @@ -52,29 +59,38 @@ type ec2Interface interface { } type ec2InstanceSet struct { - ec2config ec2InstanceSetConfig - dispatcherID 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, dispatcherID cloud.InstanceSetID, logger logrus.FieldLogger) (prv cloud.InstanceSet, err error) { +func newEC2InstanceSet(config json.RawMessage, instanceSetID cloud.InstanceSetID, _ cloud.SharedResourceTags, logger logrus.FieldLogger) (prv cloud.InstanceSet, err error) { instanceSet := &ec2InstanceSet{ - dispatcherID: dispatcherID, - logger: logger, + instanceSetID: instanceSetID, + logger: logger, } err = json.Unmarshal(config, &instanceSet.ec2config) 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 == "" { @@ -107,10 +123,10 @@ func awsKeyFingerprint(pk ssh.PublicKey) (md5fp string, sha1fp string, err error 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 @@ -132,7 +148,7 @@ func (instanceSet *ec2InstanceSet) Create( var ok bool if keyname, ok = instanceSet.keys[md5keyFingerprint]; !ok { keyout, err := instanceSet.client.DescribeKeyPairs(&ec2.DescribeKeyPairsInput{ - Filters: []*ec2.Filter{&ec2.Filter{ + Filters: []*ec2.Filter{{ Name: aws.String("fingerprint"), Values: []*string{&md5keyFingerprint, &sha1keyFingerprint}, }}, @@ -157,23 +173,19 @@ func (instanceSet *ec2InstanceSet) Create( } instanceSet.keysMtx.Unlock() - ec2tags := []*ec2.Tag{ - &ec2.Tag{ - Key: aws.String(arvadosDispatchID), - Value: aws.String(string(instanceSet.dispatcherID)), - }, - &ec2.Tag{ - Key: aws.String("arvados-class"), - Value: aws.String("dynamic-compute"), - }, - } + ec2tags := []*ec2.Tag{} for k, v := range newTags { ec2tags = append(ec2tags, &ec2.Tag{ - Key: aws.String(tagPrefix + k), + Key: aws.String(k), Value: aws.String(v), }) } + var groups []string + for sg := range instanceSet.ec2config.SecurityGroupIDs { + groups = append(groups, sg) + } + rii := ec2.RunInstancesInput{ ImageId: aws.String(string(imageID)), InstanceType: &instanceType.ProviderType, @@ -182,25 +194,25 @@ func (instanceSet *ec2InstanceSet) Create( KeyName: &keyname, NetworkInterfaces: []*ec2.InstanceNetworkInterfaceSpecification{ - &ec2.InstanceNetworkInterfaceSpecification{ + { AssociatePublicIpAddress: aws.Bool(false), DeleteOnTermination: aws.Bool(true), DeviceIndex: aws.Int64(0), - Groups: aws.StringSlice(instanceSet.ec2config.SecurityGroupIDs), + Groups: aws.StringSlice(groups), SubnetId: &instanceSet.ec2config.SubnetID, }}, DisableApiTermination: aws.Bool(false), InstanceInitiatedShutdownBehavior: aws.String("terminate"), - UserData: aws.String(base64.StdEncoding.EncodeToString([]byte("#!/bin/sh\n" + initCommand + "\n"))), TagSpecifications: []*ec2.TagSpecification{ - &ec2.TagSpecification{ + { ResourceType: aws.String("instance"), Tags: ec2tags, }}, + UserData: aws.String(base64.StdEncoding.EncodeToString([]byte("#!/bin/sh\n" + initCommand + "\n"))), } if instanceType.AddedScratch > 0 { - rii.BlockDeviceMappings = []*ec2.BlockDeviceMapping{&ec2.BlockDeviceMapping{ + rii.BlockDeviceMappings = []*ec2.BlockDeviceMapping{{ DeviceName: aws.String("/dev/xvdt"), Ebs: &ec2.EbsBlockDevice{ DeleteOnTermination: aws.Bool(true), @@ -219,7 +231,7 @@ func (instanceSet *ec2InstanceSet) Create( } rsv, err := instanceSet.client.RunInstances(&rii) - + err = wrapError(err, &instanceSet.throttleDelayCreate) if err != nil { return nil, err } @@ -230,15 +242,18 @@ func (instanceSet *ec2InstanceSet) Create( }, nil } -func (instanceSet *ec2InstanceSet) Instances(cloud.InstanceTags) (instances []cloud.Instance, err error) { - dii := &ec2.DescribeInstancesInput{ - Filters: []*ec2.Filter{&ec2.Filter{ - Name: aws.String("tag:" + arvadosDispatchID), - Values: []*string{aws.String(string(instanceSet.dispatcherID))}, - }}} - +func (instanceSet *ec2InstanceSet) Instances(tags cloud.InstanceTags) (instances []cloud.Instance, err error) { + var filters []*ec2.Filter + for k, v := range tags { + filters = append(filters, &ec2.Filter{ + Name: aws.String("tag:" + k), + Values: []*string{aws.String(v)}, + }) + } + dii := &ec2.DescribeInstancesInput{Filters: filters} for { dio, err := instanceSet.client.DescribeInstances(dii) + err = wrapError(err, &instanceSet.throttleDelayInstances) if err != nil { return nil, err } @@ -257,7 +272,7 @@ func (instanceSet *ec2InstanceSet) Instances(cloud.InstanceTags) (instances []cl } } -func (az *ec2InstanceSet) Stop() { +func (instanceSet *ec2InstanceSet) Stop() { } type ec2Instance struct { @@ -278,15 +293,10 @@ func (inst *ec2Instance) ProviderType() string { } func (inst *ec2Instance) SetTags(newTags cloud.InstanceTags) error { - ec2tags := []*ec2.Tag{ - &ec2.Tag{ - Key: aws.String(arvadosDispatchID), - Value: aws.String(string(inst.provider.dispatcherID)), - }, - } + var ec2tags []*ec2.Tag for k, v := range newTags { ec2tags = append(ec2tags, &ec2.Tag{ - Key: aws.String(tagPrefix + k), + Key: aws.String(k), Value: aws.String(v), }) } @@ -303,9 +313,7 @@ func (inst *ec2Instance) Tags() cloud.InstanceTags { tags := make(map[string]string) for _, t := range inst.instance.Tags { - if strings.HasPrefix(*t.Key, tagPrefix) { - tags[(*t.Key)[len(tagPrefix):]] = *t.Value - } + tags[*t.Key] = *t.Value } return tags @@ -321,9 +329,8 @@ func (inst *ec2Instance) Destroy() error { func (inst *ec2Instance) Address() string { if inst.instance.PrivateIpAddress != nil { return *inst.instance.PrivateIpAddress - } else { - return "" } + return "" } func (inst *ec2Instance) RemoteUser() string { @@ -333,3 +340,60 @@ 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 +}