lib/cloud/ec2/ec2_test.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4 //
   5 //
   6 // How to manually run individual tests against the real cloud:
   7 //
   8 // $ go test -v git.arvados.org/arvados.git/lib/cloud/ec2 -live-ec2-cfg ec2config.yml -check.f=TestCreate
   9 //
  10 // Tests should be run individually and in the order they are listed in the file:
  11 //
  12 // Example ec2config.yml:
  13 //
  14 // ImageIDForTestSuite: ami-xxxxxxxxxxxxxxxxx
  15 // DriverParameters:
  16 //       AccessKeyID: XXXXXXXXXXXXXX
  17 //       SecretAccessKey: xxxxxxxxxxxxxxxxxxxx
  18 //       Region: us-east-1
  19 //       SecurityGroupIDs: [sg-xxxxxxxx]
  20 //       SubnetID: subnet-xxxxxxxx
  21 //       AdminUsername: crunch
  22
  23 package ec2
  24
  25 import (
  26         "encoding/json"
  27         "flag"
  28         "sync/atomic"
  29         "testing"
  30         "time"
  31
  32         "git.arvados.org/arvados.git/lib/cloud"
  33         "git.arvados.org/arvados.git/lib/dispatchcloud/test"
  34         "git.arvados.org/arvados.git/sdk/go/arvados"
  35         "git.arvados.org/arvados.git/sdk/go/config"
  36         "github.com/aws/aws-sdk-go/aws"
  37         "github.com/aws/aws-sdk-go/aws/awserr"
  38         "github.com/aws/aws-sdk-go/service/ec2"
  39         "github.com/sirupsen/logrus"
  40         check "gopkg.in/check.v1"
  41 )
  42
  43 var live = flag.String("live-ec2-cfg", "", "Test with real EC2 API, provide config file")
  44
  45 // Gocheck boilerplate
  46 func Test(t *testing.T) {
  47         check.TestingT(t)
  48 }
  49
  50 type EC2InstanceSetSuite struct{}
  51
  52 var _ = check.Suite(&EC2InstanceSetSuite{})
  53
  54 type testConfig struct {
  55         ImageIDForTestSuite string
  56         DriverParameters    json.RawMessage
  57 }
  58
  59 type ec2stub struct {
  60         c       *check.C
  61         reftime time.Time
  62 }
  63
  64 func (e *ec2stub) ImportKeyPair(input *ec2.ImportKeyPairInput) (*ec2.ImportKeyPairOutput, error) {
  65         return nil, nil
  66 }
  67
  68 func (e *ec2stub) DescribeKeyPairs(input *ec2.DescribeKeyPairsInput) (*ec2.DescribeKeyPairsOutput, error) {
  69         return &ec2.DescribeKeyPairsOutput{}, nil
  70 }
  71
  72 func (e *ec2stub) RunInstances(input *ec2.RunInstancesInput) (*ec2.Reservation, error) {
  73         return &ec2.Reservation{Instances: []*ec2.Instance{{
  74                 InstanceId:   aws.String("i-123"),
  75                 InstanceType: aws.String("t2.micro"),
  76                 Tags:         input.TagSpecifications[0].Tags,
  77         }}}, nil
  78 }
  79
  80 func (e *ec2stub) DescribeInstances(input *ec2.DescribeInstancesInput) (*ec2.DescribeInstancesOutput, error) {
  81         return &ec2.DescribeInstancesOutput{
  82                 Reservations: []*ec2.Reservation{{
  83                         Instances: []*ec2.Instance{{
  84                                 InstanceId:        aws.String("i-123"),
  85                                 InstanceLifecycle: aws.String("spot"),
  86                                 InstanceType:      aws.String("t2.micro"),
  87                                 PrivateIpAddress:  aws.String("10.1.2.3"),
  88                                 State:             &ec2.InstanceState{Name: aws.String("running")},
  89                         }, {
  90                                 InstanceId:        aws.String("i-124"),
  91                                 InstanceLifecycle: aws.String("spot"),
  92                                 InstanceType:      aws.String("t2.micro"),
  93                                 PrivateIpAddress:  aws.String("10.1.2.4"),
  94                                 State:             &ec2.InstanceState{Name: aws.String("running")},
  95                         }},
  96                 }},
  97         }, nil
  98 }
  99
 100 func (e *ec2stub) DescribeInstanceStatusPages(input *ec2.DescribeInstanceStatusInput, fn func(*ec2.DescribeInstanceStatusOutput, bool) bool) error {
 101         fn(&ec2.DescribeInstanceStatusOutput{
 102                 InstanceStatuses: []*ec2.InstanceStatus{{
 103                         InstanceId:       aws.String("i-123"),
 104                         AvailabilityZone: aws.String("aa-east-1a"),
 105                 }, {
 106                         InstanceId:       aws.String("i-124"),
 107                         AvailabilityZone: aws.String("aa-east-1a"),
 108                 }},
 109         }, true)
 110         return nil
 111 }
 112
 113 func (e *ec2stub) DescribeSpotPriceHistoryPages(input *ec2.DescribeSpotPriceHistoryInput, fn func(*ec2.DescribeSpotPriceHistoryOutput, bool) bool) error {
 114         if !fn(&ec2.DescribeSpotPriceHistoryOutput{
 115                 SpotPriceHistory: []*ec2.SpotPrice{
 116                         &ec2.SpotPrice{
 117                                 InstanceType:     aws.String("t2.micro"),
 118                                 AvailabilityZone: aws.String("aa-east-1a"),
 119                                 SpotPrice:        aws.String("0.005"),
 120                                 Timestamp:        aws.Time(e.reftime.Add(-9 * time.Minute)),
 121                         },
 122                         &ec2.SpotPrice{
 123                                 InstanceType:     aws.String("t2.micro"),
 124                                 AvailabilityZone: aws.String("aa-east-1a"),
 125                                 SpotPrice:        aws.String("0.015"),
 126                                 Timestamp:        aws.Time(e.reftime.Add(-5 * time.Minute)),
 127                         },
 128                 },
 129         }, false) {
 130                 return nil
 131         }
 132         fn(&ec2.DescribeSpotPriceHistoryOutput{
 133                 SpotPriceHistory: []*ec2.SpotPrice{
 134                         &ec2.SpotPrice{
 135                                 InstanceType:     aws.String("t2.micro"),
 136                                 AvailabilityZone: aws.String("aa-east-1a"),
 137                                 SpotPrice:        aws.String("0.01"),
 138                                 Timestamp:        aws.Time(e.reftime.Add(-2 * time.Minute)),
 139                         },
 140                 },
 141         }, true)
 142         return nil
 143 }
 144
 145 func (e *ec2stub) CreateTags(input *ec2.CreateTagsInput) (*ec2.CreateTagsOutput, error) {
 146         return nil, nil
 147 }
 148
 149 func (e *ec2stub) TerminateInstances(input *ec2.TerminateInstancesInput) (*ec2.TerminateInstancesOutput, error) {
 150         return nil, nil
 151 }
 152
 153 func GetInstanceSet(c *check.C) (*ec2InstanceSet, cloud.ImageID, arvados.Cluster) {
 154         cluster := arvados.Cluster{
 155                 InstanceTypes: arvados.InstanceTypeMap(map[string]arvados.InstanceType{
 156                         "tiny": {
 157                                 Name:         "tiny",
 158                                 ProviderType: "t2.micro",
 159                                 VCPUs:        1,
 160                                 RAM:          4000000000,
 161                                 Scratch:      10000000000,
 162                                 Price:        .02,
 163                                 Preemptible:  false,
 164                         },
 165                         "tiny-with-extra-scratch": {
 166                                 Name:         "tiny-with-extra-scratch",
 167                                 ProviderType: "t2.micro",
 168                                 VCPUs:        1,
 169                                 RAM:          4000000000,
 170                                 Price:        .02,
 171                                 Preemptible:  false,
 172                                 AddedScratch: 20000000000,
 173                         },
 174                         "tiny-preemptible": {
 175                                 Name:         "tiny-preemptible",
 176                                 ProviderType: "t2.micro",
 177                                 VCPUs:        1,
 178                                 RAM:          4000000000,
 179                                 Scratch:      10000000000,
 180                                 Price:        .02,
 181                                 Preemptible:  true,
 182                         },
 183                 })}
 184         if *live != "" {
 185                 var exampleCfg testConfig
 186                 err := config.LoadFile(&exampleCfg, *live)
 187                 c.Assert(err, check.IsNil)
 188
 189                 ap, err := newEC2InstanceSet(exampleCfg.DriverParameters, "test123", nil, logrus.StandardLogger())
 190                 c.Assert(err, check.IsNil)
 191                 return ap.(*ec2InstanceSet), cloud.ImageID(exampleCfg.ImageIDForTestSuite), cluster
 192         }
 193         ap := ec2InstanceSet{
 194                 ec2config: ec2InstanceSetConfig{
 195                         SpotPriceUpdateInterval: arvados.Duration(time.Hour),
 196                 },
 197                 instanceSetID: "test123",
 198                 logger:        logrus.StandardLogger(),
 199                 client:        &ec2stub{c: c, reftime: time.Now().UTC()},
 200                 keys:          make(map[string]string),
 201         }
 202         return &ap, cloud.ImageID("blob"), cluster
 203 }
 204
 205 func (*EC2InstanceSetSuite) TestCreate(c *check.C) {
 206         ap, img, cluster := GetInstanceSet(c)
 207         pk, _ := test.LoadTestKey(c, "../../dispatchcloud/test/sshkey_dispatch")
 208
 209         inst, err := ap.Create(cluster.InstanceTypes["tiny"],
 210                 img, map[string]string{
 211                         "TestTagName": "test tag value",
 212                 }, "umask 0600; echo -n test-file-data >/var/run/test-file", pk)
 213         c.Assert(err, check.IsNil)
 214
 215         tags := inst.Tags()
 216         c.Check(tags["TestTagName"], check.Equals, "test tag value")
 217         c.Logf("inst.String()=%v Address()=%v Tags()=%v", inst.String(), inst.Address(), tags)
 218
 219 }
 220
 221 func (*EC2InstanceSetSuite) TestCreateWithExtraScratch(c *check.C) {
 222         ap, img, cluster := GetInstanceSet(c)
 223         pk, _ := test.LoadTestKey(c, "../../dispatchcloud/test/sshkey_dispatch")
 224
 225         inst, err := ap.Create(cluster.InstanceTypes["tiny-with-extra-scratch"],
 226                 img, map[string]string{
 227                         "TestTagName": "test tag value",
 228                 }, "umask 0600; echo -n test-file-data >/var/run/test-file", pk)
 229
 230         c.Assert(err, check.IsNil)
 231
 232         tags := inst.Tags()
 233         c.Check(tags["TestTagName"], check.Equals, "test tag value")
 234         c.Logf("inst.String()=%v Address()=%v Tags()=%v", inst.String(), inst.Address(), tags)
 235
 236 }
 237
 238 func (*EC2InstanceSetSuite) TestCreatePreemptible(c *check.C) {
 239         ap, img, cluster := GetInstanceSet(c)
 240         pk, _ := test.LoadTestKey(c, "../../dispatchcloud/test/sshkey_dispatch")
 241
 242         inst, err := ap.Create(cluster.InstanceTypes["tiny-preemptible"],
 243                 img, map[string]string{
 244                         "TestTagName": "test tag value",
 245                 }, "umask 0600; echo -n test-file-data >/var/run/test-file", pk)
 246
 247         c.Assert(err, check.IsNil)
 248
 249         tags := inst.Tags()
 250         c.Check(tags["TestTagName"], check.Equals, "test tag value")
 251         c.Logf("inst.String()=%v Address()=%v Tags()=%v", inst.String(), inst.Address(), tags)
 252
 253 }
 254
 255 func (*EC2InstanceSetSuite) TestTagInstances(c *check.C) {
 256         ap, _, _ := GetInstanceSet(c)
 257         l, err := ap.Instances(nil)
 258         c.Assert(err, check.IsNil)
 259
 260         for _, i := range l {
 261                 tg := i.Tags()
 262                 tg["TestTag2"] = "123 test tag 2"
 263                 c.Check(i.SetTags(tg), check.IsNil)
 264         }
 265 }
 266
 267 func (*EC2InstanceSetSuite) TestListInstances(c *check.C) {
 268         ap, _, _ := GetInstanceSet(c)
 269         l, err := ap.Instances(nil)
 270         c.Assert(err, check.IsNil)
 271
 272         for _, i := range l {
 273                 tg := i.Tags()
 274                 c.Logf("%v %v %v", i.String(), i.Address(), tg)
 275         }
 276 }
 277
 278 func (*EC2InstanceSetSuite) TestDestroyInstances(c *check.C) {
 279         ap, _, _ := GetInstanceSet(c)
 280         l, err := ap.Instances(nil)
 281         c.Assert(err, check.IsNil)
 282
 283         for _, i := range l {
 284                 c.Check(i.Destroy(), check.IsNil)
 285         }
 286 }
 287
 288 func (*EC2InstanceSetSuite) TestInstancePriceHistory(c *check.C) {
 289         ap, img, cluster := GetInstanceSet(c)
 290         pk, _ := test.LoadTestKey(c, "../../dispatchcloud/test/sshkey_dispatch")
 291         tags := cloud.InstanceTags{"arvados-ec2-driver": "test"}
 292
 293         defer func() {
 294                 instances, err := ap.Instances(tags)
 295                 c.Assert(err, check.IsNil)
 296                 for _, inst := range instances {
 297                         c.Logf("cleanup: destroy instance %s", inst)
 298                         c.Check(inst.Destroy(), check.IsNil)
 299                 }
 300         }()
 301
 302         ap.ec2config.EBSPrice = 0.1 // $/GiB/month
 303         inst1, err := ap.Create(cluster.InstanceTypes["tiny-preemptible"], img, tags, "true", pk)
 304         c.Assert(err, check.IsNil)
 305         defer inst1.Destroy()
 306         inst2, err := ap.Create(cluster.InstanceTypes["tiny-preemptible"], img, tags, "true", pk)
 307         c.Assert(err, check.IsNil)
 308         defer inst2.Destroy()
 309
 310         // in live mode, we need to wait for the instances to reach
 311         // running state before we can discover their availability
 312         // zones and look up the appropriate prices.
 313         var instances []cloud.Instance
 314         for deadline := time.Now().Add(5 * time.Minute); ; {
 315                 if deadline.Before(time.Now()) {
 316                         c.Fatal("timed out")
 317                 }
 318                 instances, err = ap.Instances(tags)
 319                 running := 0
 320                 for _, inst := range instances {
 321                         if *inst.(*ec2Instance).instance.InstanceLifecycle == "spot" {
 322                                 running++
 323                         }
 324                 }
 325                 if running >= 2 {
 326                         c.Logf("instances are running, and identifiable as spot instances")
 327                         break
 328                 }
 329                 c.Logf("waiting for instances to be identifiable as spot instances...")
 330                 time.Sleep(10 * time.Second)
 331         }
 332
 333         for _, inst := range instances {
 334                 hist := inst.PriceHistory(arvados.InstanceType{})
 335                 c.Logf("%s price history: %v", inst.ID(), hist)
 336                 c.Check(len(hist) > 0, check.Equals, true)
 337
 338                 histWithScratch := inst.PriceHistory(arvados.InstanceType{AddedScratch: 640 << 30})
 339                 c.Logf("%s price history with 640 GiB scratch: %v", inst.ID(), histWithScratch)
 340
 341                 for i, ip := range hist {
 342                         c.Check(ip.Price, check.Not(check.Equals), 0.0)
 343                         if i > 0 {
 344                                 c.Check(ip.StartTime.Before(hist[i-1].StartTime), check.Equals, true)
 345                         }
 346                         c.Check(ip.Price < histWithScratch[i].Price, check.Equals, true)
 347                 }
 348         }
 349 }
 350
 351 func (*EC2InstanceSetSuite) TestWrapError(c *check.C) {
 352         retryError := awserr.New("Throttling", "", nil)
 353         wrapped := wrapError(retryError, &atomic.Value{})
 354         _, ok := wrapped.(cloud.RateLimitError)
 355         c.Check(ok, check.Equals, true)
 356
 357         quotaError := awserr.New("InsufficientInstanceCapacity", "", nil)
 358         wrapped = wrapError(quotaError, nil)
 359         _, ok = wrapped.(cloud.QuotaError)
 360         c.Check(ok, check.Equals, true)
 361 }