func (db *SQLDB) ClaimActualLRP(logger lager.Logger, processGuid string, index int32, instanceKey *models.ActualLRPInstanceKey) (*models.ActualLRPGroup, *models.ActualLRPGroup, error) {
logger = logger.WithData(lager.Data{"process_guid": processGuid, "index": index, "instance_key": instanceKey})
logger.Info("starting")
defer logger.Info("complete")
var beforeActualLRP models.ActualLRP
var actualLRP *models.ActualLRP
err := db.transact(logger, func(logger lager.Logger, tx *sql.Tx) error {
var err error
actualLRP, err = db.fetchActualLRPForUpdate(logger, processGuid, index, false, tx)
if err != nil {
logger.Error("failed-fetching-actual-lrp-for-share", err)
return err
}
beforeActualLRP = *actualLRP
if !actualLRP.AllowsTransitionTo(&actualLRP.ActualLRPKey, instanceKey, models.ActualLRPStateClaimed) {
logger.Error("cannot-transition-to-claimed", nil, lager.Data{"from_state": actualLRP.State, "same_instance_key": actualLRP.ActualLRPInstanceKey.Equal(instanceKey)})
return models.ErrActualLRPCannotBeClaimed
}
if actualLRP.State == models.ActualLRPStateClaimed && actualLRP.ActualLRPInstanceKey.Equal(instanceKey) {
return nil
}
actualLRP.ModificationTag.Increment()
actualLRP.State = models.ActualLRPStateClaimed
actualLRP.ActualLRPInstanceKey = *instanceKey
actualLRP.PlacementError = ""
actualLRP.ActualLRPNetInfo = models.ActualLRPNetInfo{}
actualLRP.Since = db.clock.Now().UnixNano()
_, err = db.update(logger, tx, actualLRPsTable,
SQLAttributes{
"state": actualLRP.State,
"cell_id": actualLRP.CellId,
"instance_guid": actualLRP.InstanceGuid,
"modification_tag_index": actualLRP.ModificationTag.Index,
"placement_error": actualLRP.PlacementError,
"since": actualLRP.Since,
"net_info": []byte{},
},
"process_guid = ? AND instance_index = ? AND evacuating = ?",
processGuid, index, false,
)
if err != nil {
logger.Error("failed-claiming-actual-lrp", err)
return db.convertSQLError(err)
}
return nil
})
return &models.ActualLRPGroup{Instance: &beforeActualLRP}, &models.ActualLRPGroup{Instance: actualLRP}, err
}
func (db *SQLDB) UnclaimActualLRP(logger lager.Logger, key *models.ActualLRPKey) (*models.ActualLRPGroup, *models.ActualLRPGroup, error) {
logger = logger.WithData(lager.Data{"key": key})
var beforeActualLRP models.ActualLRP
var actualLRP *models.ActualLRP
processGuid := key.ProcessGuid
index := key.Index
err := db.transact(logger, func(logger lager.Logger, tx *sql.Tx) error {
var err error
actualLRP, err = db.fetchActualLRPForUpdate(logger, processGuid, index, false, tx)
if err != nil {
logger.Error("failed-fetching-actual-lrp-for-share", err)
return err
}
beforeActualLRP = *actualLRP
if actualLRP.State == models.ActualLRPStateUnclaimed {
logger.Debug("already-unclaimed")
return models.ErrActualLRPCannotBeUnclaimed
}
logger.Info("starting")
defer logger.Info("complete")
now := db.clock.Now().UnixNano()
actualLRP.ModificationTag.Increment()
actualLRP.State = models.ActualLRPStateUnclaimed
actualLRP.ActualLRPInstanceKey.CellId = ""
actualLRP.ActualLRPInstanceKey.InstanceGuid = ""
actualLRP.Since = now
actualLRP.ActualLRPNetInfo = models.ActualLRPNetInfo{}
_, err = db.update(logger, tx, actualLRPsTable,
SQLAttributes{
"state": actualLRP.State,
"cell_id": actualLRP.CellId,
"instance_guid": actualLRP.InstanceGuid,
"modification_tag_index": actualLRP.ModificationTag.Index,
"since": actualLRP.Since,
"net_info": []byte{},
},
"process_guid = ? AND instance_index = ? AND evacuating = ?",
processGuid, index, false,
)
if err != nil {
logger.Error("failed-to-unclaim-actual-lrp", err)
return db.convertSQLError(err)
}
return nil
})
return &models.ActualLRPGroup{Instance: &beforeActualLRP}, &models.ActualLRPGroup{Instance: actualLRP}, err
}
func (db *ETCDDB) unclaimActualLRPWithIndex(
logger lager.Logger,
lrp *models.ActualLRP,
storeIndex uint64,
actualLRPKey *models.ActualLRPKey,
actualLRPInstanceKey *models.ActualLRPInstanceKey,
) (change stateChange, err error) {
logger = logger.Session("unclaim-actual-lrp-with-index")
defer logger.Debug("complete", lager.Data{"state_change": change, "error": err})
if !lrp.ActualLRPKey.Equal(actualLRPKey) {
logger.Error("failed-actual-lrp-key-differs", models.ErrActualLRPCannotBeUnclaimed)
return stateDidNotChange, models.ErrActualLRPCannotBeUnclaimed
}
if lrp.State == models.ActualLRPStateUnclaimed {
logger.Info("already-unclaimed")
return stateDidNotChange, nil
}
if !lrp.ActualLRPInstanceKey.Equal(actualLRPInstanceKey) {
logger.Error("failed-actual-lrp-instance-key-differs", models.ErrActualLRPCannotBeUnclaimed)
return stateDidNotChange, models.ErrActualLRPCannotBeUnclaimed
}
lrp.Since = db.clock.Now().UnixNano()
lrp.State = models.ActualLRPStateUnclaimed
lrp.ActualLRPInstanceKey = models.ActualLRPInstanceKey{}
lrp.ActualLRPNetInfo = models.EmptyActualLRPNetInfo()
lrp.ModificationTag.Increment()
err = lrp.Validate()
if err != nil {
logger.Error("failed-to-validate-unclaimed-lrp", err)
return stateDidNotChange, models.NewError(models.Error_InvalidRecord, err.Error())
}
lrpData, serialErr := db.serializeModel(logger, lrp)
if serialErr != nil {
logger.Error("failed-to-marshal-unclaimed-lrp", serialErr)
return stateDidNotChange, serialErr
}
_, err = db.client.CompareAndSwap(ActualLRPSchemaPath(actualLRPKey.ProcessGuid, actualLRPKey.Index), lrpData, 0, storeIndex)
if err != nil {
logger.Error("failed-to-compare-and-swap", err)
return stateDidNotChange, models.ErrActualLRPCannotBeUnclaimed
}
logger.Debug("changed-to-unclaimed")
return stateDidChange, nil
}
func (db *SQLDB) FailActualLRP(logger lager.Logger, key *models.ActualLRPKey, placementError string) (*models.ActualLRPGroup, *models.ActualLRPGroup, error) {
logger = logger.WithData(lager.Data{"actual_lrp_key": key, "placement_error": placementError})
logger.Info("starting")
defer logger.Info("complete")
var beforeActualLRP models.ActualLRP
var actualLRP *models.ActualLRP
err := db.transact(logger, func(logger lager.Logger, tx *sql.Tx) error {
var err error
actualLRP, err = db.fetchActualLRPForUpdate(logger, key.ProcessGuid, key.Index, false, tx)
if err != nil {
logger.Error("failed-to-get-actual-lrp", err)
return err
}
beforeActualLRP = *actualLRP
if actualLRP.State != models.ActualLRPStateUnclaimed {
logger.Error("cannot-fail-actual-lrp", nil, lager.Data{"from_state": actualLRP.State})
return models.ErrActualLRPCannotBeFailed
}
now := db.clock.Now().UnixNano()
actualLRP.ModificationTag.Increment()
actualLRP.PlacementError = placementError
actualLRP.Since = now
evacuating := false
_, err = db.update(logger, tx, actualLRPsTable,
SQLAttributes{
"modification_tag_index": actualLRP.ModificationTag.Index,
"placement_error": truncateString(actualLRP.PlacementError, 1024),
"since": actualLRP.Since,
},
"process_guid = ? AND instance_index = ? AND evacuating = ?",
key.ProcessGuid, key.Index, evacuating,
)
if err != nil {
logger.Error("failed-failing-actual-lrp", err)
return db.convertSQLError(err)
}
return nil
})
return &models.ActualLRPGroup{Instance: &beforeActualLRP}, &models.ActualLRPGroup{Instance: actualLRP}, err
}
)
var _ = Describe("Evacuation", func() {
var (
actualLRP *models.ActualLRP
guid string
index int32
)
BeforeEach(func() {
guid = "some-guid"
index = int32(1)
actualLRP = model_helpers.NewValidActualLRP(guid, index)
actualLRP.CrashCount = 0
actualLRP.CrashReason = ""
actualLRP.Since = fakeClock.Now().UnixNano()
actualLRP.ModificationTag = models.ModificationTag{}
actualLRP.ModificationTag.Increment()
actualLRP.ModificationTag.Increment()
_, err := sqlDB.CreateUnclaimedActualLRP(logger, &actualLRP.ActualLRPKey)
Expect(err).NotTo(HaveOccurred())
_, _, err = sqlDB.ClaimActualLRP(logger, guid, index, &actualLRP.ActualLRPInstanceKey)
Expect(err).NotTo(HaveOccurred())
_, _, err = sqlDB.StartActualLRP(logger, &actualLRP.ActualLRPKey, &actualLRP.ActualLRPInstanceKey, &actualLRP.ActualLRPNetInfo)
Expect(err).NotTo(HaveOccurred())
})
Describe("EvacuateActualLRP", func() {
var ttl uint64
func (db *SQLDB) CrashActualLRP(logger lager.Logger, key *models.ActualLRPKey, instanceKey *models.ActualLRPInstanceKey, crashReason string) (*models.ActualLRPGroup, *models.ActualLRPGroup, bool, error) {
logger = logger.WithData(lager.Data{"key": key, "instance_key": instanceKey, "crash_reason": crashReason})
logger.Info("starting")
defer logger.Info("complete")
var immediateRestart = false
var beforeActualLRP models.ActualLRP
var actualLRP *models.ActualLRP
err := db.transact(logger, func(logger lager.Logger, tx *sql.Tx) error {
var err error
actualLRP, err = db.fetchActualLRPForUpdate(logger, key.ProcessGuid, key.Index, false, tx)
if err != nil {
logger.Error("failed-to-get-actual-lrp", err)
return err
}
beforeActualLRP = *actualLRP
latestChangeTime := time.Duration(db.clock.Now().UnixNano() - actualLRP.Since)
var newCrashCount int32
if latestChangeTime > models.CrashResetTimeout && actualLRP.State == models.ActualLRPStateRunning {
newCrashCount = 1
} else {
newCrashCount = actualLRP.CrashCount + 1
}
if !actualLRP.AllowsTransitionTo(&actualLRP.ActualLRPKey, instanceKey, models.ActualLRPStateCrashed) {
logger.Error("failed-to-transition-to-crashed", nil, lager.Data{"from_state": actualLRP.State, "same_instance_key": actualLRP.ActualLRPInstanceKey.Equal(instanceKey)})
return models.ErrActualLRPCannotBeCrashed
}
actualLRP.ModificationTag.Increment()
actualLRP.State = models.ActualLRPStateCrashed
actualLRP.ActualLRPInstanceKey.InstanceGuid = ""
actualLRP.ActualLRPInstanceKey.CellId = ""
actualLRP.ActualLRPNetInfo = models.ActualLRPNetInfo{}
actualLRP.CrashCount = newCrashCount
actualLRP.CrashReason = crashReason
evacuating := false
if actualLRP.ShouldRestartImmediately(models.NewDefaultRestartCalculator()) {
actualLRP.State = models.ActualLRPStateUnclaimed
immediateRestart = true
}
now := db.clock.Now().UnixNano()
actualLRP.Since = now
_, err = db.update(logger, tx, actualLRPsTable,
SQLAttributes{
"state": actualLRP.State,
"cell_id": actualLRP.CellId,
"instance_guid": actualLRP.InstanceGuid,
"modification_tag_index": actualLRP.ModificationTag.Index,
"crash_count": actualLRP.CrashCount,
"crash_reason": truncateString(actualLRP.CrashReason, 1024),
"since": actualLRP.Since,
"net_info": []byte{},
},
"process_guid = ? AND instance_index = ? AND evacuating = ?",
key.ProcessGuid, key.Index, evacuating,
)
if err != nil {
logger.Error("failed-to-crash-actual-lrp", err)
return db.convertSQLError(err)
}
return nil
})
return &models.ActualLRPGroup{Instance: &beforeActualLRP}, &models.ActualLRPGroup{Instance: actualLRP}, immediateRestart, err
}
func (db *SQLDB) StartActualLRP(logger lager.Logger, key *models.ActualLRPKey, instanceKey *models.ActualLRPInstanceKey, netInfo *models.ActualLRPNetInfo) (*models.ActualLRPGroup, *models.ActualLRPGroup, error) {
logger = logger.WithData(lager.Data{"actual_lrp_key": key, "actual_lrp_instance_key": instanceKey, "net_info": netInfo})
var beforeActualLRP models.ActualLRP
var actualLRP *models.ActualLRP
err := db.transact(logger, func(logger lager.Logger, tx *sql.Tx) error {
var err error
actualLRP, err = db.fetchActualLRPForUpdate(logger, key.ProcessGuid, key.Index, false, tx)
if err == models.ErrResourceNotFound {
actualLRP, err = db.createRunningActualLRP(logger, key, instanceKey, netInfo, tx)
return err
}
if err != nil {
logger.Error("failed-to-get-actual-lrp", err)
return err
}
beforeActualLRP = *actualLRP
if actualLRP.ActualLRPKey.Equal(key) &&
actualLRP.ActualLRPInstanceKey.Equal(instanceKey) &&
actualLRP.ActualLRPNetInfo.Equal(netInfo) &&
actualLRP.State == models.ActualLRPStateRunning {
logger.Debug("nothing-to-change")
return nil
}
if !actualLRP.AllowsTransitionTo(key, instanceKey, models.ActualLRPStateRunning) {
logger.Error("failed-to-transition-actual-lrp-to-started", nil)
return models.ErrActualLRPCannotBeStarted
}
logger.Info("starting")
defer logger.Info("completed")
now := db.clock.Now().UnixNano()
evacuating := false
actualLRP.ActualLRPInstanceKey = *instanceKey
actualLRP.ActualLRPNetInfo = *netInfo
actualLRP.State = models.ActualLRPStateRunning
actualLRP.Since = now
actualLRP.ModificationTag.Increment()
actualLRP.PlacementError = ""
netInfoData, err := db.serializeModel(logger, &actualLRP.ActualLRPNetInfo)
if err != nil {
logger.Error("failed-to-serialize-net-info", err)
return err
}
_, err = db.update(logger, tx, actualLRPsTable,
SQLAttributes{
"state": actualLRP.State,
"cell_id": actualLRP.CellId,
"instance_guid": actualLRP.InstanceGuid,
"modification_tag_index": actualLRP.ModificationTag.Index,
"placement_error": actualLRP.PlacementError,
"since": actualLRP.Since,
"net_info": netInfoData,
},
"process_guid = ? AND instance_index = ? AND evacuating = ?",
key.ProcessGuid, key.Index, evacuating,
)
if err != nil {
logger.Error("failed-starting-actual-lrp", err)
return db.convertSQLError(err)
}
return nil
})
return &models.ActualLRPGroup{Instance: &beforeActualLRP}, &models.ActualLRPGroup{Instance: actualLRP}, err
}
ActualLRPKey: models.NewActualLRPKey(
"process-guid-1",
2,
"domain-1",
),
ActualLRPInstanceKey: models.NewActualLRPInstanceKey(
"instance-guid-1",
"cell-id-1",
),
State: models.ActualLRPStateClaimed,
Since: 4444,
}
evacuatingLRP2 = actualLRP2
evacuatingLRP2.State = models.ActualLRPStateRunning
evacuatingLRP2.Since = 3417
})
JustBeforeEach(func() {
request := newTestRequest(requestBody)
handler.ActualLRPGroups(logger, responseRecorder, request)
})
Context("when reading actual lrps from DB succeeds", func() {
var actualLRPGroups []*models.ActualLRPGroup
BeforeEach(func() {
actualLRPGroups =
[]*models.ActualLRPGroup{
{Instance: &actualLRP1},
{Instance: &actualLRP2, Evacuating: &evacuatingLRP2},
func (db *SQLDB) EvacuateActualLRP(
logger lager.Logger,
lrpKey *models.ActualLRPKey,
instanceKey *models.ActualLRPInstanceKey,
netInfo *models.ActualLRPNetInfo,
ttl uint64,
) (*models.ActualLRPGroup, error) {
logger = logger.Session("evacuate-lrp", lager.Data{"lrp_key": lrpKey, "instance_key": instanceKey, "net_info": netInfo})
logger.Debug("starting")
defer logger.Debug("complete")
var actualLRP *models.ActualLRP
err := db.transact(logger, func(logger lager.Logger, tx *sql.Tx) error {
var err error
processGuid := lrpKey.ProcessGuid
index := lrpKey.Index
actualLRP, err = db.fetchActualLRPForUpdate(logger, processGuid, index, true, tx)
if err == models.ErrResourceNotFound {
logger.Debug("creating-evacuating-lrp")
actualLRP, err = db.createEvacuatingActualLRP(logger, lrpKey, instanceKey, netInfo, ttl, tx)
return err
}
if err != nil {
logger.Error("failed-locking-lrp", err)
return err
}
if actualLRP.ActualLRPKey.Equal(lrpKey) &&
actualLRP.ActualLRPInstanceKey.Equal(instanceKey) &&
reflect.DeepEqual(actualLRP.ActualLRPNetInfo, *netInfo) {
logger.Debug("evacuating-lrp-already-exists")
return nil
}
now := db.clock.Now().UnixNano()
actualLRP.ModificationTag.Increment()
actualLRP.ActualLRPKey = *lrpKey
actualLRP.ActualLRPInstanceKey = *instanceKey
actualLRP.Since = now
actualLRP.ActualLRPNetInfo = *netInfo
netInfoData, err := db.serializeModel(logger, netInfo)
if err != nil {
logger.Error("failed-serializing-net-info", err)
return err
}
_, err = db.update(logger, tx, "actual_lrps",
SQLAttributes{
"domain": actualLRP.Domain,
"instance_guid": actualLRP.InstanceGuid,
"cell_id": actualLRP.CellId,
"net_info": netInfoData,
"state": actualLRP.State,
"since": actualLRP.Since,
"modification_tag_index": actualLRP.ModificationTag.Index,
},
"process_guid = ? AND instance_index = ? AND evacuating = ?",
actualLRP.ProcessGuid, actualLRP.Index, true,
)
if err != nil {
logger.Error("failed-update-evacuating-lrp", err)
return db.convertSQLError(err)
}
return nil
})
return &models.ActualLRPGroup{Evacuating: actualLRP}, err
}
BeforeEach(func() {
guid = "the-guid"
index = 1
ttl = 60
actualLRP = model_helpers.NewValidActualLRP(guid, index)
etcdHelper.SetRawEvacuatingActualLRP(actualLRP, ttl)
node, err := storeClient.Get(etcd.EvacuatingActualLRPSchemaPath(guid, index), false, false)
fakeStoreClient.GetReturns(node, err)
})
Context("when the something about the actual LRP has changed", func() {
BeforeEach(func() {
clock.IncrementBySeconds(5)
actualLRP.Since = clock.Now().UnixNano()
})
Context("when the lrp key changes", func() {
BeforeEach(func() {
actualLRP.Domain = "some-other-domain"
})
It("persists the evacuating lrp in etcd", func() {
group, err := etcdDB.EvacuateActualLRP(logger, &actualLRP.ActualLRPKey, &actualLRP.ActualLRPInstanceKey, &actualLRP.ActualLRPNetInfo, ttl)
Expect(err).NotTo(HaveOccurred())
actualLRP.ModificationTag.Increment()
actualLRPGroup, err := etcdDB.ActualLRPGroupByProcessGuidAndIndex(logger, guid, index)
Expect(err).NotTo(HaveOccurred())
Expect(actualLRPGroup.Evacuating).To(BeEquivalentTo(actualLRP))
