Moved 'failure-detector' config from 'akka.actor.deployment.address' to 'akka.remote'. Made AccrualFailureDetector configurable from config.
This commit is contained in:
Jonas Bonér
parent
39d169612f
commit
f4740a4798
9 changed files with 27 additions and 106 deletions
|
|
@ -22,7 +22,6 @@ class DeployerSpec extends AkkaSpec {
|
|||
None,
|
||||
RoundRobin,
|
||||
NrOfInstances(3),
|
||||
BannagePeriodFailureDetector(10 seconds),
|
||||
RemoteScope(List(
|
||||
RemoteAddress("wallace", 2552), RemoteAddress("gromit", 2552))))))
|
||||
// ClusterScope(
|
||||
|
|
|
|||
|
|
@ -22,7 +22,6 @@ class ConfiguredLocalRoutingSpec extends AkkaSpec {
|
|||
None,
|
||||
RoundRobin,
|
||||
NrOfInstances(5),
|
||||
NoOpFailureDetector,
|
||||
LocalScope))
|
||||
|
||||
val helloLatch = new CountDownLatch(5)
|
||||
|
|
@ -58,7 +57,6 @@ class ConfiguredLocalRoutingSpec extends AkkaSpec {
|
|||
None,
|
||||
RoundRobin,
|
||||
NrOfInstances(10),
|
||||
NoOpFailureDetector,
|
||||
LocalScope))
|
||||
|
||||
val connectionCount = 10
|
||||
|
|
@ -103,7 +101,6 @@ class ConfiguredLocalRoutingSpec extends AkkaSpec {
|
|||
None,
|
||||
RoundRobin,
|
||||
NrOfInstances(5),
|
||||
NoOpFailureDetector,
|
||||
LocalScope))
|
||||
|
||||
val helloLatch = new CountDownLatch(5)
|
||||
|
|
@ -138,7 +135,6 @@ class ConfiguredLocalRoutingSpec extends AkkaSpec {
|
|||
None,
|
||||
Random,
|
||||
NrOfInstances(7),
|
||||
NoOpFailureDetector,
|
||||
LocalScope))
|
||||
|
||||
val stopLatch = new CountDownLatch(7)
|
||||
|
|
@ -172,7 +168,6 @@ class ConfiguredLocalRoutingSpec extends AkkaSpec {
|
|||
None,
|
||||
Random,
|
||||
NrOfInstances(10),
|
||||
NoOpFailureDetector,
|
||||
LocalScope))
|
||||
|
||||
val connectionCount = 10
|
||||
|
|
@ -217,7 +212,6 @@ class ConfiguredLocalRoutingSpec extends AkkaSpec {
|
|||
None,
|
||||
Random,
|
||||
NrOfInstances(6),
|
||||
NoOpFailureDetector,
|
||||
LocalScope))
|
||||
|
||||
val helloLatch = new CountDownLatch(6)
|
||||
|
|
|
|||
|
|
@ -162,11 +162,11 @@ class LocalActorRefProvider(val app: AkkaApplication) extends ActorRefProvider {
|
|||
(if (systemService) None else deployer.lookupDeployment(address)) match { // see if the deployment already exists, if so use it, if not create actor
|
||||
|
||||
// create a local actor
|
||||
case None | Some(DeploymentConfig.Deploy(_, _, DeploymentConfig.Direct, _, _, DeploymentConfig.LocalScope)) ⇒
|
||||
case None | Some(DeploymentConfig.Deploy(_, _, DeploymentConfig.Direct, _, DeploymentConfig.LocalScope)) ⇒
|
||||
new LocalActorRef(app, props, supervisor, address, systemService) // create a local actor
|
||||
|
||||
// create a routed actor ref
|
||||
case deploy @ Some(DeploymentConfig.Deploy(_, _, routerType, nrOfInstances, _, DeploymentConfig.LocalScope)) ⇒
|
||||
case deploy @ Some(DeploymentConfig.Deploy(_, _, routerType, nrOfInstances, DeploymentConfig.LocalScope)) ⇒
|
||||
|
||||
val routerFactory: () ⇒ Router = DeploymentConfig.routerTypeFor(routerType) match {
|
||||
case RouterType.Direct ⇒ () ⇒ new DirectRouter
|
||||
|
|
|
|||
|
|
@ -51,7 +51,7 @@ class Deployer(val app: AkkaApplication) extends ActorDeployer {
|
|||
def deploy(deployment: Deploy): Unit = instance.deploy(deployment)
|
||||
|
||||
def isLocal(deployment: Deploy): Boolean = deployment match {
|
||||
case Deploy(_, _, _, _, _, LocalScope) | Deploy(_, _, _, _, _, _: LocalScope) ⇒ true
|
||||
case Deploy(_, _, _, _, LocalScope) | Deploy(_, _, _, _, _: LocalScope) ⇒ true
|
||||
case _ ⇒ false
|
||||
}
|
||||
|
||||
|
|
@ -144,50 +144,6 @@ class Deployer(val app: AkkaApplication) extends ActorDeployer {
|
|||
}
|
||||
}
|
||||
|
||||
// --------------------------------
|
||||
// akka.actor.deployment.<address>.failure-detector.<detector>
|
||||
// --------------------------------
|
||||
val failureDetectorOption: Option[FailureDetector] = addressConfig.getSection("failure-detector") match {
|
||||
case Some(failureDetectorConfig) ⇒
|
||||
failureDetectorConfig.keys.toList match {
|
||||
case Nil ⇒ None
|
||||
case detector :: Nil ⇒
|
||||
detector match {
|
||||
case "no-op" ⇒
|
||||
Some(NoOpFailureDetector)
|
||||
|
||||
case "remove-connection-on-first-failure" ⇒
|
||||
Some(RemoveConnectionOnFirstFailureFailureDetector)
|
||||
|
||||
case "bannage-period" ⇒
|
||||
throw new ConfigurationException(
|
||||
"Configuration for [" + addressPath + ".failure-detector.bannage-period] must have a 'time-to-ban' option defined")
|
||||
|
||||
case "bannage-period.time-to-ban" ⇒
|
||||
failureDetectorConfig.getSection("bannage-period") map { section ⇒
|
||||
val timeToBan = Duration(section.getInt("time-to-ban", 60), app.AkkaConfig.DefaultTimeUnit)
|
||||
BannagePeriodFailureDetector(timeToBan)
|
||||
}
|
||||
|
||||
case "custom" ⇒
|
||||
failureDetectorConfig.getSection("custom") map { section ⇒
|
||||
val implementationClass = section.getString("class").getOrElse(throw new ConfigurationException(
|
||||
"Configuration for [" + addressPath +
|
||||
".failure-detector.custom] must have a 'class' element with the fully qualified name of the failure detector class"))
|
||||
CustomFailureDetector(implementationClass)
|
||||
}
|
||||
|
||||
case _ ⇒ None
|
||||
}
|
||||
case detectors ⇒
|
||||
throw new ConfigurationException(
|
||||
"Configuration for [" + addressPath +
|
||||
".failure-detector] can not have multiple sections - found [" + detectors.mkString(", ") + "]")
|
||||
}
|
||||
case None ⇒ None
|
||||
}
|
||||
val failureDetector = failureDetectorOption getOrElse { NoOpFailureDetector } // fall back to default failure detector
|
||||
|
||||
// --------------------------------
|
||||
// akka.actor.deployment.<address>.create-as
|
||||
// --------------------------------
|
||||
|
|
@ -235,7 +191,7 @@ class Deployer(val app: AkkaApplication) extends ActorDeployer {
|
|||
}
|
||||
}
|
||||
|
||||
Some(Deploy(address, recipe, router, nrOfInstances, failureDetector, RemoteScope(remoteAddresses)))
|
||||
Some(Deploy(address, recipe, router, nrOfInstances, RemoteScope(remoteAddresses)))
|
||||
|
||||
case None ⇒ // check for 'cluster' config section
|
||||
|
||||
|
|
@ -266,9 +222,7 @@ class Deployer(val app: AkkaApplication) extends ActorDeployer {
|
|||
val address = tokenizer.nextElement.asInstanceOf[String]
|
||||
|
||||
protocol match {
|
||||
//case "host" ⇒ Host(address)
|
||||
case "node" ⇒ Node(address)
|
||||
//case "ip" ⇒ IP(address)
|
||||
case _ ⇒ raiseHomeConfigError()
|
||||
}
|
||||
}
|
||||
|
|
@ -279,7 +233,7 @@ class Deployer(val app: AkkaApplication) extends ActorDeployer {
|
|||
// --------------------------------
|
||||
clusterConfig.getSection("replication") match {
|
||||
case None ⇒
|
||||
Some(Deploy(address, recipe, router, nrOfInstances, failureDetector, deploymentConfig.ClusterScope(preferredNodes, Transient)))
|
||||
Some(Deploy(address, recipe, router, nrOfInstances, deploymentConfig.ClusterScope(preferredNodes, Transient)))
|
||||
|
||||
case Some(replicationConfig) ⇒
|
||||
val storage = replicationConfig.getString("storage", "transaction-log") match {
|
||||
|
|
@ -298,7 +252,7 @@ class Deployer(val app: AkkaApplication) extends ActorDeployer {
|
|||
".cluster.replication.strategy] needs to be either [\"write-through\"] or [\"write-behind\"] - was [" +
|
||||
unknown + "]")
|
||||
}
|
||||
Some(Deploy(address, recipe, router, nrOfInstances, failureDetector, deploymentConfig.ClusterScope(preferredNodes, Replication(storage, strategy))))
|
||||
Some(Deploy(address, recipe, router, nrOfInstances, deploymentConfig.ClusterScope(preferredNodes, Replication(storage, strategy))))
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -19,7 +19,6 @@ object DeploymentConfig {
|
|||
recipe: Option[ActorRecipe],
|
||||
routing: Routing = Direct,
|
||||
nrOfInstances: NrOfInstances = ZeroNrOfInstances,
|
||||
failureDetector: FailureDetector = NoOpFailureDetector,
|
||||
scope: Scope = LocalScope)
|
||||
|
||||
// --------------------------------
|
||||
|
|
@ -51,21 +50,6 @@ object DeploymentConfig {
|
|||
case object LeastRAM extends Routing
|
||||
case object LeastMessages extends Routing
|
||||
|
||||
// --------------------------------
|
||||
// --- FailureDetector
|
||||
// --------------------------------
|
||||
sealed trait FailureDetector
|
||||
case class BannagePeriodFailureDetector(timeToBan: Duration) extends FailureDetector
|
||||
case class CustomFailureDetector(className: String) extends FailureDetector
|
||||
|
||||
// For Java API
|
||||
case class NoOpFailureDetector() extends FailureDetector
|
||||
case class RemoveConnectionOnFirstFailureFailureDetector() extends FailureDetector
|
||||
|
||||
// For Scala API
|
||||
case object NoOpFailureDetector extends FailureDetector
|
||||
case object RemoveConnectionOnFirstFailureFailureDetector extends FailureDetector
|
||||
|
||||
// --------------------------------
|
||||
// --- Scope
|
||||
// --------------------------------
|
||||
|
|
@ -176,16 +160,6 @@ object DeploymentConfig {
|
|||
// case IP(address) ⇒ throw new UnsupportedOperationException("Specifying preferred node name by 'IP address' is not yet supported. Use the node name like: preferred-nodes = [\"node:node1\"]")
|
||||
}
|
||||
|
||||
def failureDetectorTypeFor(failureDetector: FailureDetector): FailureDetectorType = failureDetector match {
|
||||
case NoOpFailureDetector ⇒ FailureDetectorType.NoOp
|
||||
case NoOpFailureDetector() ⇒ FailureDetectorType.NoOp
|
||||
case BannagePeriodFailureDetector(timeToBan) ⇒ FailureDetectorType.BannagePeriod(timeToBan)
|
||||
case RemoveConnectionOnFirstFailureFailureDetector ⇒ FailureDetectorType.RemoveConnectionOnFirstFailure
|
||||
case RemoveConnectionOnFirstFailureFailureDetector() ⇒ FailureDetectorType.RemoveConnectionOnFirstFailure
|
||||
case CustomFailureDetector(implClass) ⇒ FailureDetectorType.Custom(implClass)
|
||||
case unknown ⇒ throw new UnsupportedOperationException("Unknown FailureDetector [" + unknown + "]")
|
||||
}
|
||||
|
||||
def routerTypeFor(routing: Routing): RouterType = routing match {
|
||||
case _: Direct | Direct ⇒ RouterType.Direct
|
||||
case _: RoundRobin | RoundRobin ⇒ RouterType.RoundRobin
|
||||
|
|
@ -255,7 +229,7 @@ class DeploymentConfig(val app: AkkaApplication) {
|
|||
def isHomeNode(homes: Iterable[Home]): Boolean = homes exists (home ⇒ nodeNameFor(home) == app.nodename)
|
||||
|
||||
def replicationSchemeFor(deployment: Deploy): Option[ReplicationScheme] = deployment match {
|
||||
case Deploy(_, _, _, _, _, ClusterScope(_, replicationScheme)) ⇒ Some(replicationScheme)
|
||||
case Deploy(_, _, _, _, ClusterScope(_, replicationScheme)) ⇒ Some(replicationScheme)
|
||||
case _ ⇒ None
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -12,6 +12,8 @@ import scala.annotation.tailrec
|
|||
|
||||
import System.{ currentTimeMillis ⇒ newTimestamp }
|
||||
|
||||
import akka.AkkaApplication
|
||||
|
||||
/**
|
||||
* Implementation of 'The Phi Accrual Failure Detector' by Hayashibara et al. as defined in their paper:
|
||||
* [http://ddg.jaist.ac.jp/pub/HDY+04.pdf]
|
||||
|
|
@ -22,11 +24,15 @@ import System.{ currentTimeMillis ⇒ newTimestamp }
|
|||
* <p/>
|
||||
* Default threshold is 8, but can be configured in the Akka config.
|
||||
*/
|
||||
class AccrualFailureDetector(
|
||||
val threshold: Int = 8,
|
||||
val maxSampleSize: Int = 1000) {
|
||||
class AccrualFailureDetector(val threshold: Int = 8, val maxSampleSize: Int = 1000) {
|
||||
|
||||
final val PhiFactor = 1.0 / math.log(10.0)
|
||||
def this(app: AkkaApplication) {
|
||||
this(
|
||||
app.config.getInt("akka.remote.failure-detector.theshold", 8),
|
||||
app.config.getInt("akka.remote.failure-detector.max-sample-size", 1000))
|
||||
}
|
||||
|
||||
private final val PhiFactor = 1.0 / math.log(10.0)
|
||||
|
||||
private case class FailureStats(mean: Double = 0.0D, variance: Double = 0.0D, deviation: Double = 0.0D)
|
||||
|
||||
|
|
|
|||
|
|
@ -37,7 +37,7 @@ class Remote(val app: AkkaApplication) {
|
|||
val shouldCompressData = config.getBool("akka.remote.use-compression", false)
|
||||
val remoteSystemDaemonAckTimeout = Duration(config.getInt("akka.remote.remote-daemon-ack-timeout", 30), DefaultTimeUnit).toMillis.toInt
|
||||
|
||||
val failureDetector = new AccrualFailureDetector(FailureDetectorThreshold, FailureDetectorMaxSampleSize)
|
||||
val failureDetector = new AccrualFailureDetector(app)
|
||||
|
||||
// val gossiper = new Gossiper(this)
|
||||
|
||||
|
|
|
|||
|
|
@ -58,7 +58,7 @@ class RemoteActorRefProvider(val app: AkkaApplication) extends ActorRefProvider
|
|||
case null ⇒
|
||||
val actor: ActorRef = try {
|
||||
deployer.lookupDeploymentFor(address) match {
|
||||
case Some(DeploymentConfig.Deploy(_, _, routerType, nrOfInstances, failureDetectorType, DeploymentConfig.RemoteScope(remoteAddresses))) ⇒
|
||||
case Some(DeploymentConfig.Deploy(_, _, routerType, nrOfInstances, DeploymentConfig.RemoteScope(remoteAddresses))) ⇒
|
||||
|
||||
// FIXME move to AccrualFailureDetector as soon as we have the Gossiper up and running and remove the option to select impl in the akka.conf file since we only have one
|
||||
// val failureDetector = DeploymentConfig.failureDetectorTypeFor(failureDetectorType) match {
|
||||
|
|
|
|||
|
|
@ -75,20 +75,6 @@ akka {
|
|||
# default is '1'
|
||||
# if the "direct" router is used then this element is ignored (always '1')
|
||||
|
||||
failure-detector { # failure detection scheme to use
|
||||
bannage-period { # available: no-op {}
|
||||
time-to-ban = 10 # remove-connection-on-first-failure {}
|
||||
} # bannage-period { ... }
|
||||
|
||||
# Options:
|
||||
# remove-connection-on-first-failure {} # threshold { ... }
|
||||
# custom { # custom { ... } - fully qualified class name of the router class
|
||||
# class = "com.biz.app.MyCustomFailureDetector"
|
||||
# }
|
||||
# threshold {
|
||||
# }
|
||||
}
|
||||
|
||||
#create-as {
|
||||
# class = "com.biz.app.MyActor" # FIXME document 'create-as'
|
||||
#}
|
||||
|
|
@ -236,6 +222,14 @@ akka {
|
|||
|
||||
remote-daemon-ack-timeout = 30 # Timeout for ACK of cluster operations, lik checking actor out etc.
|
||||
|
||||
failure-detector { # accrual failure detection config
|
||||
threshold = 8 # defines the failure detector threshold
|
||||
# A low threshold is prone to generate many wrong suspicions but ensures a
|
||||
# quick detection in the event of a real crash. Conversely, a high threshold
|
||||
# generates fewer mistakes but needs more time to detect actual crashes
|
||||
max-sample-size = 1000
|
||||
}
|
||||
|
||||
server {
|
||||
port = 2552 # The default remote server port clients should connect to. Default is 2552 (AKKA)
|
||||
message-frame-size = 1048576 # Increase this if you want to be able to send messages with large payloads
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue