In the initial version, only an Actorsystem (name Akka-Network-Ping) is created.
No Actors are created yet, but there's a command line parser that accepts the following commands from the command-line:
[pingerCount] pi|ping [pingCount] [pingInterval]
s|status
q|quit
The ping (alias pi) will, in a subsequent exercise, create pingerCount Pinger actor(s) that will ping a single PingServer actor. The PingServer actor should respond to a Ping message with a Response message to the actor who sent the Ping message.
In this state of the application, messages are logged at INFO level.
Note that the logging output is directed to a log file named pingpong.log in the project's root folder.
The overall Actor hierarchy will looks as follows:
/user
|
/PingResponseCoordinator
|
+-----------------+----------------+
| |
/pingMaster /pingServer
|
+---------+---------+---------+
| | | |
/pinger1 /pinger2 ... /pingerN
In the first exercise, you will set-up actors PingResponseCoordinator, PingMaster, Pinger and PingServer.
In this exercise, you will define four actors, PingResponseCoordinator, PingMaster, Pinger and PingServer, each with an empty behaviour and a definition of the messages, Ping and Response, relevant to the protocol we want to implement.
Your tasks:
- Change the behaviour of the PingServer actor so that when it receives a
Pingmessage, it responds to the sender of the message with aResponsemessage. Note that both messages carry extra information:
- sequenceNumber: a number that is unique for each
Pingmessage for a particularPingactor.
-
In the
Responseresponse, the value ofrefcontained in thePingmessage should be copied as-is. -
Change the behaviour of the
Pingeractor so that, after creation, it will sendpingCountPingmessages to thePingServeractor. ThesequenceNumbershould be incremented between subsequentPingmessages. For the time being, ignorepingInterval: just send the messages in rapid succession. -
Change the
PingResponseCoordinatoractor. When an instance of this actor is created, it should create aPingServerand aPingMasteractor as children. Name these actorspingServerandpingMasterrespectively. -
When
PingResponseCoordinatorreceives aCreatePingermessage, it should forward this message to thePingMasteractor. -
Create the
PingResponseCoordinatoractor inPingResponseApp(look for the TODO in the code). -
Change the
PingMasterbehaviour as to create aPingerchild actor when it receives aCreatePingermessage. -
Adapt the
createPingermethod in thePingResponseApp(look for the TODO in the code) as to send the appropriate messages to the right actor so that the requested number ofPingeractors are created. -
Run the application, run some ping commands and verify that the output is what is expected.
This code implements a solution for the stated tasks. However, some questions may be raised:
-
How to implement the
pingIntervalbehaviour - whenpingCount> 1: the firstPingshould be sent immediately and subsequentPingmessages should be sent sequentially atpingIntervalintervals. -
In the current application, there's an obvious memory leak:
Pingactors are created and, after having performed their task, are never stopped... -
A number of tests have been added to the project that test the functionality of the
PingerandPingServeractors.
##Exercise 2 - Implementing pingInterval
-
Change
Pingerto send thePingmessages at the specified interval. Have a look at an actor's access to the scheduler (context.system.scheduler.*). Note that we always sendpingCountmessages, i.e. regardless of the number ofResponsemessages that are sent back in response. -
Fix the memory leak, for example, let the
Pingactor stop itself when it has done its job. For this, look at what's available under the actor's context.
##Exercise 2 - Solution & discussion
The sending of Ping messages at regular intervals can be implemented in many ways... In the given solution, the Pinger actor uses the scheduler to schedule delivery of the required Ping messages to itself at the right moment.
-
Also note the utilisation of
context.stop(self)to stop thePingeractor. -
Notice the need to bring an (implicit) execution scope into context - this is needed by the
scheduleOncemethod
##Exercise 3 - Introduce reaction time in PingServer actor
In this exercise, we will let PingServer actor respond to Ping messages with a delay configurable in the application config settings. Change the signature of the PingServer actor as to accept this delay parameter of type FiniteDuration.
Name the configuration parameter AkkaNetworkPing.Response.responseDelay
Perform some ping tests with the responseDelay set to 2 seconds. What is the impact on the behaviour of the application? What is causing this effect?
##Exercise 3 - Solution & discussion
In the proposed solution, we make sure that the PingServer actor is really busy while responding to a Ping. We do this by ignoring any incoming message during during the processing of the Ping. In order to achieve this, we utilise the Akka Stash trait.
##Exercise 3 - Solution - Alternative Pinger implementation
An alternative for the utilisation of mutable state in the Pinger actor.
##Exercise 4 - Scale the PingServer actor
As observed during some testing, the single instance of the PingServer actor has become a limiting factor for the ping-pong throughput.
Use a pooled router with round-robin routing strategy configured via Akka configuration to scale the app.
##Exercise 4 - Solution & discussion
Making the Akka-Network-Ping scale can be achieved by removing the bottleneck in the PingServer actor by turning it into a pooled router. Notice the elegance by which this can be done: a simply change in the deployment configuration and a change on one line in the source code.
##Exercise 5 - Introduction of an unreliable PingServer actor
Let's make PingServer a bit unreliable by having it not send a reply to a Ping message at random:
- Define a configuration parameter named
AkkaNetworkPing.Response.reliabilitywith a value between 0 and 100. - Change the signature of
PingServeras to accept thereliabilityparameter of typeInt. - Change the behaviour of
PingServerso that when it receives aPing, it throws a dice by generating a random number between 0 and 100. If this number is greater thanreliability, respond. Otherwise, ignore the message and continue waiting for a newPing.
##Exercise 5 - Solution & discussion
After the modification, we see that, in case of dropped Ping messages, the bookkeeping of the Pinger no longer fits. As a consequence, it never stop because it will keep waiting for responses that will never arrive.
Observe this by running the application a number of times, setting reliability to for example 60 until at least one message is dropped.
##Exercise 6 - Delegating work to one-off actors
In this exercise you will change the behaviour of the Pinger actor in a number of ways:
-
Instead of
Pingersending aPingmessage itself, it will delegate this task to a one-off actor namedPingerWorker -
PingerWorkerwill send aPingmessage it receives fromPingertoPingServer. -
PingerWorkerwill wait for a reply fromPingServerfor a specified amount of time:pingTimeoutof typeFiniteDuration. -
If a reply is received in time, it is sent to
Pinger. Otherwise,PingerWorkerthrows an exception.
Implement the task in the following steps:
-
A
PingWorkeractor class has been set-up already.- It defines an exception definition (
TimeoutException). - It defines a case class
Timedoutto be used in the protocol. - Use the scheduler to send a message to
selfwhen the timeout period is completed. - Modify the behaviour as specified in the introduction.
- It defines an exception definition (
-
Modify
Pingeras follows:- Let it create the appropriate number of
PingerWorkeractors and foreach of them, send thePing(already scheduled in the code) to these actors. - Check the behaviour for other required modifications.
- Let it create the appropriate number of
Run the application and verify if it works as expected. Set the configuration parameters in such a way that a timeout is triggered and observe what happens. Is the observed behaviour the desired one ?
##Exercise 6 - Solution & discussion
When PingServer's reliability is set to a low level (e.g. 50), time-outs can be triggered easily. What we see in that case is that the default Akka SupervisoryStragegy is to restart a failed actor. This is clearly not what is needed: the PingerWorker actor should be stopped instead.
Let's correct this in the next exercise.
##Exercise 7 - Tuning an actor's supervision strategy
In this exercise you will define the appropriate supervisory strategy for the PingerWorker actor. Also, we want to correct the bookkeeping in the Pinger Actor - the actor who's supervising the PingerWorker.
A new message PingTimedout message has been created to facilitate things: it can be sent to self to assist the bookkeeping process.
Your tasks:
- Implement the correct supervisory strategy for
PingerWorker - Correct the bookkeeping so that the
Pingeractor is stopped after having received all replies (or timeout indications).
##Exercise 7 - Solution & discussion
Note that the decider in actor Pinger captures the TimeoutException and specifies that the faulty actor should be stopped. Also, for bookkeeping purposes, it sends a PingTimedout message to itself which in turn leads to an update of the internal state of the actor.
The ask pattern can be used to asynchronously get a response to a message sent to an actor.
We will use this to get the number of Pinger actors at a particular moment in time.
You will get this information from the PingMain actor by sending it a GetStatus (case object) defined in PingMain. This actor should respond with a Status message holding the requested count. Hence, this should be reflected in the behaviour of the PingMain actor.
You will need to obtain the ActorRef for actor PingMain. Have a look at ActorSelection in ActorSystem.
Also note that ask needs the following in order to function:
- an (implicit)
Timeoutvalue. - an (implicit) Execution context.
An ask will return a Future giving you a handle to a (future) result. Note that due to type erasure, you will need to apply mapTo to indicate the correct return type. Next, you'll finish things off by installing callbacks on it to handle successful completion or failure.
##Exercise 8 - Solution & discussion
Congrats - you made it !!!
##What's next ?
You've covered a lot of important features of Akka while solving the exercises in this hacking session. Do note however that there's a lot more to cover such as:
- Akka FSM & FSMLogging
askpattern in combination withpipeTo- Pooled router versus Group router & different routing strategies
- Supervisor strategies -
OneForOneStrategyversusOneForAllStrategy - Akka persistence
- Local versus Remote deployment
- Akka clustering
- Akka sharding
- ....
##Suggestions for further exploration
You might extend AkkaNetworkPing by:
- Implementing retrying the sending of
Pingmessages in case no (Response) is replied within a certain time. - Keep track of the number of ping requests, actual pings sent, number of timeouts.
- Report of the stats by extending the stats command
- Implement
Pinger,PingerWorkerandPingServerusing Akka FSM - Adding more tests...
- Experiment with
ask/pipeTo - ...
Some excellent Akka books to read: