Answer
The cluster node maintains a ready listing which maps a key and
a callback perform. The bottom line is chosen relying on the precise
standards to invoke the callback.
For instance if it must be invoked every time
a message from different cluster node is obtained, it may be the
Correlation Identifer of the message.
Within the case of Replicated Log it’s the
Excessive-Water Mark. The callback handles
the response and decides if the consumer request may be fulfilled.
Think about the instance of a key-value retailer the place, knowledge is replicated
on a number of servers. Right here, Quorum can be utilized
to resolve when a replication may be thought of profitable
to provoke a response to the consumer.
The cluster node then tracks the requests despatched to different cluster nodes,
and a callback is registered with every request.
Every request is marked with a Correlation Identifer,
which is used to map response to the request.
The ready listing is then notified to invoke the callback
when the response from different cluster nodes are obtained.
For the sake of this instance, let’s name our three cluster nodes
athens, byzantium and cyrene.
The consumer connects with athens to retailer “title” as “Microservices”.
Athens replicates it on byzantium and cyrene; so it sends
a request to itself to retailer the key-value and sends
requests to each byzantium and cyrene concurrently.
To trace responses, athens creates a WriteQuorumResponseCallback
and provides it to the ready listing for every of the requests despatched.
For each response obtained, the WriteQuorumResponseCallback is
invoked to deal with the response. It checks whether or not the required quantity
of responses have been obtained.
As soon as the response is obtained from byzantium, the quorum is reached
and the pending consumer request is accomplished.
Cyrene can reply later, however the response may be despatched to the consumer
with out ready for it.
The code appears just like the pattern beneath:
Observe that each cluster node maintains its personal occasion of a ready listing.
The ready listing tracks the important thing and related callback and
shops the timestamp at which the callback was registered.
The timestamp is used to test whether or not the callbacks should be expired
if responses have not been obtained throughout the anticipated time.
public class RequestWaitingList<Key, Response> { personal Map<Key, CallbackDetails> pendingRequests = new ConcurrentHashMap<>(); public void add(Key key, RequestCallback<Response> callback) { pendingRequests.put(key, new CallbackDetails(callback, clock.nanoTime())); }
class CallbackDetails { RequestCallback requestCallback; lengthy createTime; public CallbackDetails(RequestCallback requestCallback, lengthy createTime) { this.requestCallback = requestCallback; this.createTime = createTime; } public RequestCallback getRequestCallback() { return requestCallback; } public lengthy elapsedTime(lengthy now) { return now - createTime; } }
public interface RequestCallback<T> { void onResponse(T r); void onError(Throwable e); }
It’s requested to deal with the response or error
as soon as the response has been obtained from the opposite cluster node.
class RequestWaitingList…
public void handleResponse(Key key, Response response) { if (!pendingRequests.containsKey(key)) { return; } CallbackDetails callbackDetails = pendingRequests.take away(key); callbackDetails.getRequestCallback().onResponse(response); }
class RequestWaitingList…
public void handleError(int requestId, Throwable e) { CallbackDetails callbackDetails = pendingRequests.take away(requestId); callbackDetails.getRequestCallback().onError(e); }
The ready listing can then be used to deal with quorum responses
with the implementation wanting one thing like this:
static class WriteQuorumCallback implements RequestCallback<RequestOrResponse> { personal last int quorum; personal unstable int expectedNumberOfResponses; personal unstable int receivedResponses; personal unstable int receivedErrors; personal unstable boolean performed; personal last RequestOrResponse request; personal last ClientConnection clientConnection; public WriteQuorumCallback(int totalExpectedResponses, RequestOrResponse clientRequest, ClientConnection clientConnection) { this.expectedNumberOfResponses = totalExpectedResponses; this.quorum = expectedNumberOfResponses / 2 + 1; this.request = clientRequest; this.clientConnection = clientConnection; } @Override public void onResponse(RequestOrResponse response) { receivedResponses++; if (receivedResponses == quorum && !performed) { respondToClient("Success"); performed = true; } } @Override public void onError(Throwable t) { receivedErrors++; if (receivedErrors == quorum && !performed) { respondToClient("Error"); performed = true; } } personal void respondToClient(String response) { clientConnection.write(new RequestOrResponse(RequestId.SetValueResponse.getId(), response.getBytes(), request.getCorrelationId())); } }
Each time a cluster node sends requests to different nodes,
it provides a callback to the ready listing mapping with the Correlation Identifer
of the request despatched.
class ClusterNode…
personal void handleSetValueClientRequestRequiringQuorum(Listing<InetAddressAndPort> replicas, RequestOrResponse request, ClientConnection clientConnection) { int totalExpectedResponses = replicas.dimension(); RequestCallback requestCallback = new WriteQuorumCallback(totalExpectedResponses, request, clientConnection); for (InetAddressAndPort duplicate : replicas) { int correlationId = nextRequestId(); requestWaitingList.add(correlationId, requestCallback); strive { SocketClient consumer = new SocketClient(duplicate); consumer.sendOneway(new RequestOrResponse(RequestId.SetValueRequest.getId(), request.getMessageBodyJson(), correlationId, listenAddress)); } catch (IOException e) { requestWaitingList.handleError(correlationId, e); } } }
As soon as the response is obtained, the ready listing is requested to deal with it:
class ClusterNode…
personal void handleSetValueResponse(RequestOrResponse response) { requestWaitingList.handleResponse(response.getCorrelationId(), response); }
The ready listing will then invoke the related WriteQuorumCallback.
The WriteQuorumCallback occasion verifies if
the quorum responses have been obtained and invokes the callback
to reply to the consumer.
Expiring Lengthy Pending Requests
Generally, responses from the opposite cluster nodes are
delayed. In these situations the ready listing typically has
a mechanism to run out requests after a timeout:
class RequestWaitingList…
personal SystemClock clock; personal ScheduledExecutorService executor = Executors.newSingleThreadScheduledExecutor(); personal lengthy expirationIntervalMillis = 2000; public RequestWaitingList(SystemClock clock) { this.clock = clock; executor.scheduleWithFixedDelay(this::expire, expirationIntervalMillis, expirationIntervalMillis, MILLISECONDS); } personal void expire() { lengthy now = clock.nanoTime(); Listing<Key> expiredRequestKeys = getExpiredRequestKeys(now); expiredRequestKeys.stream().forEach(expiredRequestKey -> { CallbackDetails request = pendingRequests.take away(expiredRequestKey); request.requestCallback.onError(new TimeoutException("Request expired")); }); } personal Listing<Key> getExpiredRequestKeys(lengthy now) { return pendingRequests.entrySet().stream().filter(entry -> entry.getValue().elapsedTime(now) > expirationIntervalMillis).map(e -> e.getKey()).gather(Collectors.toList()); }