Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance. Project price only 1 $
You can buy this project and download/modify it how often you want.
package io.mats3.matsbrokermonitor.activemq;
import java.time.Clock;
import java.time.Instant;
import java.util.ArrayDeque;
import java.util.Deque;
import java.util.Iterator;
import java.util.Optional;
import java.util.OptionalDouble;
import java.util.concurrent.ConcurrentNavigableMap;
import java.util.concurrent.ConcurrentSkipListMap;
import java.util.concurrent.CopyOnWriteArrayList;
import java.util.concurrent.ThreadLocalRandom;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.function.Consumer;
import javax.jms.Connection;
import javax.jms.ConnectionFactory;
import javax.jms.DeliveryMode;
import javax.jms.JMSException;
import javax.jms.MapMessage;
import javax.jms.Message;
import javax.jms.MessageConsumer;
import javax.jms.MessageProducer;
import javax.jms.Queue;
import javax.jms.Session;
import javax.jms.Topic;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.slf4j.MDC;
/**
* @author Endre Stølsvik 2021-12-20 18:00 - http://stolsvik.com/, [email protected]
*/
public class ActiveMqBrokerStatsQuerierImpl implements ActiveMqBrokerStatsQuerier, Statics {
private static final Logger log = LoggerFactory.getLogger(ActiveMqBrokerStatsQuerierImpl.class);
private static final Logger log_update = LoggerFactory.getLogger(ActiveMqBrokerStatsQuerierImpl.class.getName()
+ ".Update");
private static final String CORRELATION_ID_PREFIX_SCHEDULED = "Scheduled";
private static final String CORRELATION_ID_PREFIX_FORCED = "Forced";
private static final String CORRELATION_ID_PARAMETER_FULL_UPDATE = "fullUpdate";
private static final String CORRELATION_ID_PARAMETER_NODE_ID = "nodeId";
private String _nodeId = Long.toString(Math.abs(ThreadLocalRandom.current().nextLong()), 36)
+ Long.toString(Math.abs(ThreadLocalRandom.current().nextLong()), 36);
private final ConnectionFactory _connectionFactory;
private final Clock _clock;
private final long _updateIntervalMillis;
/**
* 2.5 minutes query interval.
*/
static ActiveMqBrokerStatsQuerierImpl create(ConnectionFactory connectionFactory) {
return create(connectionFactory, DEFAULT_UPDATE_INTERVAL_MILLIS);
}
/**
* Using specified query interval, should not be too small, in particular if lots of queues and topics, as each
* queue and topic gets its own statistics reply message: Less than 30 seconds would be much too often.
*/
static ActiveMqBrokerStatsQuerierImpl create(ConnectionFactory connectionFactory, long updateIntervalMillis) {
return new ActiveMqBrokerStatsQuerierImpl(connectionFactory, Clock.systemUTC(), updateIntervalMillis);
}
private ActiveMqBrokerStatsQuerierImpl(ConnectionFactory connectionFactory, Clock clock,
long updateIntervalMillis) {
_connectionFactory = connectionFactory;
_clock = clock;
_updateIntervalMillis = updateIntervalMillis;
}
private enum RunStatus {
NOT_STARTED,
RUNNING,
CLOSED
}
private String _matsDestinationPrefix = "mats.";
private final Deque _waitObject_And_ForceUpdateOutstandingCorrelationIds = new ArrayDeque<>();
// RunStatus: NOT_STARTED -> RUNNING -> CLOSED. Not restartable.
private volatile RunStatus _runStatus = RunStatus.NOT_STARTED;
private volatile Thread _sendStatsRequestMessages_Thread;
private volatile Thread _receiveDestinationsStatsReplyMessages_Thread;
private volatile Connection _receiveDestinationsStatsReplyMessages_Connection;
private volatile long _lastFullUpdatePropagatedMillis;
private volatile long _lastStatsUpdateMessageReceived = System.currentTimeMillis();
// Reset in request thread, used in receive thread.
private final AtomicInteger _countOfDestinationsReceivedAfterRequest = new AtomicInteger();
// Only used from receive thread.
private int _countOfDestinationsReceivedSinceLastBatch;
private volatile long _nanosAtStart_RequestQuery = 0;
private volatile BrokerStatsDto _currentBrokerStatsDto;
private final ConcurrentNavigableMap _currentDestinationStatsDtos = new ConcurrentSkipListMap<>();
private final CopyOnWriteArrayList> _listeners = new CopyOnWriteArrayList<>();
@Override
public void setNodeId(String nodeId) {
_nodeId = nodeId;
}
public void start() {
if (_runStatus == RunStatus.CLOSED) {
throw new IllegalStateException("Asked to start, but runStatus==CLOSED, so have already been closed.");
}
if (_runStatus == RunStatus.RUNNING) {
log.info("Asked to start, but already running. Ignoring.");
return;
}
// First set _runStatus to RUNNING, so threads will be happy
_runStatus = RunStatus.RUNNING;
log.info("Starting ActiveMQ Broker statistics querier.");
String id = "(Querier@" + Integer.toHexString(System.identityHashCode(this)) + ")";
_sendStatsRequestMessages_Thread = new Thread(this::sendStatsRequestMessages,
"MatsBrokerMonitor.ActiveMQ: Send Statistics request messages " + id);
_receiveDestinationsStatsReplyMessages_Thread = new Thread(this::receiveStatisticsReplyMessagesRunnable,
"MatsBrokerMonitor.ActiveMQ: Receive&Process Statistics reply messages " + id);
// :: First start reply consumer
_receiveDestinationsStatsReplyMessages_Thread.start();
// .. then starting requester (it will chill a small tad after getting connection before doing first request)
_sendStatsRequestMessages_Thread.start();
}
@Override
public void close() {
if (_runStatus == RunStatus.CLOSED) {
log.warn("Asked to close, but runStatus==CLOSED, so must already have been stopped. Ignoring.");
return;
}
if (_runStatus == RunStatus.NOT_STARTED) {
log.warn("Asked to close, but runStatus==NOT_STARTED, so have never been started."
+ " Setting directly to CLOSED.");
_runStatus = RunStatus.CLOSED;
return;
}
// First set _runStatus to CLOSED, so that any loops and checks will exit.
_runStatus = RunStatus.CLOSED;
log.info("Asked to close. Set runStatus to CLOSED, closing Connections and interrupting threads.");
// Notify the request sender - it'll close the Connection on its way out.
synchronized (_waitObject_And_ForceUpdateOutstandingCorrelationIds) {
_waitObject_And_ForceUpdateOutstandingCorrelationIds.notifyAll();
}
// Closing Connections for the receiver - they'll wake up from 'con.receive()'.
closeConnectionIfNonNullIgnoreException(_receiveDestinationsStatsReplyMessages_Connection);
// Check that all threads exit
try {
_sendStatsRequestMessages_Thread.join(TIMEOUT_MILLIS_GRACEFUL_THREAD_SHUTDOWN);
_receiveDestinationsStatsReplyMessages_Thread.join(TIMEOUT_MILLIS_GRACEFUL_THREAD_SHUTDOWN);
}
catch (InterruptedException e) {
/* ignore */
}
// .. interrupt the request sender if the above didn't work.
interruptThread(_sendStatsRequestMessages_Thread);
// .. interrupt the receiver too if they haven't gotten out.
interruptThread(_receiveDestinationsStatsReplyMessages_Thread);
// Null out Threads
_sendStatsRequestMessages_Thread = null;
_receiveDestinationsStatsReplyMessages_Thread = null;
}
@Override
public void setMatsDestinationPrefix(String matsDestinationPrefix) {
if (_runStatus != RunStatus.NOT_STARTED) {
throw new IllegalStateException("Tried setting matsDestinationPrefix, but runStatus != NOT_STARTED.");
}
_matsDestinationPrefix = matsDestinationPrefix;
}
private static void closeConnectionIfNonNullIgnoreException(Connection connection) {
if (connection == null) {
return;
}
try {
connection.close();
}
catch (JMSException e) {
log.warn("Got [" + e.getClass().getSimpleName() + "] when trying to close Connection [" + connection
+ "], ignoring.");
}
}
private void interruptThread(Thread thread) {
if (thread != null) {
thread.interrupt();
}
}
@Override
public void registerListener(Consumer listener) {
_listeners.add(listener);
}
private String constructCorrelationIdToSend(boolean forced, boolean fullUpdate, String correlationId) {
return (forced ? CORRELATION_ID_PREFIX_FORCED : CORRELATION_ID_PREFIX_SCHEDULED)
+ ":" + CORRELATION_ID_PARAMETER_FULL_UPDATE + "." + fullUpdate
+ ":" + CORRELATION_ID_PARAMETER_NODE_ID + "." + _nodeId
+ ":" + correlationId;
}
private static CorrSplit splitCorrelation(String sentCorrelationid) {
int firstColon = sentCorrelationid.indexOf(':');
String prefix = sentCorrelationid.substring(0, firstColon);
int secondColon = sentCorrelationid.indexOf(':', firstColon + 1);
String full = sentCorrelationid.substring(firstColon + 1, secondColon);
int thirdColon = sentCorrelationid.indexOf(':', secondColon + 1);
String node = sentCorrelationid.substring(secondColon + 1, thirdColon);
// The actual CorrelationId is the rest of the String
String correlationId = sentCorrelationid.substring(thirdColon + 1);
boolean forced = CORRELATION_ID_PREFIX_FORCED.equals(prefix);
boolean fullUpdate = full.contains("true");
String nodeId = node.substring(CORRELATION_ID_PARAMETER_NODE_ID.length() + 1);
return new CorrSplit(forced, fullUpdate, nodeId, correlationId);
}
private static class CorrSplit {
final boolean forced;
final boolean fullUpdate;
final String nodeId;
final String correlationId;
public CorrSplit(boolean forced, boolean fullUpdate, String nodeId, String correlationId) {
this.forced = forced;
this.fullUpdate = fullUpdate;
this.nodeId = nodeId;
this.correlationId = correlationId;
}
}
@Override
public void forceUpdate(String correlationId, boolean fullUpdate) {
synchronized (_waitObject_And_ForceUpdateOutstandingCorrelationIds) {
_waitObject_And_ForceUpdateOutstandingCorrelationIds.add(constructCorrelationIdToSend(true, fullUpdate,
correlationId));
// ?: Are there WAY too many outstanding correlation Ids?
if (_waitObject_And_ForceUpdateOutstandingCorrelationIds
.size() > MAX_NUMBER_OF_OUTSTANDING_CORRELATION_IDS) {
// -> Yes, so then we're actually screwed - there are obviously no response from the broker
log.error("We have " + MAX_NUMBER_OF_OUTSTANDING_CORRELATION_IDS + " outstanding forced updates with"
+ " correlationIds, which implies that the broker does not respond. We'll just ditch them all"
+ " to prevent OOME; you need to figure out the problem: Probably the"
+ " Statistics[Broker]Plugin is not installed on the ActiveMQ server.");
_waitObject_And_ForceUpdateOutstandingCorrelationIds.clear();
}
_waitObject_And_ForceUpdateOutstandingCorrelationIds.notifyAll();
}
}
private static class ActiveMqBrokerStatsEventImpl implements ActiveMqBrokerStatsEvent {
private final String _correlationId;
private final boolean _fullUpdate;
private final boolean _statsEventOriginatedOnThisNode;
private final String _originatingNodeId;
private final double _requestReplyLatencyMillis;
public ActiveMqBrokerStatsEventImpl(String correlationId, boolean fullUpdate,
boolean statsEventOriginatedOnThisNode,
String originatingNodeId, double requestReplyLatencyMillis) {
_correlationId = correlationId;
_fullUpdate = fullUpdate;
_statsEventOriginatedOnThisNode = statsEventOriginatedOnThisNode;
_originatingNodeId = originatingNodeId;
_requestReplyLatencyMillis = requestReplyLatencyMillis;
}
/**
* Only present if forced update.
*/
@Override
public Optional getCorrelationId() {
return Optional.ofNullable(_correlationId);
}
@Override
public boolean isFullUpdate() {
return _fullUpdate;
}
@Override
public boolean isStatsEventOriginatedOnThisNode() {
return _statsEventOriginatedOnThisNode;
}
@Override
public Optional getOriginatingNodeId() {
return Optional.ofNullable(_originatingNodeId);
}
@Override
public OptionalDouble getStatsRequestReplyLatencyMillis() {
return _requestReplyLatencyMillis >= 0
? OptionalDouble.of(_requestReplyLatencyMillis)
: OptionalDouble.empty();
}
}
public Optional getCurrentBrokerStatsDto() {
return Optional.ofNullable(_currentBrokerStatsDto);
}
public ConcurrentNavigableMap getCurrentDestinationStatsDtos() {
return _currentDestinationStatsDtos;
}
// ===== IMPLEMENTATION =====
private void sendStatsRequestMessages() {
Connection sendRequestMessages_Connection = null;
while (_runStatus == RunStatus.RUNNING) {
try {
sendRequestMessages_Connection = _connectionFactory.createConnection();
sendRequestMessages_Connection.start();
Session session = sendRequestMessages_Connection.createSession(false, Session.AUTO_ACKNOWLEDGE);
MessageProducer producer = session.createProducer(null);
// If there are no receiver (StatisticsPlugin not installed), then try to prevent build-up of msgs.
producer.setTimeToLive(5 * 60 * 1000); // 5 minutes.
// Just to point out that there is no need to save these messages if they cannot be handled.
producer.setDeliveryMode(DeliveryMode.NON_PERSISTENT);
/*
* So, here I'll throw in a bit of premature optimization: If the MatsDestinationPrefix is a proper
* "path-based" prefix with a dot as last character, like it is per default ("mats."), we'll use a
* specific query that only requests these. However, we'll then also need to ask specifically for the
* DLQs, which again are pre-prefixed with "DLQ.", and for the global DLQ just in case the user hasn't
* configured his message broker sanely with /individual DLQ policies/.
*
* Thus: If the MatsDestinationPrefix isn't set, or it is a crap prefix that cannot be used in the
* wildcard syntax of ActiveMQ (e.g. "endre:", which doesn't end in a dot), we ask one query to get
* every single queue and topic (well, two, since one for queues and one for topics, go figure) - but
* then get tons of replies which we don't care about (at least all advisory topics for every single
* queue and topic, and these statistics queues and topics - as well as any other queues that the user
* also run on the ActiveMQ instance). If it is a good prefix that can be used with the wildcard syntax,
* we have to ask three queries (well, four), but only get exactly the ones we care about.
*
* (The reason for this badness is my current main user of Mats which for legacy reasons employ such a
* bad prefix.)
*/
Queue requestQueuesQueue_main;
Topic requestTopicsTopic_main;
Queue requestQueuesQueue_individualDlq;
Queue requestQueuesQueue_globalDlq;
// ?: Do we have a proper "path-style" prefix, i.e. ending with a "."?
if ((_matsDestinationPrefix != null) && _matsDestinationPrefix.endsWith(".")) {
// -> Yes, this is a proper "path-style" prefix, so we can reduce the query from "all" to "mats
// only".
String queryRequestDestination_specific = QUERY_REQUEST_DESTINATION_PREFIX + "."
+ _matsDestinationPrefix + ">";
String queryRequestDestination_specific_DLQ = QUERY_REQUEST_DESTINATION_PREFIX + ".DLQ."
+ _matsDestinationPrefix + ">";
log.info("The matsDestinationPrefix was set to [" + _matsDestinationPrefix + "], and it is a proper"
+ " \"path-style\" prefix (i.e. ending with a dot), thus restricting the query destination"
+ " by employing [" + queryRequestDestination_specific + "], ["
+ queryRequestDestination_specific_DLQ + "]"
+ " and the Global DLQ [" + Statics.ACTIVE_MQ_GLOBAL_DLQ_NAME + "]");
requestQueuesQueue_main = session.createQueue(queryRequestDestination_specific);
requestTopicsTopic_main = session.createTopic(queryRequestDestination_specific);
requestQueuesQueue_individualDlq = session.createQueue(queryRequestDestination_specific_DLQ);
requestQueuesQueue_globalDlq = session.createQueue(QUERY_REQUEST_DESTINATION_PREFIX + "."
+ Statics.ACTIVE_MQ_GLOBAL_DLQ_NAME);
}
else {
// -> No, this is a bad prefix that cannot utilize the wildcard syntax, so have to ask for every
// single queue and topic.
String queryRequestDestination_all = QUERY_REQUEST_DESTINATION_PREFIX + ".>";
log.info("The matsDestinationPrefix was set to [" + _matsDestinationPrefix + "], but this isn't"
+ " a proper \"path-style\" prefix (it is not ending with a dot), thus cannot restrict"
+ " the query to mats-specific destinations, but must query for all destinations by"
+ " employing [" + queryRequestDestination_all + "]");
requestQueuesQueue_main = session.createQueue(queryRequestDestination_all);
requestTopicsTopic_main = session.createTopic(queryRequestDestination_all);
// Since the two above gets all queues and topics, the DLQs will also be gotten.
requestQueuesQueue_individualDlq = null;
requestQueuesQueue_globalDlq = null;
}
// :: Create queue for the "zero-termination", sent in a separate request query
// NOTE: We just ask for a queue which shall not be there, but request the "zero termination".
Queue requestQueuesQueue_zeroQueuesMatchWithNullTermination = session.createQueue(
QUERY_REQUEST_DESTINATION_PREFIX + ".No_Match_Whatsoever_"
+ Long.toString(Math.abs(ThreadLocalRandom.current().nextLong()), 36));
// Request BrokerStats topic
Topic requestBrokerTopic = session.createTopic(QUERY_REQUEST_BROKER);
// Reply topic for all statistics messages (this thread)
Topic replyStatisticsTopic = session.createTopic(QUERY_REPLY_STATISTICS_TOPIC);
// Chill a small tad before sending first request, so that receivers hopefully have started.
// Notice: It isn't particularly bad if they haven't, they'll just miss the first request/reply.
chill((long) (CHILL_MILLIS_BEFORE_FIRST_STATS_REQUEST
* (1 + ThreadLocalRandom.current().nextDouble())));
String correlationIdToSend = null;
while (_runStatus == RunStatus.RUNNING) {
// :: Request stats for Broker
Message requestBrokerMsg = session.createMessage();
requestBrokerMsg.setJMSReplyTo(replyStatisticsTopic);
producer.send(requestBrokerTopic, requestBrokerMsg);
// ::: Destinations
// (Notice: Directing replyTo for both Queues and Topics reply to same receiver.)
// :: Request stats for Queues
Message requestQueuesMsg = session.createMessage();
set_includeFirstMessageTimestamp(requestQueuesMsg);
requestQueuesMsg.setJMSReplyTo(replyStatisticsTopic);
// :: Request stats for Topics
Message requestTopicsMsg = session.createMessage();
set_includeFirstMessageTimestamp(requestTopicsMsg);
requestTopicsMsg.setJMSReplyTo(replyStatisticsTopic);
// :: Send destination stats messages
// Clear the count of destinations received since this request (we start counting from now)
_countOfDestinationsReceivedAfterRequest.set(0);
// We start timing from now
_nanosAtStart_RequestQuery = System.nanoTime();
producer.send(requestQueuesQueue_main, requestQueuesMsg);
producer.send(requestTopicsTopic_main, requestTopicsMsg);
// ?: Are we using a proper "path-style" mats destination prefix?
if (requestQueuesQueue_individualDlq != null) {
// -> Yes, we're using a "path-style" mats destination prefix.
// The main queue (and topic) query is then a specific query for only those relevant queues.
// Therefore, we'll also need to ask for the individual DLQs, and for the ActiveMQ Global DLQ.
// (Asking for the Global DLQ is just to point things out for users that haven't configured
// individual DLQ policy. We won't try to fix this problem fully, i.e. won't "distribute"
// these DLQs to the individual mats stages they really belong to, as it is pretty much
// infeasible if there are more than very few, and that the proper fix is very simple: Use a
// frikkin' individual DLQ policy.)
Message requestIndividualDlqMsg = session.createMessage();
set_includeFirstMessageTimestamp(requestIndividualDlqMsg);
requestIndividualDlqMsg.setJMSReplyTo(replyStatisticsTopic);
Message requestGlobalDlqMsg = session.createMessage();
set_includeFirstMessageTimestamp(requestGlobalDlqMsg);
requestGlobalDlqMsg.setJMSReplyTo(replyStatisticsTopic);
producer.send(requestQueuesQueue_individualDlq, requestIndividualDlqMsg);
producer.send(requestQueuesQueue_globalDlq, requestGlobalDlqMsg);
}
// :: Send the null-termination query (query w/o any destination replies, only the null-terminator)
// Do we currently have a specific correlationId?
if (correlationIdToSend == null) {
// -> No specific correlation Id, so make a "normal" Scheduled correlationId.
// Should we do a full-update, since long time since last full update?
boolean timeBasedFullUpdate = (System.currentTimeMillis()
- _lastFullUpdatePropagatedMillis) > FULL_UPDATE_INTERVAL;
// A random correlationId. Not really used.
String random = Long.toString(ThreadLocalRandom.current().nextLong(), 36);
correlationIdToSend = constructCorrelationIdToSend(false, timeBasedFullUpdate, random);
}
// .. construct the null-termination message
Message requestNullTermination = session.createMessage();
requestNullTermination.setJMSReplyTo(replyStatisticsTopic);
// .. set the special property which directs the StatisticsPlugin to "empty terminate" the replies.
requestNullTermination.setBooleanProperty(QUERY_REQUEST_DENOTE_END_LIST, true);
requestNullTermination.setJMSCorrelationID(correlationIdToSend);
producer.send(requestQueuesQueue_zeroQueuesMatchWithNullTermination, requestNullTermination);
// We've done this job, clear out correlationId
correlationIdToSend = null;
// :: Chill and loop
synchronized (_waitObject_And_ForceUpdateOutstandingCorrelationIds) {
// :: Go into wait: Either we're waking by interval, or by forced update
// Loop till we decide that we should do request.
while (_runStatus == RunStatus.RUNNING) {
// ?: Do we have a forced update waiting?
if (!_waitObject_And_ForceUpdateOutstandingCorrelationIds.isEmpty()) {
// -> Yes, so pop the force update correlationId before looping.
correlationIdToSend = _waitObject_And_ForceUpdateOutstandingCorrelationIds.remove();
// Break out of wait-loop
break;
}
// E-> No forced update, so calculate how long to wait
// 10% randomness, 5% to both sides
long randomRange = (long) Math.max(150, _updateIntervalMillis * .1d);
long randomness = Math.round(ThreadLocalRandom.current().nextDouble() * randomRange);
long currentWait = Math.max(1, _updateIntervalMillis - (randomRange / 2) + randomness);
// Do wait.
_waitObject_And_ForceUpdateOutstandingCorrelationIds.wait(currentWait);
// ?: Not running anymore?
if (_runStatus != RunStatus.RUNNING) {
// -> Break out of wait loop (and then all the way out, due to loop conditionals.).
break;
}
// ?: Forced update waiting?
if (!_waitObject_And_ForceUpdateOutstandingCorrelationIds.isEmpty()) {
// -> Loop (waitloop) to get correlationId (and then break).
continue;
}
// :: Check if we actually should do the request, or if the other node has already done it.
// ?: Is the last received stats message older than 75% of our interval?
if (_lastStatsUpdateMessageReceived < (System.currentTimeMillis()
- (_updateIntervalMillis * 0.75d))) {
// -> Yes, the message is overdue enough, so let's do the request
log.debug("Last received StatsUpdateMessage is too old (I am probably first of the"
+ " nodes, or the only node); I will do the request");
// Break out of wait-loop
break;
}
else {
log.debug("Last received update is pretty fresh (probably gotten from another node);"
+ " I will NOT do the request.");
}
}
}
}
}
catch (Throwable t) {
// ?: Exiting?
if (_runStatus != RunStatus.RUNNING) {
// -> Yes, exiting, so get out.
break;
}
log.warn("Got a [" + t.getClass().getSimpleName() + "] in the query-loop."
+ " Attempting to close JMS Connection if gotten, then chill-waiting, then trying again.", t);
closeConnectionIfNonNullIgnoreException(sendRequestMessages_Connection);
chill(CHILL_MILLIS_WAIT_AFTER_THROWABLE_IN_RECEIVE_LOOPS);
}
}
// To exit, we're signalled via interrupt - it is our job to close Connection (it is a local variable)
closeConnectionIfNonNullIgnoreException(sendRequestMessages_Connection);
log.info("Got asked to exit, and that we do!");
}
private void set_includeFirstMessageTimestamp(Message msg) throws JMSException {
msg.setBooleanProperty(QUERY_REQUEST_DESTINATION_INCLUDE_FIRST_MESSAGE_TIMESTAMP, true);
}
// Only used within receiveStatisticsReplyMessagesRunnable-thread.
@SuppressWarnings("unchecked")
private void receiveStatisticsReplyMessagesRunnable() {
OUTERLOOP: while (_runStatus == RunStatus.RUNNING) {
try {
_receiveDestinationsStatsReplyMessages_Connection = _connectionFactory.createConnection();
_receiveDestinationsStatsReplyMessages_Connection.start();
Session session = _receiveDestinationsStatsReplyMessages_Connection.createSession(false,
Session.AUTO_ACKNOWLEDGE);
Topic replyTopic = session.createTopic(QUERY_REPLY_STATISTICS_TOPIC);
MessageConsumer consumer = session.createConsumer(replyTopic);
long nanosAtEnd_lastMessageReceived = 0;
while (_runStatus == RunStatus.RUNNING) {
// Note: The logic here is that we'll get a bunch of statistics reply messages, ended by either
// an explicit "empty terminator", or by timing out. This first, timed receive is for the "bunch",
// while the later, indefinite receive is for waiting for the next batch.
// Since I do not feel utterly confident about the ordering of the messages vs. the multiple
// requests and the last "empty terminator", we should also handle that a bunch is split into
// two parts: Some replies, the empty terminator, and the rest of the replies. This is not handled
// /good/, but we will at least receive the stats messages. However, the fired event will be too
// early. This will definitely be a "eventually consistent" scenario - so lets hope the ordering
// holds, in which case we have a very good consistency wrt. when sending the update event.
// The net effect here is that we will receive stats updates for everything we want, but the
// ordering and clean division into "sets" based on the requests being sent cannot be totally
// guaranteed - both due to mentioned ordering, but also the fact that there should be multiple
// nodes running this (so that we can get interleaving of requests), and that downstream can at
// any point request a forced update request, which might come right after a scheduled update req.
// However, since we do stack everything we get up into a persistent memory map on this side, we
// will just get more and more current information. As long as we eventually notify our listeners
// (the ActiveMqMatsBrokerMonitor), the downstream will get a good view about the broker state.
// (The map is scavenged based on time: If there are old destinations that haven't gotten an update
// for a longish time, it is assumed that the destination is gone from the broker).
// Go into timed receive (receiving "the rest of the bunch", but handling startup)
MapMessage statsMsg = (MapMessage) consumer.receive(
TIMEOUT_MILLIS_FOR_LAST_MESSAGE_IN_BATCH_FOR_DESTINATION_STATS);
// NOTE: We're using an undocumented feature of ActiveMQ's StatisticsBrokerPlugin whereby if we add
// a special marker to the query, the replies will be "empty terminated" by an empty MapMessage.
// ?: Check run flag - exit if we're not running anymore.
if (_runStatus != RunStatus.RUNNING) {
break OUTERLOOP;
}
// ?: Did we either get a null, or get an "empty" MapMessage?
// (Null would be timeout for stats, empty MapMessage would be the "null termination" for stats.)
if ((statsMsg == null) || (statsMsg.getObject("brokerName") == null)) {
// -> null or empty, but runStatus==RUNNING, so this was a timeout or "terminator".
// Now evaluate whether this was the end of a batch of destination stats messages, and if
// so, fire the event.
evaluateEndOfBatch(statsMsg, nanosAtEnd_lastMessageReceived);
// ----- So, either startup, or we've finished a batch of messages:
// Go into indefinite receive: Waiting for _next batch_ of messages triggered by _next stats
// request_. (Or null message, resulting from outside close due to shutdown)
statsMsg = (MapMessage) consumer.receive();
}
// ----- We've either got a message from the loop, or a new message after a batch.
// ?: Was this a null-message, most probably denoting that we're exiting?
if (statsMsg == null) {
// -> Yes, null message received.
log.info("Received null message from consumer.receive(), assuming shutdown.");
if (_runStatus != RunStatus.RUNNING) {
break OUTERLOOP;
}
throw new UnexpectedNullMessageReceivedException("Null message received,"
+ " but runFlag still true?!");
}
// :: Evaluate whether it is a BrokerStatistics or DestinationStatistics message
if (statsMsg.getObject("destinationName") != null) {
// -> Destination stats
// This was a destination stats message - count it
_countOfDestinationsReceivedAfterRequest.incrementAndGet();
_countOfDestinationsReceivedSinceLastBatch++;
// .. log time we received this destination stats message
nanosAtEnd_lastMessageReceived = System.nanoTime();
DestinationStatsDto destinationStatsDto = mapMessageToDestinationStatsDto(statsMsg);
_currentDestinationStatsDtos.put(destinationStatsDto.destinationName, destinationStatsDto);
// Update that we've gotten a stats-message
_lastStatsUpdateMessageReceived = System.currentTimeMillis();
if (log.isTraceEnabled()) log.trace("Got DestinationStats: " + destinationStatsDto);
}
else if (statsMsg.getObject("storeUsage") != null) {
// -> Broker stats
_currentBrokerStatsDto = mapMessageToBrokerStatsDto(statsMsg);
// Update that we've gotten a stats-message
_lastStatsUpdateMessageReceived = System.currentTimeMillis();
if (log.isTraceEnabled()) log.trace("Got BrokerStats: " + _currentBrokerStatsDto);
}
else {
log.info("Got a JMS Message that was neither destination stats nor broker stats."
+ " This is probably an unexpected 'null termination' which happened because more than"
+ " one instance of the querier sent stats query at the same time, thus getting two"
+ " replies concurrently. Should not happen often since we have forced randomness in"
+ " query interval. If it does happen often, either your interval is too low,"
+ " you have high load or latency, very many queues - or gimme a call! \n" + statsMsg);
}
}
}
catch (Throwable t) {
// ?: Exiting?
if (_runStatus != RunStatus.RUNNING) {
// -> Yes, exiting, so get out.
break;
}
log.warn("Got a [" + t.getClass().getSimpleName() + "] in the receive-loop."
+ " Attempting to close JMS Connection if gotten, then chill-waiting, then trying again.", t);
closeConnectionIfNonNullIgnoreException(_receiveDestinationsStatsReplyMessages_Connection);
chill(CHILL_MILLIS_WAIT_AFTER_THROWABLE_IN_RECEIVE_LOOPS);
}
}
// To exit, we're signalled via the JMS Connection being closed; Our job is just to null it on our way out.
_receiveDestinationsStatsReplyMessages_Connection = null;
log.info("Got asked to exit, and that we do!");
}
private void evaluateEndOfBatch(MapMessage statsMsg, long nanosAtEnd_lastMessageReceived) throws JMSException {
// :: Extract information from the JMS Message's CorrelationId.
// ?: Have we gotten (bunch of) stats messages by now?
if (_countOfDestinationsReceivedSinceLastBatch > 0) {
// -> We've gotten 1 or more stats messages.
// We've now (hopefully) received a full set of destination stats.
// :: Pick out pieces of the raw correlation if we came here based on a final last message.
// (Does not work if we came here due to timeout, as we don't then have a message.)
String correlationId = null;
boolean isFullUpdate = false;
String originatingNodeId = null;
boolean requestSameNode = false;
if (statsMsg != null) {
String raw = statsMsg.getJMSCorrelationID();
if (raw != null) {
CorrSplit corrSplit = splitCorrelation(raw);
// Only use the correlationId if it was supplied via a forceUpdate.
correlationId = corrSplit.forced ? corrSplit.correlationId : null;
isFullUpdate = corrSplit.fullUpdate;
originatingNodeId = corrSplit.nodeId;
requestSameNode = _nodeId.equals(originatingNodeId);
}
}
// :: Log the update with metrics.
MDC.put("mats.mbm.fullUpdate", Boolean.toString(isFullUpdate));
MDC.put("mats.mbm.requestSameNode", Boolean.toString(requestSameNode));
if (correlationId != null) {
MDC.put("mats.mbm.correlationId", correlationId);
}
String msg = "#MBM_UPDATE# (Request " + (requestSameNode ? "IS" : "is NOT") + " from this node)"
+ " We've received a batch of ["
+ _countOfDestinationsReceivedSinceLastBatch
+ "] destination stats messages, current number of destinations ["
+ _currentDestinationStatsDtos.size()
+ "]";
if (requestSameNode) {
long nanosRequestFinalReplyLatency = nanosAtEnd_lastMessageReceived
- _nanosAtStart_RequestQuery;
long nanosBetweenQueryAndNow = System.nanoTime() - _nanosAtStart_RequestQuery;
long extraLatency = nanosBetweenQueryAndNow - nanosRequestFinalReplyLatency;
MDC.put("mats.mbm.requestReplyLatencyMillis",
Double.toString(ms3(nanosRequestFinalReplyLatency)));
MDC.put("mats.mbm.extraLatency",
Double.toString(ms3(extraLatency)));
msg += ", request-to-final-reply latency:"
+ " [" + ms3(nanosRequestFinalReplyLatency) + "] ms"
+ " - extra latency [" + ms3(extraLatency) + "] ms";
}
msg += ". Notifying local listeners.";
log_update.info(msg);
MDC.clear();
/*
* :: If this is a full update, then note this. We decide whether full update on the request-sending side,
* and note the last full update on ActiveMq _replies received_ side, so that if the other node did the full
* update _request_, all nodes will notice and agree on when the last full update was propagated. (The
* sender will automatically tag request as full-update if > X mins since last)
*/
if (isFullUpdate) {
_lastFullUpdatePropagatedMillis = System.currentTimeMillis();
}
/*
* Clean out old destinations by scavenging stats which no longer is getting updates (i.e. not existing
* anymore).
*
* Would have been nice to run this as a full-update when we scavenge some destinations, but that could be
* messy as we might not be sure the multiple MBM nodes are in sync wrt. "old" destinations. So, we'll just
* let the next scheduled full update do the job.
*/
long longAgo = System.currentTimeMillis() - SCAVENGE_OLD_STATS_SECONDS * 1000;
Iterator currentStatsIterator = _currentDestinationStatsDtos
.values().iterator();
while (currentStatsIterator.hasNext()) {
DestinationStatsDto stats = currentStatsIterator.next();
if (stats.statsReceived.toEpochMilli() < longAgo) {
log.info("Removing destination which haven't gotten updates for ["
+ SCAVENGE_OLD_STATS_SECONDS
+ "] seconds: [" + stats.destinationName + "]");
currentStatsIterator.remove();
}
}
// :: Notify listeners
ActiveMqBrokerStatsEventImpl event = new ActiveMqBrokerStatsEventImpl(correlationId,
isFullUpdate, requestSameNode, originatingNodeId,
requestSameNode ? (System.nanoTime() - _nanosAtStart_RequestQuery) / 1_000_000d : -1);
for (Consumer listener : _listeners) {
listener.accept(event);
}
_countOfDestinationsReceivedSinceLastBatch = 0;
}
else {
// -> No, we haven't got any messages - probably startup
log.debug("We haven't gotten any stats messages - wait for first one.");
}
}
private BrokerStatsDto mapMessageToBrokerStatsDto(MapMessage mm) throws JMSException {
BrokerStatsDto dto = new BrokerStatsDto();
mapMessageToCommonStatsDto(mm, dto);
dto.stompSsl = (String) mm.getObject("stomp+ssl");
dto.ssl = (String) mm.getObject("ssl");
dto.stomp = (String) mm.getObject("stomp");
dto.openwire = (String) mm.getObject("openwire");
dto.vm = (String) mm.getObject("vm");
dto.dataDirectory = (String) mm.getObject("dataDirectory");
dto.tempUsage = (long) mm.getObject("tempUsage");
dto.tempLimit = (long) mm.getObject("tempLimit");
dto.tempPercentUsage = (int) mm.getObject("tempPercentUsage");
dto.storeUsage = (long) mm.getObject("storeUsage");
dto.storePercentUsage = (int) mm.getObject("storePercentUsage");
dto.storeLimit = (long) mm.getObject("storeLimit");
return dto;
}
private DestinationStatsDto mapMessageToDestinationStatsDto(MapMessage mm) throws JMSException {
DestinationStatsDto dto = new DestinationStatsDto();
mapMessageToCommonStatsDto(mm, dto);
dto.destinationName = (String) mm.getObject("destinationName");
dto.firstMessageTimestamp = getTimestampFromMapMessage(mm, "firstMessageTimestamp");
return dto;
}
private Optional getTimestampFromMapMessage(MapMessage mm, String key) throws JMSException {
long millis = mm.getLong(key);
return millis != 0 ? Optional.of(Instant.ofEpochMilli(millis)) : Optional.empty();
}
private void mapMessageToCommonStatsDto(MapMessage mm, CommonStatsDto dto) throws JMSException {
dto.statsReceived = Instant.now(_clock);
dto.brokerId = (String) mm.getObject("brokerId");
dto.brokerName = (String) mm.getObject("brokerName");
dto.brokerTime = Instant.ofEpochMilli(mm.getJMSTimestamp());
dto.size = (long) mm.getObject("size");
dto.enqueueCount = (long) mm.getObject("enqueueCount");
dto.dequeueCount = (long) mm.getObject("dequeueCount");
dto.expiredCount = (long) mm.getObject("expiredCount");
dto.dispatchCount = (long) mm.getObject("dispatchCount");
dto.inflightCount = (long) mm.getObject("inflightCount");
dto.producerCount = (long) mm.getObject("producerCount");
dto.consumerCount = (long) mm.getObject("consumerCount");
dto.minEnqueueTime = (double) mm.getObject("minEnqueueTime");
dto.averageEnqueueTime = (double) mm.getObject("averageEnqueueTime");
dto.maxEnqueueTime = (double) mm.getObject("maxEnqueueTime");
dto.memoryUsage = (long) mm.getObject("memoryUsage");
dto.memoryLimit = (long) mm.getObject("memoryLimit");
dto.memoryPercentUsage = (int) mm.getObject("memoryPercentUsage");
dto.averageMessageSize = (long) mm.getObject("averageMessageSize");
dto.messagesCached = (long) mm.getObject("messagesCached");
}
private static class UnexpectedNullMessageReceivedException extends Exception {
public UnexpectedNullMessageReceivedException(String message) {
super(message);
}
}
private static void chill(long millis) {
log.debug("Chilling [" + millis + "] ms.");
try {
Thread.sleep(millis);
}
catch (InterruptedException e) {
log.info("Got interrupted while chill-waiting [" + millis + "].");
}
}
}
