Browse Source

Merge pull request #14425 from volodymyr-babak/edge-zombie-session-fix

Prevent dead Kafka consumer groups causing Edge sync failures
pull/14443/head
Viacheslav Klimov 10 months ago
committed by GitHub
parent
commit
93b625474f
No known key found for this signature in database GPG Key ID: B5690EEEBB952194
  1. 91
      application/src/main/java/org/thingsboard/server/service/edge/rpc/EdgeGrpcService.java
  2. 27
      application/src/main/java/org/thingsboard/server/service/edge/rpc/KafkaEdgeGrpcSession.java

91
application/src/main/java/org/thingsboard/server/service/edge/rpc/EdgeGrpcService.java

@ -69,6 +69,7 @@ import org.thingsboard.server.service.telemetry.TelemetrySubscriptionService;
import java.io.IOException; import java.io.IOException;
import java.io.InputStream; import java.io.InputStream;
import java.util.ArrayList; import java.util.ArrayList;
import java.util.Collection;
import java.util.HashMap; import java.util.HashMap;
import java.util.List; import java.util.List;
import java.util.Map; import java.util.Map;
@ -82,6 +83,7 @@ import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.Lock; import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantLock; import java.util.concurrent.locks.ReentrantLock;
import java.util.function.Consumer; import java.util.function.Consumer;
import java.util.function.Function;
import static org.thingsboard.server.service.state.DefaultDeviceStateService.ACTIVITY_STATE; import static org.thingsboard.server.service.state.DefaultDeviceStateService.ACTIVITY_STATE;
import static org.thingsboard.server.service.state.DefaultDeviceStateService.LAST_CONNECT_TIME; import static org.thingsboard.server.service.state.DefaultDeviceStateService.LAST_CONNECT_TIME;
@ -94,6 +96,7 @@ import static org.thingsboard.server.service.state.DefaultDeviceStateService.LAS
public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase implements EdgeRpcService { public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase implements EdgeRpcService {
private final ConcurrentMap<EdgeId, EdgeGrpcSession> sessions = new ConcurrentHashMap<>(); private final ConcurrentMap<EdgeId, EdgeGrpcSession> sessions = new ConcurrentHashMap<>();
private final ConcurrentMap<UUID, EdgeGrpcSession> sessionsById = new ConcurrentHashMap<>();
private final ConcurrentMap<EdgeId, Lock> sessionNewEventsLocks = new ConcurrentHashMap<>(); private final ConcurrentMap<EdgeId, Lock> sessionNewEventsLocks = new ConcurrentHashMap<>();
private final Map<EdgeId, Boolean> sessionNewEvents = new HashMap<>(); private final Map<EdgeId, Boolean> sessionNewEvents = new HashMap<>();
private final ConcurrentMap<EdgeId, ScheduledFuture<?>> sessionEdgeEventChecks = new ConcurrentHashMap<>(); private final ConcurrentMap<EdgeId, ScheduledFuture<?>> sessionEdgeEventChecks = new ConcurrentHashMap<>();
@ -283,6 +286,7 @@ public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase i
destroySession(session); destroySession(session);
session.cleanUp(); session.cleanUp();
sessions.remove(edgeId); sessions.remove(edgeId);
sessionsById.remove(session.getSessionId());
final Lock newEventLock = sessionNewEventsLocks.computeIfAbsent(edgeId, id -> new ReentrantLock()); final Lock newEventLock = sessionNewEventsLocks.computeIfAbsent(edgeId, id -> new ReentrantLock());
newEventLock.lock(); newEventLock.lock();
try { try {
@ -332,9 +336,15 @@ public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase i
TenantId tenantId = edge.getTenantId(); TenantId tenantId = edge.getTenantId();
log.info("[{}][{}] edge [{}] connected successfully.", tenantId, edgeGrpcSession.getSessionId(), edgeId); log.info("[{}][{}] edge [{}] connected successfully.", tenantId, edgeGrpcSession.getSessionId(), edgeId);
if (sessions.containsKey(edgeId)) { if (sessions.containsKey(edgeId)) {
destroySession(sessions.get(edgeId)); EdgeGrpcSession existing = sessions.get(edgeId);
if (existing != null) {
log.info("[{}][{}] Replacing existing session [{}] for edge [{}]", tenantId, edgeGrpcSession.getSessionId(), existing.getSessionId(), edgeId);
destroySession(existing);
sessionsById.remove(existing.getSessionId());
}
} }
sessions.put(edgeId, edgeGrpcSession); sessions.put(edgeId, edgeGrpcSession);
sessionsById.put(edgeGrpcSession.getSessionId(), edgeGrpcSession);
final Lock newEventLock = sessionNewEventsLocks.computeIfAbsent(edgeId, id -> new ReentrantLock()); final Lock newEventLock = sessionNewEventsLocks.computeIfAbsent(edgeId, id -> new ReentrantLock());
newEventLock.lock(); newEventLock.lock();
try { try {
@ -492,9 +502,9 @@ public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase i
private void onEdgeDisconnect(Edge edge, UUID sessionId) { private void onEdgeDisconnect(Edge edge, UUID sessionId) {
EdgeId edgeId = edge.getId(); EdgeId edgeId = edge.getId();
log.info("[{}][{}] edge disconnected!", edgeId, sessionId); log.info("[{}][{}] edge disconnected!", edgeId, sessionId);
EdgeGrpcSession toRemove = sessions.get(edgeId); EdgeGrpcSession current = sessions.get(edgeId);
if (toRemove.getSessionId().equals(sessionId)) { if (current != null && current.getSessionId().equals(sessionId)) {
toRemove = sessions.remove(edgeId); EdgeGrpcSession toRemove = sessions.remove(edgeId);
final Lock newEventLock = sessionNewEventsLocks.computeIfAbsent(edgeId, id -> new ReentrantLock()); final Lock newEventLock = sessionNewEventsLocks.computeIfAbsent(edgeId, id -> new ReentrantLock());
newEventLock.lock(); newEventLock.lock();
try { try {
@ -503,6 +513,7 @@ public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase i
newEventLock.unlock(); newEventLock.unlock();
} }
destroySession(toRemove); destroySession(toRemove);
sessionsById.remove(sessionId);
TenantId tenantId = toRemove.getEdge().getTenantId(); TenantId tenantId = toRemove.getEdge().getTenantId();
save(tenantId, edgeId, ACTIVITY_STATE, false); save(tenantId, edgeId, ACTIVITY_STATE, false);
long lastDisconnectTs = System.currentTimeMillis(); long lastDisconnectTs = System.currentTimeMillis();
@ -510,7 +521,18 @@ public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase i
pushRuleEngineMessage(toRemove.getEdge().getTenantId(), edge, lastDisconnectTs, TbMsgType.DISCONNECT_EVENT); pushRuleEngineMessage(toRemove.getEdge().getTenantId(), edge, lastDisconnectTs, TbMsgType.DISCONNECT_EVENT);
cancelScheduleEdgeEventsCheck(edgeId); cancelScheduleEdgeEventsCheck(edgeId);
} else { } else {
log.debug("[{}] edge session [{}] is not available anymore, nothing to remove. most probably this session is already outdated!", edgeId, sessionId); log.info("[{}] edge session [{}] is not current anymore. Attempting to destroy it by sessionId.", edgeId, sessionId);
EdgeGrpcSession stale = sessionsById.remove(sessionId);
if (stale != null) {
try {
destroySession(stale);
log.info("[{}][{}] Successfully destroyed stale session for edge [{}]", stale.getTenantId(), sessionId, edgeId);
} catch (Exception e) {
log.warn("[{}][{}] Failed to destroy stale session for edge [{}]", stale.getTenantId(), sessionId, edgeId, e);
}
} else {
log.debug("[{}] No session found by sessionId [{}] to destroy", edgeId, sessionId);
}
} }
edgeIdServiceIdCache.evict(edgeId); edgeIdServiceIdCache.evict(edgeId);
} }
@ -522,6 +544,9 @@ public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase i
session.getTenantId(), session.getEdge().getId(), session.getEdge().getName(), session.getSessionId()); session.getTenantId(), session.getEdge().getId(), session.getEdge().getName(), session.getSessionId());
zombieSessions.add(session); zombieSessions.add(session);
} }
} catch (Exception e) {
log.warn("[{}][{}] Exception during session destroy for edge [{}] with session id [{}]",
session.getTenantId(), session.getEdge().getId(), session.getEdge().getName(), session.getSessionId(), e);
} }
} }
@ -631,25 +656,9 @@ public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase i
private void cleanupZombieSessions() { private void cleanupZombieSessions() {
try { try {
List<EdgeId> toRemove = new ArrayList<>(); tryToDestroyZombieSessions(getZombieSessions(sessions.values()), s -> sessions.remove(s.getEdge().getId()));
for (EdgeGrpcSession session : sessions.values()) { tryToDestroyZombieSessions(getZombieSessions(sessionsById.values()), s -> sessionsById.remove(s.getSessionId()));
if (session instanceof KafkaEdgeGrpcSession kafkaSession &&
!kafkaSession.isConnected() &&
kafkaSession.getConsumer() != null &&
kafkaSession.getConsumer().getConsumer() != null &&
!kafkaSession.getConsumer().getConsumer().isStopped()) {
toRemove.add(kafkaSession.getEdge().getId());
}
}
for (EdgeId edgeId : toRemove) {
log.info("[{}] Destroying session for edge because edge is not connected", edgeId);
EdgeGrpcSession removed = sessions.get(edgeId);
if (removed instanceof KafkaEdgeGrpcSession kafkaSession) {
if (kafkaSession.destroy()) {
sessions.remove(edgeId);
}
}
}
zombieSessions.removeIf(zombie -> { zombieSessions.removeIf(zombie -> {
if (zombie.destroy()) { if (zombie.destroy()) {
log.info("[{}][{}] Successfully cleaned up zombie session [{}] for edge [{}].", log.info("[{}][{}] Successfully cleaned up zombie session [{}] for edge [{}].",
@ -666,4 +675,38 @@ public class EdgeGrpcService extends EdgeRpcServiceGrpc.EdgeRpcServiceImplBase i
} }
} }
private List<EdgeGrpcSession> getZombieSessions(Collection<EdgeGrpcSession> sessions) {
List<EdgeGrpcSession> result = new ArrayList<>();
for (EdgeGrpcSession session : sessions) {
if (isKafkaSessionAndZombie(session)) {
result.add(session);
}
}
return result;
}
private void tryToDestroyZombieSessions(List<EdgeGrpcSession> sessionsToRemove, Function<EdgeGrpcSession, EdgeGrpcSession> removeFunc) {
for (EdgeGrpcSession toRemove : sessionsToRemove) {
log.info("[{}] Destroying session for edge because edge is not connected", toRemove.getEdge().getId());
if (toRemove.destroy()) {
removeFunc.apply(toRemove);
}
}
}
private boolean isKafkaSessionAndZombie(EdgeGrpcSession session) {
if (session instanceof KafkaEdgeGrpcSession kafkaSession) {
log.debug("[{}] kafkaSession.isConnected() = {}, kafkaSession.getConsumer().getConsumer().isStopped() = {}",
kafkaSession.getEdge().getId(),
kafkaSession.isConnected(),
kafkaSession.getConsumer() != null ? kafkaSession.getConsumer().getConsumer() != null ? kafkaSession.getConsumer().getConsumer().isStopped() : null : null);
return !kafkaSession.isConnected() &&
kafkaSession.getConsumer() != null &&
kafkaSession.getConsumer().getConsumer() != null &&
!kafkaSession.getConsumer().getConsumer().isStopped();
}
return false;
}
} }

27
application/src/main/java/org/thingsboard/server/service/edge/rpc/KafkaEdgeGrpcSession.java

@ -101,8 +101,21 @@ public class KafkaEdgeGrpcSession extends EdgeGrpcSession {
@Override @Override
public ListenableFuture<Boolean> processEdgeEvents() { public ListenableFuture<Boolean> processEdgeEvents() {
if (!isConnected() || isSyncInProgress() || isHighPriorityProcessing) {
log.warn("[{}][{}] Session is not ready (connected={}, syncInProgress={}, highPriority={}), skip starting edge event consumer",
tenantId, edge != null ? edge.getId() : null, isConnected(), isSyncInProgress(), isHighPriorityProcessing);
return Futures.immediateFuture(Boolean.FALSE);
}
if (consumer == null || (consumer.getConsumer() != null && consumer.getConsumer().isStopped())) { if (consumer == null || (consumer.getConsumer() != null && consumer.getConsumer().isStopped())) {
try { try {
if (consumerExecutor != null && !consumerExecutor.isShutdown()) {
try {
consumerExecutor.shutdown();
awaitConsumerTermination();
} catch (Exception e) {
log.warn("[{}][{}] Failed to shutdown previous consumer executor", tenantId, edge.getId(), e);
}
}
this.consumerExecutor = Executors.newSingleThreadExecutor(ThingsBoardThreadFactory.forName("edge-event-consumer")); this.consumerExecutor = Executors.newSingleThreadExecutor(ThingsBoardThreadFactory.forName("edge-event-consumer"));
this.consumer = QueueConsumerManager.<TbProtoQueueMsg<ToEdgeEventNotificationMsg>>builder() this.consumer = QueueConsumerManager.<TbProtoQueueMsg<ToEdgeEventNotificationMsg>>builder()
.name("TB Edge events [" + edge.getId() + "]") .name("TB Edge events [" + edge.getId() + "]")
@ -133,6 +146,7 @@ public class KafkaEdgeGrpcSession extends EdgeGrpcSession {
public boolean destroy() { public boolean destroy() {
try { try {
if (consumer != null) { if (consumer != null) {
log.info("[{}][{}] Stopping edge event consumer...", tenantId, edge != null ? edge.getId() : null);
consumer.stop(); consumer.stop();
} }
} catch (Exception e) { } catch (Exception e) {
@ -141,16 +155,25 @@ public class KafkaEdgeGrpcSession extends EdgeGrpcSession {
} }
consumer = null; consumer = null;
try { try {
if (consumerExecutor != null) { if (consumerExecutor != null && !consumerExecutor.isShutdown()) {
consumerExecutor.shutdown(); consumerExecutor.shutdown();
awaitConsumerTermination();
} }
} catch (Exception e) { } catch (Exception e) {
log.warn("[{}][{}] Failed to shutdown consumer executor", tenantId, edge.getId(), e); log.warn("[{}][{}] Failed to shutdown edge event consumer executor", tenantId, edge.getId(), e);
return false; return false;
} }
return true; return true;
} }
private void awaitConsumerTermination() {
try {
consumerExecutor.awaitTermination(5, java.util.concurrent.TimeUnit.SECONDS);
} catch (InterruptedException ie) {
log.warn("[{}][{}] Interrupted while awaiting consumer executor termination", tenantId, edge.getId());
}
}
@Override @Override
public void cleanUp() { public void cleanUp() {
String topic = topicService.buildEdgeEventNotificationsTopicPartitionInfo(tenantId, edge.getId()).getTopic(); String topic = topicService.buildEdgeEventNotificationsTopicPartitionInfo(tenantId, edge.getId()).getTopic();

Loading…
Cancel
Save