All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.jgroups.protocols.pbcast.FLUSH Maven / Gradle / Ivy

Go to download

This artifact provides a single jar that contains all classes required to use remote EJB and JMS, including all dependencies. It is intended for use by those not using maven, maven users should just import the EJB and JMS BOM's instead (shaded JAR's cause lots of problems with maven, as it is very easy to inadvertently end up with different versions on classes on the class path).

There is a newer version: 34.0.0.Final
Show newest version
package org.jgroups.protocols.pbcast;

import org.jgroups.*;
import org.jgroups.annotations.*;
import org.jgroups.conf.AttributeType;
import org.jgroups.stack.Protocol;
import org.jgroups.util.*;

import java.io.DataInput;
import java.io.DataOutput;
import java.io.IOException;
import java.util.*;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.TimeoutException;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.locks.Condition;
import java.util.concurrent.locks.ReentrantLock;
import java.util.function.Supplier;

/**
 * Flush, as it name implies, forces group members to flush their pending messages while blocking
 * them to send any additional messages. The process of flushing acquiesces the group so that state
 * transfer or a join can be done. It is also called stop-the-world model as nobody will be able to
 * send messages while a flush is in process.
 * 
 * 

* Flush is needed for: *

* (1) State transfer. When a member requests state transfer, the coordinator tells everyone to stop * sending messages and waits for everyone's ack. Then it asks the application for its state and * ships it back to the requester. After the requester has received and set the state successfully, * the coordinator tells everyone to resume sending messages. *

* (2) View changes (e.g.a join). Before installing a new view V2, flushing would ensure that all * messages *sent* in the current view V1 are indeed *delivered* in V1, rather than in V2 (in all * non-faulty members). This is essentially Virtual Synchrony. * * * * @author Vladimir Blagojevic * @since 2.4 */ @MBean(description = "Flushes the cluster") public class FLUSH extends Protocol { private static final FlushStartResult SUCCESS_START_FLUSH = new FlushStartResult(Boolean.TRUE,null); /* * ------------------------------------------ Properties------------------------------------------ */ @Property(description = "Max time to keep channel blocked in flush. Default is 8000 msec",type=AttributeType.TIME) protected long timeout = 8000; @Property(description = "Timeout (per atttempt) to quiet the cluster during the first flush phase. " + "Default is 2000 msec",type=AttributeType.TIME) protected long start_flush_timeout = 2000; @Property(description = "Timeout to wait for UNBLOCK after STOP_FLUSH is issued. Default is 2000 msec",type=AttributeType.TIME) protected long end_flush_timeout = 2000; @Property(description = "Retry timeout after an unsuccessful attempt to quiet the cluster (first flush phase)." + "Default is 3000 msec", type=AttributeType.TIME) protected long retry_timeout = 2000; @Property(description = "Reconciliation phase toggle. Default is true") protected boolean enable_reconciliation = true; @Property(description="When set, FLUSH is bypassed, same effect as if FLUSH wasn't in the config at all") protected boolean bypass=false; /* * --------------------------------------------- JMX ---------------------------------------------- */ private long startFlushTime; private long totalTimeInFlush; private int numberOfFlushes; private double averageFlushDuration; /* * --------------------------------------------- Fields------------------------------------------------------ */ @GuardedBy("sharedLock") private View currentView=new View(new ViewId(), new ArrayList<>()); /** * Group member that requested FLUSH. For view installations flush coordinator is the group * coordinator For state transfer flush coordinator is the state requesting member */ @GuardedBy("sharedLock") private Address flushCoordinator; @GuardedBy("sharedLock") private final List

flushMembers=new ArrayList<>(); private final AtomicInteger viewCounter = new AtomicInteger(0); @GuardedBy("sharedLock") private final Map flushCompletedMap=new HashMap<>(); @GuardedBy("sharedLock") private final List
flushNotCompletedMap=new ArrayList<>(); @GuardedBy("sharedLock") private final Set
suspected=new TreeSet<>(); @GuardedBy("sharedLock") private final List
reconcileOks=new ArrayList<>(); private final Object sharedLock = new Object(); private final ReentrantLock blockMutex = new ReentrantLock(); private final Condition notBlockedDown = blockMutex.newCondition(); /** * Indicates if FLUSH.down() is currently blocking threads Condition predicate associated with * blockMutex */ @ManagedAttribute(description="Is message sending currently blocked") @GuardedBy("blockMutex") private volatile boolean isBlockingFlushDown = true; @GuardedBy("sharedLock") private boolean flushCompleted = false; private final Promise flush_promise = new Promise<>(); private final Promise flush_unblock_promise = new Promise<>(); private final AtomicBoolean flushInProgress = new AtomicBoolean(false); private final AtomicBoolean sentBlock = new AtomicBoolean(false); private final AtomicBoolean sentUnblock = new AtomicBoolean(false); public long getStartFlushTimeout() { return start_flush_timeout; } public FLUSH setStartFlushTimeout(long start_flush_timeout) { this.start_flush_timeout = start_flush_timeout; return this; } public long getRetryTimeout() { return retry_timeout; } public FLUSH setRetryTimeout(long retry_timeout) { this.retry_timeout = retry_timeout; return this; } public void start() throws Exception { Map map = new HashMap<>(); map.put("flush_supported", Boolean.TRUE); up_prot.up(new Event(Event.CONFIG, map)); down_prot.down(new Event(Event.CONFIG, map)); viewCounter.set(0); blockMutex.lock(); try { isBlockingFlushDown = true; } finally { blockMutex.unlock(); } } public void stop() { synchronized (sharedLock) { currentView = new View(new ViewId(), new ArrayList<>()); flushCompletedMap.clear(); flushNotCompletedMap.clear(); flushMembers.clear(); suspected.clear(); flushCoordinator = null; } } /* -------------------JMX attributes and operations --------------------- */ @ManagedAttribute(type=AttributeType.TIME) public double getAverageFlushDuration() { return averageFlushDuration; } @ManagedAttribute(type=AttributeType.TIME) public long getTotalTimeInFlush() { return totalTimeInFlush; } @ManagedAttribute public int getNumberOfFlushes() { return numberOfFlushes; } @ManagedOperation(description="Sets the bypass flag") public boolean setBypass(boolean flag) { boolean ret=bypass; bypass=flag; return ret; } @ManagedOperation(description = "Request cluster flush") public void startFlush() { startFlush(new Event(Event.SUSPEND)); } private void startFlush(Event evt) { List
flushParticipants =evt.getArg(); startFlush(flushParticipants); } private void startFlush(List
flushParticipants) { if (!flushInProgress.get()) { flush_promise.reset(); synchronized(sharedLock) { if(flushParticipants == null) flushParticipants=new ArrayList<>(currentView.getMembers()); } onSuspend(flushParticipants); try { FlushStartResult r = flush_promise.getResultWithTimeout(start_flush_timeout); if(r.failed()) throw new RuntimeException(r.getFailureCause()); } catch (TimeoutException e) { Set
missingMembers; synchronized(sharedLock) { missingMembers=new HashSet<>(flushMembers); missingMembers.removeAll(flushCompletedMap.keySet()); } rejectFlush(flushParticipants, currentViewId()); throw new RuntimeException(local_addr + " timed out waiting for flush responses from " + missingMembers + " after " + start_flush_timeout + " ms. Rejected flush to participants " + flushParticipants, e); } } else { throw new RuntimeException("Flush attempt is in progress"); } } @ManagedOperation(description = "Request end of flush in a cluster") public void stopFlush() { down(new Event(Event.RESUME)); } /* * ------------------- end JMX attributes and operations --------------------- */ public Object down(Event evt) { if(!bypass){ switch (evt.getType()) { case Event.CONNECT: case Event.CONNECT_USE_FLUSH: return handleConnect(evt,true); case Event.CONNECT_WITH_STATE_TRANSFER: case Event.CONNECT_WITH_STATE_TRANSFER_USE_FLUSH: return handleConnect(evt, false); case Event.SUSPEND: startFlush(evt); return null; // only for testing, see FLUSH#testFlushWithCrashedFlushCoordinator case Event.SUSPEND_BUT_FAIL: if (!flushInProgress.get()) { flush_promise.reset(); ArrayList
flushParticipants = null; synchronized (sharedLock) { flushParticipants = new ArrayList<>(currentView.getMembers()); } onSuspend(flushParticipants); } break; case Event.RESUME: onResume(evt); return null; } } return down_prot.down(evt); } public Object down(Message msg) { if(!bypass) { Address dest = msg.getDest(); if (dest == null) { // mcasts FlushHeader fh =msg.getHeader(this.id); if (fh != null && fh.type == FlushHeader.FLUSH_BYPASS) { return down_prot.down(msg); } else { blockMessageDuringFlush(); } } else { // unicasts are irrelevant in virtual synchrony, let them through return down_prot.down(msg); } } return down_prot.down(msg); } private Object handleConnect(Event evt, boolean waitForUnblock) { if (sentBlock.compareAndSet(false, true)) { sendBlockUpToChannel(); } Object result = down_prot.down(evt); if (result instanceof Throwable) { // set the var back to its original state if we cannot // connect successfully sentBlock.set(false); } if(waitForUnblock) waitForUnblock(); return result; } private void blockMessageDuringFlush() { boolean shouldSuspendByItself = false; blockMutex.lock(); try { while (isBlockingFlushDown) { if (log.isDebugEnabled()) log.debug(local_addr + ": blocking for " + (timeout <= 0 ? "ever" : timeout + "ms")); if(timeout <= 0) { notBlockedDown.await(); } else { shouldSuspendByItself = !notBlockedDown.await(timeout, TimeUnit.MILLISECONDS); } if (shouldSuspendByItself) { isBlockingFlushDown = false; log.warn(local_addr + ": unblocking after " + timeout + "ms"); flush_promise.setResult(new FlushStartResult(Boolean.TRUE,null)); notBlockedDown.signalAll(); } } } catch (InterruptedException e) { Thread.currentThread().interrupt(); } finally { blockMutex.unlock(); } } public Object up(Event evt) { if(!bypass) { switch (evt.getType()) { case Event.VIEW_CHANGE: // JGRP-618: FLUSH coordinator transfer reorders block/unblock/view events in applications (TCP stack only) up_prot.up(evt); View newView =evt.getArg(); boolean coordinatorLeft = onViewChange(newView); boolean singletonMember = newView.size() == 1 && newView.containsMember(local_addr); boolean isThisOurFirstView = viewCounter.addAndGet(1) == 1; // if this is channel's first view and its the only member of the group - no flush // was run but the channel application should still receive BLOCK,VIEW,UNBLOCK // also if coordinator of flush left each member should run stopFlush individually. if ((isThisOurFirstView && singletonMember) || coordinatorLeft) onStopFlush(); return null; case Event.TMP_VIEW: View tmpView =evt.getArg(); if (!tmpView.containsMember(local_addr)) onViewChange(tmpView); break; case Event.SUSPECT: Collection
suspects=evt.arg() instanceof Address? Collections.singletonList(evt.arg()) : evt.arg(); onSuspect(suspects); break; case Event.SUSPEND: startFlush(evt); return null; case Event.RESUME: onResume(evt); return null; case Event.UNBLOCK: flush_unblock_promise.setResult(Boolean.TRUE); break; } } return up_prot.up(evt); } public Object up(Message msg) { if(!bypass) { final FlushHeader fh =msg.getHeader(this.id); if (fh != null) { final Tuple,Digest> tuple=readParticipantsAndDigest(msg.getArray(), msg.getOffset(), msg.getLength()); switch (fh.type) { case FlushHeader.FLUSH_BYPASS: return up_prot.up(msg); case FlushHeader.START_FLUSH: Collection fp = tuple.getVal1(); boolean amIParticipant = (fp != null && fp.contains(local_addr)) || msg.getSrc().equals(local_addr); if (amIParticipant) { handleStartFlush(msg, fh); } else { if (log.isDebugEnabled()) log.debug(local_addr + ": received START_FLUSH but I'm not flush participant, not responding"); } break; case FlushHeader.FLUSH_RECONCILE: handleFlushReconcile(msg); break; case FlushHeader.FLUSH_RECONCILE_OK: onFlushReconcileOK(msg); break; case FlushHeader.STOP_FLUSH: onStopFlush(); break; case FlushHeader.ABORT_FLUSH: Collection flushParticipants = tuple.getVal1(); boolean participant = flushParticipants != null && flushParticipants.contains(local_addr); if (log.isDebugEnabled()) { log.debug(local_addr + ": received ABORT_FLUSH from flush coordinator " + msg.getSrc() + ", am I flush participant=" + participant); } if (participant) resetForNextFlush(); break; case FlushHeader.FLUSH_NOT_COMPLETED: if (log.isDebugEnabled()) { log.debug(local_addr + ": received FLUSH_NOT_COMPLETED from " + msg.getSrc()); } boolean flushCollision = false; synchronized (sharedLock) { flushNotCompletedMap.add(msg.getSrc()); flushCollision = !flushCompletedMap.isEmpty(); if (flushCollision) { flushNotCompletedMap.clear(); flushCompletedMap.clear(); } } if (log.isDebugEnabled()) log.debug(local_addr + ": received FLUSH_NOT_COMPLETED from " + msg.getSrc() + " collision=" + flushCollision); // reject flush if we have at least one OK and at least one FAIL if (flushCollision) { Runnable r =() -> rejectFlush(tuple.getVal1(), fh.viewID); new Thread(r).start(); } // however, flush should fail/retry as soon as one FAIL is received flush_promise.setResult(new FlushStartResult(Boolean.FALSE, new Exception("Flush failed for " + msg.getSrc()))); break; case FlushHeader.FLUSH_COMPLETED: if (isCurrentFlushMessage(fh)) onFlushCompleted(msg.getSrc(), msg, fh); break; } return null; // do not pass FLUSH msg up } else { // https://issues.redhat.com/browse/JGRP-575: for processing of application messages after we join, // lets wait for STOP_FLUSH to complete before we start allowing message up if (msg.getDest() != null) return up_prot.up(msg); // allow unicasts to pass, virtual synchrony only applies to multicasts } } return up_prot.up(msg); } public void up(MessageBatch batch) { if(bypass) { up_prot.up(batch); return; } Iterator it=batch.iterator(); while(it.hasNext()) { Message msg=it.next(); if(msg.getHeader(id) != null) { it.remove(); up(msg); // let the existing code handle this } else { if(msg.getDest() != null) { // skip unicast messages, process them right away it.remove(); up_prot.up(msg); } } } if(!batch.isEmpty()) up_prot.up(batch); } private void waitForUnblock() { try { flush_unblock_promise.getResultWithTimeout(end_flush_timeout); } catch (TimeoutException t) { if (log.isWarnEnabled()) log.warn(local_addr + ": waiting for UNBLOCK timed out after " + end_flush_timeout + " ms"); } finally { flush_unblock_promise.reset(); } } private void onFlushReconcileOK(Message msg) { if (log.isDebugEnabled()) log.debug(local_addr + ": received reconcile ok from " + msg.getSrc()); synchronized (sharedLock) { reconcileOks.add(msg.getSrc()); if (reconcileOks.size() >= flushMembers.size()) { flush_promise.setResult(SUCCESS_START_FLUSH); if (log.isDebugEnabled()) log.debug(local_addr + ": all FLUSH_RECONCILE_OK received"); } } } private void handleFlushReconcile(Message msg) { Address requester = msg.getSrc(); Tuple,Digest> tuple=readParticipantsAndDigest(msg.getArray(), msg.getOffset(),msg.getLength()); Digest reconcileDigest = tuple.getVal2(); if (log.isDebugEnabled()) log.debug(local_addr + ": received FLUSH_RECONCILE, passing digest to NAKACK " + reconcileDigest); // Let NAKACK reconcile missing messages down_prot.down(new Event(Event.REBROADCAST, reconcileDigest)); if (log.isDebugEnabled()) log.debug(local_addr + ": returned from FLUSH_RECONCILE, " + " sending RECONCILE_OK to " + requester); Message reconcileOk = new EmptyMessage(requester).setFlag(Message.Flag.OOB) .putHeader(this.id,new FlushHeader(FlushHeader.FLUSH_RECONCILE_OK)); down_prot.down(reconcileOk); } private void handleStartFlush(Message msg, FlushHeader fh) { Address flushRequester = msg.getSrc(); boolean proceed = flushInProgress.compareAndSet(false, true); if (proceed) { synchronized (sharedLock) { flushCoordinator = flushRequester; } onStartFlush(flushRequester, msg, fh); } else { Tuple,Digest> tuple=readParticipantsAndDigest(msg.getArray(), msg.getOffset(),msg.getLength()); Collection flushParticipants=tuple.getVal1(); Message response = new BytesMessage(flushRequester) .putHeader(this.id,new FlushHeader(FlushHeader.FLUSH_NOT_COMPLETED,fh.viewID)) .setArray(marshal(flushParticipants, null)); down_prot.down(response); if (log.isDebugEnabled()) log.debug(local_addr + ": received START_FLUSH, responded with FLUSH_NOT_COMPLETED to " + flushRequester); } } private void rejectFlush(Collection participants, long viewId) { if(participants == null) return; for (Address flushMember : participants) { if(flushMember == null) continue; Message reject = new BytesMessage(flushMember).setSrc(local_addr).setFlag(Message.Flag.OOB) .putHeader(this.id, new FlushHeader(FlushHeader.ABORT_FLUSH, viewId)) .setArray(marshal(participants, null)); down_prot.down(reject); } } public List providedDownServices() { List retval=new ArrayList<>(2); retval.add(Event.SUSPEND); retval.add(Event.RESUME); return retval; } private void sendBlockUpToChannel() { this.up(new Event(Event.BLOCK)); sentUnblock.set(false); } private void sendUnBlockUpToChannel() { sentBlock.set(false); this.up(new Event(Event.UNBLOCK)); } private boolean isCurrentFlushMessage(FlushHeader fh) { return fh.viewID == currentViewId(); } private long currentViewId() { long viewId = -1; synchronized (sharedLock) { ViewId view = currentView.getViewId(); if (view != null) { viewId = view.getId(); } } return viewId; } private boolean onViewChange(View view) { boolean coordinatorLeft = false; View oldView; synchronized (sharedLock) { suspected.retainAll(view.getMembers()); oldView = currentView; currentView = view; coordinatorLeft = !oldView.getMembers().isEmpty() && !view.getMembers().isEmpty() && !view.containsMember(oldView.getCreator()); } if (log.isDebugEnabled()) log.debug(local_addr + ": installing view " + view); return coordinatorLeft; } private void onStopFlush() { if (stats && startFlushTime > 0) { long stopFlushTime = System.currentTimeMillis(); totalTimeInFlush += (stopFlushTime - startFlushTime); if (numberOfFlushes > 0) { averageFlushDuration = totalTimeInFlush / (double) numberOfFlushes; } startFlushTime = 0; } if (log.isDebugEnabled()) log.debug(local_addr + ": received STOP_FLUSH, unblocking FLUSH.down() and sending UNBLOCK up"); resetForNextFlush(); if (sentUnblock.compareAndSet(false, true)) { // ensures that we do not repeat unblock event sendUnBlockUpToChannel(); } } private void resetForNextFlush() { synchronized (sharedLock) { flushCompletedMap.clear(); flushNotCompletedMap.clear(); flushMembers.clear(); suspected.clear(); flushCoordinator = null; flushCompleted = false; } blockMutex.lock(); try { isBlockingFlushDown = false; notBlockedDown.signalAll(); } finally { blockMutex.unlock(); } flushInProgress.set(false); } /** * Starts the flush protocol * @param members List of participants in the flush protocol. Guaranteed to be non-null */ private void onSuspend(final List
members) { Collection
participantsInFlush = null; synchronized (sharedLock) { flushCoordinator =local_addr; // start FLUSH only on group members that we need to flush participantsInFlush = members; participantsInFlush.retainAll(currentView.getMembers()); flushMembers.clear(); flushMembers.addAll(participantsInFlush); flushMembers.removeAll(suspected); } if (participantsInFlush.isEmpty()) { flush_promise.setResult(SUCCESS_START_FLUSH); } else { Message msg=new BytesMessage(null).setSrc(local_addr).setArray(marshal(participantsInFlush, null)) .putHeader(this.id, new FlushHeader(FlushHeader.START_FLUSH, currentViewId())); down_prot.down(msg); log.debug("%s: flush coordinator is starting FLUSH with participants %s", local_addr, participantsInFlush); } } private void onResume(Event evt) { List
members =evt.getArg(); long viewID = currentViewId(); boolean isParticipant = false; synchronized(sharedLock) { isParticipant = flushMembers.contains(local_addr) || (members != null && members.contains(local_addr)); } if (members == null || members.isEmpty()) { Message msg = new EmptyMessage(null).setSrc(local_addr); // Cannot be OOB since START_FLUSH is not OOB // we have to FIFO order two subsequent flushes log.debug(local_addr + ": received RESUME, sending STOP_FLUSH to all"); msg.putHeader(this.id, new FlushHeader(FlushHeader.STOP_FLUSH, viewID)); down_prot.down(msg); } else { for (Address address : members) { Message msg = new EmptyMessage(address).setSrc(local_addr); // Cannot be OOB since START_FLUSH is not OOB // we have to FIFO order two subsequent flushes log.debug(local_addr + ": received RESUME, sending STOP_FLUSH to " + address); msg.putHeader(this.id, new FlushHeader(FlushHeader.STOP_FLUSH, viewID)); down_prot.down(msg); } } if(isParticipant) waitForUnblock(); } private void onStartFlush(Address flushStarter, Message msg, FlushHeader fh) { if (stats) { startFlushTime = System.currentTimeMillis(); numberOfFlushes += 1; } boolean proceed = false; boolean amIFlushInitiator = false; Tuple,Digest> tuple=readParticipantsAndDigest(msg.getArray(), msg.getOffset(),msg.getLength()); synchronized (sharedLock) { amIFlushInitiator = flushStarter.equals(local_addr); if(!amIFlushInitiator){ flushCoordinator = flushStarter; flushMembers.clear(); if (tuple.getVal1() != null) { flushMembers.addAll(tuple.getVal1()); } flushMembers.removeAll(suspected); } proceed = flushMembers.contains(local_addr); } if (proceed) { if (sentBlock.compareAndSet(false, true)) { // ensures that we do not repeat block event // and that we do not send block event to non participants sendBlockUpToChannel(); blockMutex.lock(); try { isBlockingFlushDown = true; } finally { blockMutex.unlock(); } } else { if (log.isDebugEnabled()) log.debug(local_addr + ": received START_FLUSH, but not sending BLOCK up"); } Digest digest = (Digest) down_prot.down(Event.GET_DIGEST_EVT); Message start_msg = new BytesMessage(flushStarter) .putHeader(this.id, new FlushHeader(FlushHeader.FLUSH_COMPLETED, fh.viewID)) .setArray(marshal(tuple.getVal1(), digest)); down_prot.down(start_msg); log.debug(local_addr + ": received START_FLUSH, responded with FLUSH_COMPLETED to " + flushStarter); } } private void onFlushCompleted(Address address, final Message m, final FlushHeader header) { Message msg = null; boolean needsReconciliationPhase = false; boolean collision = false; final Tuple,Digest> tuple=readParticipantsAndDigest(m.getArray(), m.getOffset(),m.getLength()); Digest digest = tuple.getVal2(); synchronized (sharedLock) { flushCompletedMap.put(address, digest); flushCompleted = flushCompletedMap.size() >= flushMembers.size() && !flushMembers.isEmpty() && flushCompletedMap.keySet().containsAll(flushMembers); collision = !flushNotCompletedMap.isEmpty(); if (log.isDebugEnabled()) log.debug(local_addr + ": FLUSH_COMPLETED from " + address + ", completed " + flushCompleted + ", flushMembers " + flushMembers + ", flushCompleted " + flushCompletedMap.keySet()); needsReconciliationPhase = enable_reconciliation && flushCompleted && hasVirtualSynchronyGaps(); if (needsReconciliationPhase) { Digest d = findHighestSequences(currentView); reconcileOks.clear(); msg = new BytesMessage(null, marshal(flushMembers, d)).setFlag(Message.Flag.OOB) .putHeader(this.id, new FlushHeader(FlushHeader.FLUSH_RECONCILE, currentViewId())); log.debug("%s : reconciling flush mebers due to virtual synchrony gap, digest is %s, " + "flush members are %s", local_addr, d, flushMembers); flushCompletedMap.clear(); } else if (flushCompleted) { flushCompletedMap.clear(); } else if (collision) { flushNotCompletedMap.clear(); flushCompletedMap.clear(); } } if (needsReconciliationPhase) { down_prot.down(msg); } else if (flushCompleted) { flush_promise.setResult(SUCCESS_START_FLUSH); if (log.isDebugEnabled()) log.debug(local_addr + ": all FLUSH_COMPLETED received"); } else if (collision) { // reject flush if we have at least one OK and at least one FAIL Runnable r =() -> rejectFlush(tuple.getVal1(), header.viewID); new Thread(r).start(); } } private boolean hasVirtualSynchronyGaps() { ArrayList digests=new ArrayList<>(flushCompletedMap.values()); return !same(digests); } protected static boolean same(final List digests) { if(digests == null) return false; Digest first=digests.get(0); for(int i=1; i < digests.size(); i++) { Digest current=digests.get(i); if(!first.equals(current)) return false; } return true; } private Digest findHighestSequences(View view) { List digests = new ArrayList<>(flushCompletedMap.values()); return maxSeqnos(view,digests); } /** Returns a digest which contains, for all members of view, the highest delivered and received * seqno of all digests */ protected static Digest maxSeqnos(final View view, List digests) { if(view == null || digests == null) return null; MutableDigest digest=new MutableDigest(view.getMembersRaw()); digests.forEach(digest::merge); return digest; } private void onSuspect(Collection
addresses) { // handles FlushTest#testFlushWithCrashedFlushCoordinator boolean amINeighbourOfCrashedFlushCoordinator = false; ArrayList
flushMembersCopy = null; synchronized (sharedLock) { boolean flushCoordinatorSuspected=addresses != null && addresses.contains(flushCoordinator); if (flushCoordinatorSuspected) { int indexOfCoordinator = flushMembers.indexOf(flushCoordinator); int myIndex = flushMembers.indexOf(local_addr); int diff = myIndex - indexOfCoordinator; amINeighbourOfCrashedFlushCoordinator = (diff == 1 || (myIndex == 0 && indexOfCoordinator == flushMembers.size())); if (amINeighbourOfCrashedFlushCoordinator) { flushMembersCopy = new ArrayList<>(flushMembers); } } } if (amINeighbourOfCrashedFlushCoordinator) { if (log.isDebugEnabled()) log.debug(local_addr + ": flush coordinator " + flushCoordinator + " suspected," + " I am the neighbor, completing the flush "); onResume(new Event(Event.RESUME, flushMembersCopy)); } // handles FlushTest#testFlushWithCrashedNonCoordinators boolean flushOkCompleted = false; Message m = null; long viewID = 0; synchronized (sharedLock) { suspected.addAll(addresses); flushMembers.removeAll(suspected); viewID = currentViewId(); flushOkCompleted = !flushCompletedMap.isEmpty() && flushCompletedMap.keySet().containsAll(flushMembers); if (flushOkCompleted) { m = new BytesMessage(flushCoordinator).setSrc(local_addr); } log.debug(local_addr + ": suspects: " + addresses + ", completed " + flushOkCompleted + ", flushOkSet " + flushCompletedMap + ", flushMembers " + flushMembers); } if (flushOkCompleted) { Digest digest = (Digest) down_prot.down(Event.GET_DIGEST_EVT); m.putHeader(this.id, new FlushHeader(FlushHeader.FLUSH_COMPLETED, viewID)).setArray(marshal(null, digest)); down_prot.down(m); if (log.isDebugEnabled()) log.debug(local_addr + ": sent FLUSH_COMPLETED message to " + flushCoordinator); } } protected static ByteArray marshal(final Collection participants, final Digest digest) { final ByteArrayDataOutputStream out=new ByteArrayDataOutputStream(512); try { Util.writeAddresses(participants, out); Util.writeStreamable(digest,out); return out.getBuffer(); } catch(Exception ex) { return null; } } protected Tuple,Digest> readParticipantsAndDigest(byte[] buffer, int offset, int length) { if(buffer == null) return null; try { DataInput in=new ByteArrayDataInputStream(buffer, offset, length); Collection
participants=Util.readAddresses(in, ArrayList::new); Digest digest=Util.readStreamable(Digest::new, in); return new Tuple<>(participants, digest); } catch(Exception ex) { log.error("%s: failed reading particpants and digest from message: %s", local_addr, ex); return null; } } private static final class FlushStartResult { private final Boolean result; private final Exception failureCause; private FlushStartResult(Boolean result, Exception failureCause) { this.result = result; this.failureCause = failureCause; } public Boolean getResult() { return result; } public boolean failed(){ return result == Boolean.FALSE; } public Exception getFailureCause() { return failureCause; } } public static class FlushHeader extends Header { public static final byte START_FLUSH = 0; public static final byte STOP_FLUSH = 2; public static final byte FLUSH_COMPLETED = 3; public static final byte ABORT_FLUSH = 5; public static final byte FLUSH_BYPASS = 6; public static final byte FLUSH_RECONCILE = 7; public static final byte FLUSH_RECONCILE_OK = 8; public static final byte FLUSH_NOT_COMPLETED = 9; protected byte type; protected long viewID; public FlushHeader() { this(START_FLUSH, 0); } public FlushHeader(byte type) { this.type=type; } public FlushHeader(byte type, long viewID) { this(type); this.viewID=viewID; } public short getMagicId() {return 64;} public Supplier create() {return FlushHeader::new;} public byte getType() {return type;} public long getViewID() {return viewID;} @Override public int serializedSize() { return Global.BYTE_SIZE + Global.LONG_SIZE; // type and viewId } public String toString() { switch (type) { case START_FLUSH: return "FLUSH[type=START_FLUSH,viewId=" + viewID; case STOP_FLUSH: return "FLUSH[type=STOP_FLUSH,viewId=" + viewID + "]"; case ABORT_FLUSH: return "FLUSH[type=ABORT_FLUSH,viewId=" + viewID + "]"; case FLUSH_COMPLETED: return "FLUSH[type=FLUSH_COMPLETED,viewId=" + viewID + "]"; case FLUSH_BYPASS: return "FLUSH[type=FLUSH_BYPASS,viewId=" + viewID + "]"; case FLUSH_RECONCILE: return "FLUSH[type=FLUSH_RECONCILE,viewId=" + viewID; case FLUSH_RECONCILE_OK: return "FLUSH[type=FLUSH_RECONCILE_OK,viewId=" + viewID + "]"; default: return "[FLUSH: unknown type (" + type + ")]"; } } @Override public void writeTo(DataOutput out) throws IOException { out.writeByte(type); out.writeLong(viewID); } @Override public void readFrom(DataInput in) throws IOException { type = in.readByte(); viewID = in.readLong(); } } }




© 2015 - 2024 Weber Informatics LLC | Privacy Policy