/**
* Waits for renting partitions.
*
* @return {@code True} if mapping was changed.
* @throws IgniteCheckedException If failed.
*/
private boolean waitForRent() throws IgniteCheckedException {
boolean changed = false;
// Synchronously wait for all renting partitions to complete.
for (Iterator<GridDhtLocalPartition> it = locParts.values().iterator(); it.hasNext(); ) {
GridDhtLocalPartition p = it.next();
GridDhtPartitionState state = p.state();
if (state == RENTING || state == EVICTED) {
if (log.isDebugEnabled()) log.debug("Waiting for renting partition: " + p);
// Wait for partition to empty out.
p.rent(true).get();
if (log.isDebugEnabled()) log.debug("Finished waiting for renting partition: " + p);
// Remove evicted partition.
it.remove();
changed = true;
}
}
return changed;
}
public void remove() {
if (lastRet < 0) throw new IllegalStateException();
Object x = array[lastRet];
lastRet = -1;
// Traverse underlying queue to find == element,
// not just a .equals element.
lock.lock();
try {
for (Iterator it = q.iterator(); it.hasNext(); ) {
if (it.next() == x) {
it.remove();
return;
}
}
} finally {
lock.unlock();
}
}
/** starts the server listens to incoming packets */
public void run() {
try {
while (true) {
if (Log.loggingNet) Log.net("In RDPServer.run: starting new iteration");
try {
Set<DatagramChannel> activeChannels = getActiveChannels();
activeChannelCalls++;
Iterator<DatagramChannel> iter = activeChannels.iterator();
while (iter.hasNext()) {
DatagramChannel dc = iter.next();
if (Log.loggingNet) Log.net("In RDPServer.run: about to call processActiveChannel");
processActiveChannel(dc);
if (Log.loggingNet) Log.net("In RDPServer.run: returned from processActiveChannel");
}
} catch (ClosedChannelException ex) {
// ignore
} catch (Exception e) {
Log.exception("RDPServer.run caught exception", e);
}
}
} finally {
Log.warn("RDPServer.run: thread exiting");
}
}
/** Clears values for this partition. */
private void clearAll() {
GridCacheVersion clearVer = cctx.versions().next();
boolean swap = cctx.isSwapOrOffheapEnabled();
boolean rec = cctx.events().isRecordable(EVT_CACHE_REBALANCE_OBJECT_UNLOADED);
Iterator<GridDhtCacheEntry> it = map.values().iterator();
GridCloseableIterator<Map.Entry<byte[], GridCacheSwapEntry>> swapIt = null;
if (swap
&& GridQueryProcessor.isEnabled(cctx.config())) { // Indexing needs to unswap cache values.
Iterator<GridDhtCacheEntry> unswapIt = null;
try {
swapIt = cctx.swap().iterator(id);
unswapIt = unswapIterator(swapIt);
} catch (Exception e) {
U.error(log, "Failed to clear swap for evicted partition: " + this, e);
}
if (unswapIt != null) it = F.concat(it, unswapIt);
}
try {
while (it.hasNext()) {
GridDhtCacheEntry cached = it.next();
try {
if (cached.clearInternal(clearVer, swap)) {
map.remove(cached.key(), cached);
if (!cached.isInternal()) {
mapPubSize.decrement();
if (rec)
cctx.events()
.addEvent(
cached.partition(),
cached.key(),
cctx.localNodeId(),
(IgniteUuid) null,
null,
EVT_CACHE_REBALANCE_OBJECT_UNLOADED,
null,
false,
cached.rawGet(),
cached.hasValue(),
null,
null,
null);
}
}
} catch (IgniteCheckedException e) {
U.error(log, "Failed to clear cache entry for evicted partition: " + cached, e);
}
}
} finally {
U.close(swapIt, log);
}
}
/**
* Creates an <tt>ArrayBlockingQueue</tt> with the given (fixed) capacity, the specified access
* policy and initially containing the elements of the given collection, added in traversal order
* of the collection's iterator.
*
* @param capacity the capacity of this queue
* @param fair if <tt>true</tt> then queue accesses for threads blocked on insertion or removal,
* are processed in FIFO order; if <tt>false</tt> the access order is unspecified.
* @param c the collection of elements to initially contain
* @throws IllegalArgumentException if <tt>capacity</tt> is less than <tt>c.size()</tt>, or less
* than 1.
* @throws NullPointerException if the specified collection or any of its elements are null
*/
public ArrayBlockingQueue(int capacity, boolean fair, Collection<? extends E> c) {
this(capacity, fair);
if (capacity < c.size()) throw new IllegalArgumentException();
for (Iterator<? extends E> it = c.iterator(); it.hasNext(); ) add(it.next());
}
/**
* we have a packet that belongs to the passed in connection. process the packet for the
* connection. It returns true if the connection is open and the packet was a data packet
*/
boolean processExistingConnection(RDPConnection con, RDPPacket packet) {
if (Log.loggingNet)
Log.net("RDPServer.processExistingConnection: con state=" + con + ", packet=" + packet);
packetCounter.add();
int state = con.getState();
if (state == RDPConnection.LISTEN) {
// something is wrong, we shouldn't be here
// we get to this method after looking in the connections map
// but all LISTEN connections should be listed direct
// from serversockets
Log.error("RDPServer.processExistingConnection: connection shouldnt be in LISTEN state");
return false;
}
if (state == RDPConnection.SYN_SENT) {
if (!packet.isAck()) {
Log.warn("got a non-ack packet when we're in SYN_SENT");
return false;
}
if (!packet.isSyn()) {
Log.warn("got a non-syn packet when we're in SYN_SENT");
return false;
}
if (Log.loggingNet) Log.net("good: got syn-ack packet in syn_sent");
// make sure its acking our initial segment #
if (packet.getAckNum() != con.getInitialSendSeqNum()) {
if (Log.loggingNet) Log.net("syn's ack number does not match initial seq #");
return false;
}
con.setRcvCur(packet.getSeqNum());
con.setRcvIrs(packet.getSeqNum());
con.setMaxSendUnacks(packet.getSendUnacks());
con.setMaxReceiveSegmentSize(packet.getMaxRcvSegmentSize());
con.setSendUnackd(packet.getAckNum() + 1);
// ack first before setting state to open
// otherwise some other thread will get woken up and send data
// before we send the ack
if (Log.loggingNet) Log.net("new connection state: " + con);
RDPPacket replyPacket = new RDPPacket(con);
con.sendPacketImmediate(replyPacket, false);
con.setState(RDPConnection.OPEN);
return false;
}
if (state == RDPConnection.SYN_RCVD) {
if (packet.getSeqNum() <= con.getRcvIrs()) {
Log.error("seqnum is not above rcv initial seq num");
return false;
}
if (packet.getSeqNum() > (con.getRcvCur() + (con.getRcvMax() * 2))) {
Log.error("seqnum is too big");
return false;
}
if (packet.isAck()) {
if (packet.getAckNum() == con.getInitialSendSeqNum()) {
if (Log.loggingNet) Log.net("got ack for our syn - setting state to open");
con.setState(RDPConnection.OPEN); // this will notify()
// call the accept callback
// first find the serversocket
DatagramChannel dc = con.getDatagramChannel();
if (dc == null) {
throw new MVRuntimeException(
"RDPServer.processExistingConnection: no datagramchannel for connection that just turned OPEN");
}
RDPServerSocket rdpSocket = RDPServer.getRDPSocket(dc);
if (rdpSocket == null) {
throw new MVRuntimeException(
"RDPServer.processExistingConnection: no socket for connection that just turned OPEN");
}
ClientConnection.AcceptCallback acceptCB = rdpSocket.getAcceptCallback();
if (acceptCB != null) {
acceptCB.acceptConnection(con);
} else {
Log.warn("serversocket has no accept callback");
}
if (Log.loggingNet)
Log.net(
"RDPServer.processExistingConnection: got ACK, removing from unack list: "
+ packet.getSeqNum());
con.removeUnackPacket(packet.getSeqNum());
}
}
}
if (state == RDPConnection.CLOSE_WAIT) {
// reply with a reset on all packets
if (!packet.isRst()) {
RDPPacket rstPacket = RDPPacket.makeRstPacket();
con.sendPacketImmediate(rstPacket, false);
}
}
if (state == RDPConnection.OPEN) {
if (packet.isRst()) {
// the other side wants to close the connection
// set the state,
// dont call con.close() since that will send a reset packet
if (Log.loggingDebug)
Log.debug("RDPServer.processExistingConnection: got reset packet for con " + con);
if (con.getState() != RDPConnection.CLOSE_WAIT) {
con.setState(RDPConnection.CLOSE_WAIT);
con.setCloseWaitTimer();
// Only invoke callback when moving into CLOSE_WAIT
// state. This prevents two calls to connectionReset.
Log.net("RDPServer.processExistingConnection: calling reset callback");
ClientConnection.MessageCallback pcb = con.getCallback();
pcb.connectionReset(con);
}
return false;
}
if (packet.isSyn()) {
// this will close the connection (put into CLOSE_WAIT)
// send a reset packet and call the connectionReset callback
Log.error(
"RDPServer.processExistingConnection: closing connection because we got a syn packet, con="
+ con);
con.close();
return false;
}
// TODO: shouldnt it be ok for it to have same seq num?
// if it is a 0 data packet?
long rcvCur = con.getRcvCur();
if (packet.getSeqNum() <= rcvCur) {
if (Log.loggingNet)
Log.net("RDPServer.processExistingConnection: seqnum too small - acking/not process");
if (packet.getData() != null) {
if (Log.loggingNet)
Log.net(
"RDPServer.processExistingConnection: sending ack even though seqnum out of range");
RDPPacket replyPacket = new RDPPacket(con);
con.sendPacketImmediate(replyPacket, false);
}
return false;
}
if (packet.getSeqNum() > (rcvCur + (con.getRcvMax() * 2))) {
Log.error("RDPServer.processExistingConnection: seqnum too big - discarding");
return false;
}
if (packet.isAck()) {
if (Log.loggingNet)
Log.net("RDPServer.processExistingConnection: processing ack " + packet.getAckNum());
// lock for race condition (read then set)
con.getLock().lock();
try {
if (packet.getAckNum() >= con.getSendNextSeqNum()) {
// acking something we didnt even send yet
Log.error(
"RDPServer.processExistingConnection: discarding -- got ack #"
+ packet.getAckNum()
+ ", but our next send seqnum is "
+ con.getSendNextSeqNum()
+ " -- "
+ con);
return false;
}
if (con.getSendUnackd() <= packet.getAckNum()) {
con.setSendUnackd(packet.getAckNum() + 1);
if (Log.loggingNet)
Log.net(
"RDPServer.processExistingConnection: updated send_unackd num to "
+ con.getSendUnackd()
+ " (one greater than packet ack) - "
+ con);
con.removeUnackPacketUpTo(packet.getAckNum());
}
if (packet.isEak()) {
List eackList = packet.getEackList();
Iterator iter = eackList.iterator();
while (iter.hasNext()) {
Long seqNum = (Long) iter.next();
if (Log.loggingNet)
Log.net("RDPServer.processExistingConnection: got EACK: " + seqNum);
con.removeUnackPacket(seqNum.longValue());
}
}
} finally {
con.getLock().unlock();
if (Log.loggingNet)
Log.net("RDPServer.processExistingConnection: processed ack " + packet.getAckNum());
}
}
// process the data
byte[] data = packet.getData();
if ((data != null) || packet.isNul()) {
dataCounter.add();
// lock - since racecondition: we read then set
con.getLock().lock();
try {
rcvCur = con.getRcvCur(); // update rcvCur
if (Log.loggingNet) Log.net("RDPServer.processExistingConnection: rcvcur is " + rcvCur);
ClientConnection.MessageCallback pcb = con.getCallback();
if (pcb == null) {
Log.warn("RDPServer.processExistingConnection: no packet callback registered");
}
// call callback only if we havent seen it already - eackd
if (!con.hasEack(packet.getSeqNum())) {
if (con.isSequenced()) {
// this is a sequential connection,
// make sure this is the 'next' packet
// is this the next sequential packet
if (packet.getSeqNum() == (rcvCur + 1)) {
// this is the next packet
if (Log.loggingNet)
Log.net(
"RDPServer.processExistingConnection: conn is sequenced and received next packet, rcvCur="
+ rcvCur
+ ", packet="
+ packet);
if ((pcb != null) && (data != null)) {
queueForCallbackProcessing(pcb, con, packet);
}
} else {
// not the next packet, place it in queue
if (Log.loggingNet)
Log.net(
"RDPServer.processExistingConnection: conn is sequenced, BUT PACKET is OUT OF ORDER: rcvcur="
+ rcvCur
+ ", packet="
+ packet);
con.addSequencePacket(packet);
}
} else {
if ((pcb != null) && (data != null)) {
// make sure we havent already processed packet
queueForCallbackProcessing(pcb, con, packet);
}
}
} else {
if (Log.loggingNet) Log.net(con.toString() + " already seen this packet");
}
// is this the next sequential packet
if (packet.getSeqNum() == (rcvCur + 1)) {
con.setRcvCur(rcvCur + 1);
if (Log.loggingNet)
Log.net(
"RDPServer.processExistingConnection RCVD: incremented last sequenced rcvd: "
+ (rcvCur + 1));
// packet in order - dont add to eack
// Take any additional sequential packets off eack
long seqNum = rcvCur + 2;
while (con.removeEack(seqNum)) {
if (Log.loggingNet)
Log.net("RDPServer.processExistingConnection: removing/collapsing eack: " + seqNum);
con.setRcvCur(seqNum++);
}
if (con.isSequenced()) {
rcvCur++; // since we just process the last one
Log.net(
"RDPServer.processExistingConnection: connection is sequenced, processing collapsed packets.");
// send any saved sequential packets also
Iterator iter = con.getSequencePackets().iterator();
while (iter.hasNext()) {
RDPPacket p = (RDPPacket) iter.next();
if (Log.loggingNet)
Log.net(
"rdpserver: stored packet seqnum="
+ p.getSeqNum()
+ ", if equal to (rcvcur + 1)="
+ (rcvCur + 1));
if (p.getSeqNum() == (rcvCur + 1)) {
Log.net(
"RDPServer.processExistingConnection: this is the next packet, processing");
// this is the next packet - update rcvcur
rcvCur++;
// process this packet
Log.net(
"RDPServer.processExistingConnection: processing stored sequential packet "
+ p);
byte[] storedData = p.getData();
if (pcb != null && storedData != null) {
queueForCallbackProcessing(pcb, con, packet);
}
iter.remove();
}
}
} else {
if (Log.loggingNet)
Log.net("RDPServer.processExistingConnection: connection is not sequenced");
}
} else {
if (Log.loggingNet)
Log.net(
"RDPServer.processExistingConnection: RCVD OUT OF ORDER: packet seq#: "
+ packet.getSeqNum()
+ ", but last sequential rcvd packet was: "
+ con.getRcvCur()
+ " -- not incrementing counter");
if (packet.getSeqNum() > rcvCur) {
// must be at least + 2 larger than rcvCur
if (Log.loggingNet) Log.net("adding to eack list " + packet);
con.addEack(packet);
}
}
} finally {
con.getLock().unlock();
}
return true;
}
}
return false;
}
/** returns a list of rdpserversockets */
Set<DatagramChannel> getActiveChannels() throws InterruptedException, java.io.IOException {
lock.lock();
try {
while (channelMap.isEmpty()) {
channelMapNotEmpty.await();
}
} finally {
lock.unlock();
}
Set<SelectionKey> readyKeys = null;
do {
lock.lock();
try {
if (!newChannelSet.isEmpty()) {
if (Log.loggingNet) Log.net("RDPServer.getActiveChannels: newChannelSet is not null");
Iterator<DatagramChannel> iter = newChannelSet.iterator();
while (iter.hasNext()) {
DatagramChannel newDC = iter.next();
iter.remove();
newDC.register(selector, SelectionKey.OP_READ);
}
}
} finally {
lock.unlock();
}
int numReady = selector.select(); // this is a blocking call - thread safe
selectCalls++;
if (numReady == 0) {
if (Log.loggingNet) Log.net("RDPServer.getActiveChannels: selector returned 0");
continue;
}
readyKeys = selector.selectedKeys();
if (Log.loggingNet)
Log.net(
"RDPServer.getActiveChannels: called select - # of ready keys = "
+ readyKeys.size()
+ " == "
+ numReady);
} while (readyKeys == null || readyKeys.isEmpty());
lock.lock();
try {
// get a datagramchannel that is ready
Set<DatagramChannel> activeChannels = new HashSet<DatagramChannel>();
Iterator<SelectionKey> iter = readyKeys.iterator();
while (iter.hasNext()) {
SelectionKey key = iter.next();
if (Log.loggingNet)
Log.net(
"RDPServer.getActiveChannels: matched selectionkey: "
+ key
+ ", isAcceptable="
+ key.isAcceptable()
+ ", isReadable="
+ key.isReadable()
+ ", isValid="
+ key.isValid()
+ ", isWritable="
+ key.isWritable());
iter.remove(); // remove from the selected key list
if (!key.isReadable() || !key.isValid()) {
Log.error(
"RDPServer.getActiveChannels: Throwing exception: RDPServer: not readable or invalid");
throw new MVRuntimeException("RDPServer: not readable or invalid");
}
DatagramChannel dc = (DatagramChannel) key.channel();
activeChannels.add(dc);
}
if (Log.loggingNet)
Log.net(
"RDPServer.getActiveChannels: returning " + activeChannels.size() + " active channels");
return activeChannels;
} finally {
lock.unlock();
}
}
public void run() {
// every second, go through all the packets that havent been
// ack'd
List<RDPConnection> conList = new LinkedList<RDPConnection>();
long lastCounterTime = System.currentTimeMillis();
while (true) {
try {
long startTime = System.currentTimeMillis();
long interval = startTime - lastCounterTime;
if (interval > 1000) {
if (Log.loggingNet) {
Log.net(
"RDPServer counters: activeChannelCalls "
+ activeChannelCalls
+ ", selectCalls "
+ selectCalls
+ ", transmits "
+ transmits
+ ", retransmits "
+ retransmits
+ " in "
+ interval
+ "ms");
}
activeChannelCalls = 0;
selectCalls = 0;
transmits = 0;
retransmits = 0;
lastCounterTime = startTime;
}
if (Log.loggingNet) Log.net("RDPServer.RETRY: startTime=" + startTime);
// go through all the rdpconnections and re-send any
// unacked packets
conList.clear();
lock.lock();
try {
// make a copy since the values() collection is
// backed by the map
Set<RDPConnection> conCol = RDPServer.getAllConnections();
if (conCol == null) {
throw new MVRuntimeException("values() returned null");
}
conList.addAll(conCol); // make non map backed copy
} finally {
lock.unlock();
}
Iterator<RDPConnection> iter = conList.iterator();
while (iter.hasNext()) {
RDPConnection con = iter.next();
long currentTime = System.currentTimeMillis();
// is the connection in CLOSE_WAIT
if (con.getState() == RDPConnection.CLOSE_WAIT) {
long closeTime = con.getCloseWaitTimer();
long elapsedTime = currentTime - closeTime;
Log.net(
"RDPRetryThread: con is in CLOSE_WAIT: elapsed close timer(ms)="
+ elapsedTime
+ ", waiting for 30seconds to elapse. con="
+ con);
if (elapsedTime > 30000) {
// close the connection
Log.net("RDPRetryThread: removing CLOSE_WAIT connection. con=" + con);
removeConnection(con);
} else {
Log.net(
"RDPRetryThread: time left on CLOSE_WAIT timer: "
+ (30000 - (currentTime - closeTime)));
}
// con.close();
continue;
}
if (Log.loggingNet)
Log.net(
"RDPServer.RETRY: resending expired packets "
+ con
+ " - current list size = "
+ con.unackListSize());
// see if we should send a null packet, but only if con is already open
if ((con.getState() == RDPConnection.OPEN)
&& ((currentTime - con.getLastNullPacketTime()) > 30000)) {
con.getLock().lock();
try {
RDPPacket nulPacket = RDPPacket.makeNulPacket();
con.sendPacketImmediate(nulPacket, false);
con.setLastNullPacketTime();
if (Log.loggingNet) Log.net("RDPServer.retry: sent nul packet: " + nulPacket);
} finally {
con.getLock().unlock();
}
} else {
if (Log.loggingNet)
Log.net(
"RDPServer.retry: sending nul packet in "
+ (30000 - (currentTime - con.getLastNullPacketTime())));
}
con.resend(
currentTime - resendTimerMS, // resend cutoff time
currentTime - resendTimeoutMS); // giveup time
}
long endTime = System.currentTimeMillis();
if (Log.loggingNet)
Log.net(
"RDPServer.RETRY: endTime=" + endTime + ", elapse(ms)=" + (endTime - startTime));
Thread.sleep(250);
} catch (Exception e) {
Log.exception("RDPServer.RetryThread.run caught exception", e);
}
}
}
/** {@inheritDoc} */
@SuppressWarnings({"MismatchedQueryAndUpdateOfCollection"})
@Nullable
@Override
public GridDhtPartitionMap update(
@Nullable GridDhtPartitionExchangeId exchId, GridDhtPartitionFullMap partMap) {
if (log.isDebugEnabled())
log.debug(
"Updating full partition map [exchId=" + exchId + ", parts=" + fullMapString() + ']');
assert partMap != null;
lock.writeLock().lock();
try {
if (stopping) return null;
if (exchId != null && lastExchangeId != null && lastExchangeId.compareTo(exchId) >= 0) {
if (log.isDebugEnabled())
log.debug(
"Stale exchange id for full partition map update (will ignore) [lastExchId="
+ lastExchangeId
+ ", exchId="
+ exchId
+ ']');
return null;
}
if (node2part != null && node2part.compareTo(partMap) >= 0) {
if (log.isDebugEnabled())
log.debug(
"Stale partition map for full partition map update (will ignore) [lastExchId="
+ lastExchangeId
+ ", exchId="
+ exchId
+ ", curMap="
+ node2part
+ ", newMap="
+ partMap
+ ']');
return null;
}
long updateSeq = this.updateSeq.incrementAndGet();
if (exchId != null) lastExchangeId = exchId;
if (node2part != null) {
for (GridDhtPartitionMap part : node2part.values()) {
GridDhtPartitionMap newPart = partMap.get(part.nodeId());
// If for some nodes current partition has a newer map,
// then we keep the newer value.
if (newPart != null && newPart.updateSequence() < part.updateSequence()) {
if (log.isDebugEnabled())
log.debug(
"Overriding partition map in full update map [exchId="
+ exchId
+ ", curPart="
+ mapString(part)
+ ", newPart="
+ mapString(newPart)
+ ']');
partMap.put(part.nodeId(), part);
}
}
for (Iterator<UUID> it = partMap.keySet().iterator(); it.hasNext(); ) {
UUID nodeId = it.next();
if (!cctx.discovery().alive(nodeId)) {
if (log.isDebugEnabled())
log.debug(
"Removing left node from full map update [nodeId="
+ nodeId
+ ", partMap="
+ partMap
+ ']');
it.remove();
}
}
}
node2part = partMap;
Map<Integer, Set<UUID>> p2n = new HashMap<>(cctx.affinity().partitions(), 1.0f);
for (Map.Entry<UUID, GridDhtPartitionMap> e : partMap.entrySet()) {
for (Integer p : e.getValue().keySet()) {
Set<UUID> ids = p2n.get(p);
if (ids == null)
// Initialize HashSet to size 3 in anticipation that there won't be
// more than 3 nodes per partitions.
p2n.put(p, ids = U.newHashSet(3));
ids.add(e.getKey());
}
}
part2node = p2n;
boolean changed = checkEvictions(updateSeq);
consistencyCheck();
if (log.isDebugEnabled()) log.debug("Partition map after full update: " + fullMapString());
return changed ? localPartitionMap() : null;
} finally {
lock.writeLock().unlock();
}
}
