/**
* @param nodes Nodes.
* @param id ID.
* @throws IgniteCheckedException If failed.
*/
private void sendAllPartitions(
Collection<? extends ClusterNode> nodes, GridDhtPartitionExchangeId id)
throws IgniteCheckedException {
GridDhtPartitionsFullMessage m =
new GridDhtPartitionsFullMessage(id, lastVer.get(), id.topologyVersion());
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (!cacheCtx.isLocal()) {
AffinityTopologyVersion startTopVer = cacheCtx.startTopologyVersion();
boolean ready = startTopVer == null || startTopVer.compareTo(id.topologyVersion()) <= 0;
if (ready)
m.addFullPartitionsMap(cacheCtx.cacheId(), cacheCtx.topology().partitionMap(true));
}
}
// It is important that client topologies be added after contexts.
for (GridClientPartitionTopology top : cctx.exchange().clientTopologies())
m.addFullPartitionsMap(top.cacheId(), top.partitionMap(true));
if (log.isDebugEnabled())
log.debug(
"Sending full partition map [nodeIds="
+ F.viewReadOnly(nodes, F.node2id())
+ ", exchId="
+ exchId
+ ", msg="
+ m
+ ']');
cctx.io().safeSend(nodes, m, SYSTEM_POOL, null);
}
/** @param nodes Nodes. */
private Queue<ClusterNode> fallbacks(Collection<ClusterNode> nodes) {
Queue<ClusterNode> fallbacks = new LinkedList<>();
ClusterNode node = F.first(F.view(nodes, IS_LOC_NODE));
if (node != null) fallbacks.add(node);
fallbacks.addAll(node != null ? F.view(nodes, F.not(IS_LOC_NODE)) : nodes);
return fallbacks;
}
/**
* @param cacheId Cache ID.
* @return {@code True} if local client has been added.
*/
public boolean isLocalClientAdded(int cacheId) {
if (!F.isEmpty(reqs)) {
for (DynamicCacheChangeRequest req : reqs) {
if (req.start() && F.eq(req.initiatingNodeId(), cctx.localNodeId())) {
if (CU.cacheId(req.cacheName()) == cacheId) return true;
}
}
}
return false;
}
/**
* @param cacheId Cache ID to check.
* @param topVer Topology version.
* @return {@code True} if cache was added during this exchange.
*/
public boolean isCacheAdded(int cacheId, AffinityTopologyVersion topVer) {
if (!F.isEmpty(reqs)) {
for (DynamicCacheChangeRequest req : reqs) {
if (req.start() && !req.clientStartOnly()) {
if (CU.cacheId(req.cacheName()) == cacheId) return true;
}
}
}
GridCacheContext<?, ?> cacheCtx = cctx.cacheContext(cacheId);
return cacheCtx != null && F.eq(cacheCtx.startTopologyVersion(), topVer);
}
/**
* @param cctx Cache context.
* @param prj Projection (optional).
* @return Collection of data nodes in provided projection (if any).
*/
private static Collection<ClusterNode> nodes(
final GridCacheContext<?, ?> cctx,
@Nullable final ClusterGroup prj,
@Nullable final Integer part) {
assert cctx != null;
final AffinityTopologyVersion topVer = cctx.affinity().affinityTopologyVersion();
Collection<ClusterNode> affNodes = CU.affinityNodes(cctx);
if (prj == null && part == null) return affNodes;
final Set<ClusterNode> owners =
part == null
? Collections.<ClusterNode>emptySet()
: new HashSet<>(cctx.topology().owners(part, topVer));
return F.view(
affNodes,
new P1<ClusterNode>() {
@Override
public boolean apply(ClusterNode n) {
return cctx.discovery().cacheAffinityNode(n, cctx.name())
&& (prj == null || prj.node(n.id()) != null)
&& (part == null || owners.contains(n));
}
});
}
/** @return Nodes to execute on. */
private Collection<ClusterNode> nodes() {
CacheMode cacheMode = cctx.config().getCacheMode();
switch (cacheMode) {
case LOCAL:
if (prj != null)
U.warn(
log,
"Ignoring query projection because it's executed over LOCAL cache "
+ "(only local node will be queried): "
+ this);
return Collections.singletonList(cctx.localNode());
case REPLICATED:
if (prj != null || partition() != null) return nodes(cctx, prj, partition());
return cctx.affinityNode()
? Collections.singletonList(cctx.localNode())
: Collections.singletonList(F.rand(nodes(cctx, null, partition())));
case PARTITIONED:
return nodes(cctx, prj, partition());
default:
throw new IllegalStateException("Unknown cache distribution mode: " + cacheMode);
}
}
/**
* Sends query request.
*
* @param fut Distributed future.
* @param req Request.
* @param nodes Nodes.
* @throws IgniteCheckedException In case of error.
*/
@SuppressWarnings("unchecked")
private void sendRequest(
final GridCacheDistributedQueryFuture<?, ?, ?> fut,
final GridCacheQueryRequest req,
Collection<ClusterNode> nodes)
throws IgniteCheckedException {
assert fut != null;
assert req != null;
assert nodes != null;
final UUID locNodeId = cctx.localNodeId();
ClusterNode locNode = null;
Collection<ClusterNode> rmtNodes = null;
for (ClusterNode n : nodes) {
if (n.id().equals(locNodeId)) locNode = n;
else {
if (rmtNodes == null) rmtNodes = new ArrayList<>(nodes.size());
rmtNodes.add(n);
}
}
// Request should be sent to remote nodes before the query is processed on the local node.
// For example, a remote reducer has a state, we should not serialize and then send
// the reducer changed by the local node.
if (!F.isEmpty(rmtNodes)) {
cctx.io()
.safeSend(
rmtNodes,
req,
cctx.ioPolicy(),
new P1<ClusterNode>() {
@Override
public boolean apply(ClusterNode node) {
fut.onNodeLeft(node.id());
return !fut.isDone();
}
});
}
if (locNode != null) {
cctx.closures()
.callLocalSafe(
new Callable<Object>() {
@Override
public Object call() throws Exception {
req.beforeLocalExecution(cctx);
processQueryRequest(locNodeId, req);
return null;
}
});
}
}
/**
* Starts dynamic caches.
*
* @throws IgniteCheckedException If failed.
*/
private void startCaches() throws IgniteCheckedException {
cctx.cache()
.prepareCachesStart(
F.view(
reqs,
new IgnitePredicate<DynamicCacheChangeRequest>() {
@Override
public boolean apply(DynamicCacheChangeRequest req) {
return req.start();
}
}),
exchId.topologyVersion());
}
/**
* @param rmtReducer Optional reducer.
* @param rmtTransform Optional transformer.
* @param args Arguments.
* @return Future.
*/
@SuppressWarnings("IfMayBeConditional")
private <R> CacheQueryFuture<R> execute(
@Nullable IgniteReducer<T, R> rmtReducer,
@Nullable IgniteClosure<T, R> rmtTransform,
@Nullable Object... args) {
Collection<ClusterNode> nodes = nodes();
cctx.checkSecurity(SecurityPermission.CACHE_READ);
if (nodes.isEmpty())
return new GridCacheQueryErrorFuture<>(
cctx.kernalContext(), new ClusterGroupEmptyCheckedException());
if (log.isDebugEnabled())
log.debug("Executing query [query=" + this + ", nodes=" + nodes + ']');
if (cctx.deploymentEnabled()) {
try {
cctx.deploy().registerClasses(filter, rmtReducer, rmtTransform);
cctx.deploy().registerClasses(args);
} catch (IgniteCheckedException e) {
return new GridCacheQueryErrorFuture<>(cctx.kernalContext(), e);
}
}
if (subjId == null) subjId = cctx.localNodeId();
taskHash = cctx.kernalContext().job().currentTaskNameHash();
final GridCacheQueryBean bean =
new GridCacheQueryBean(
this,
(IgniteReducer<Object, Object>) rmtReducer,
(IgniteClosure<Object, Object>) rmtTransform,
args);
final GridCacheQueryManager qryMgr = cctx.queries();
boolean loc = nodes.size() == 1 && F.first(nodes).id().equals(cctx.localNodeId());
if (type == SQL_FIELDS || type == SPI)
return (CacheQueryFuture<R>)
(loc ? qryMgr.queryFieldsLocal(bean) : qryMgr.queryFieldsDistributed(bean, nodes));
else if (type == SCAN && part != null && nodes.size() > 1)
return new CacheQueryFallbackFuture<>(nodes, bean, qryMgr);
else
return (CacheQueryFuture<R>)
(loc ? qryMgr.queryLocal(bean) : qryMgr.queryDistributed(bean, nodes));
}
/**
* @param cacheId Cache ID to check.
* @return {@code True} if cache is stopping by this exchange.
*/
private boolean stopping(int cacheId) {
boolean stopping = false;
if (!F.isEmpty(reqs)) {
for (DynamicCacheChangeRequest req : reqs) {
if (cacheId == CU.cacheId(req.cacheName())) {
stopping = req.stop();
break;
}
}
}
return stopping;
}
/**
* @param nodeId Node ID.
* @param retryCnt Number of retries.
*/
private void sendAllPartitions(final UUID nodeId, final int retryCnt) {
ClusterNode n = cctx.node(nodeId);
try {
if (n != null) sendAllPartitions(F.asList(n), exchId);
} catch (IgniteCheckedException e) {
if (e instanceof ClusterTopologyCheckedException || !cctx.discovery().alive(n)) {
log.debug(
"Failed to send full partition map to node, node left grid "
+ "[rmtNode="
+ nodeId
+ ", exchangeId="
+ exchId
+ ']');
return;
}
if (retryCnt > 0) {
long timeout = cctx.gridConfig().getNetworkSendRetryDelay();
LT.error(
log,
e,
"Failed to send full partition map to node (will retry after timeout) "
+ "[node="
+ nodeId
+ ", exchangeId="
+ exchId
+ ", timeout="
+ timeout
+ ']');
cctx.time()
.addTimeoutObject(
new GridTimeoutObjectAdapter(timeout) {
@Override
public void onTimeout() {
sendAllPartitions(nodeId, retryCnt - 1);
}
});
} else
U.error(
log,
"Failed to send full partition map [node=" + n + ", exchangeId=" + exchId + ']',
e);
}
}
/**
* Processes cache query request.
*
* @param sndId Sender node id.
* @param req Query request.
*/
@SuppressWarnings("unchecked")
@Override
void processQueryRequest(UUID sndId, GridCacheQueryRequest req) {
if (req.cancel()) {
cancelIds.add(new CancelMessageId(req.id(), sndId));
if (req.fields()) removeFieldsQueryResult(sndId, req.id());
else removeQueryResult(sndId, req.id());
} else {
if (!cancelIds.contains(new CancelMessageId(req.id(), sndId))) {
if (!F.eq(req.cacheName(), cctx.name())) {
GridCacheQueryResponse res =
new GridCacheQueryResponse(
cctx.cacheId(),
req.id(),
new IgniteCheckedException(
"Received request for incorrect cache [expected="
+ cctx.name()
+ ", actual="
+ req.cacheName()));
sendQueryResponse(sndId, res, 0);
} else {
threads.put(req.id(), Thread.currentThread());
try {
GridCacheQueryInfo info = distributedQueryInfo(sndId, req);
if (info == null) return;
if (req.fields()) runFieldsQuery(info);
else runQuery(info);
} catch (Throwable e) {
U.error(log(), "Failed to run query.", e);
sendQueryResponse(
sndId, new GridCacheQueryResponse(cctx.cacheId(), req.id(), e.getCause()), 0);
if (e instanceof Error) throw (Error) e;
} finally {
threads.remove(req.id());
}
}
}
}
}
/** @return {@code True} if succeeded. */
private boolean spreadPartitions() {
try {
sendAllPartitions(rmtNodes, exchId);
return true;
} catch (IgniteCheckedException e) {
scheduleRecheck();
if (!X.hasCause(e, InterruptedException.class))
U.error(
log,
"Failed to send full partition map to nodes (will retry after timeout) [nodes="
+ F.nodeId8s(rmtNodes)
+ ", exchangeId="
+ exchId
+ ']',
e);
return false;
}
}
/**
* Constructor.
*
* @param igfsCtx IGFS context.
*/
IgfsDeleteWorker(IgfsContext igfsCtx) {
super(
"igfs-delete-worker%"
+ igfsCtx.igfs().name()
+ "%"
+ igfsCtx.kernalContext().localNodeId()
+ "%");
this.igfsCtx = igfsCtx;
meta = igfsCtx.meta();
data = igfsCtx.data();
evts = igfsCtx.kernalContext().event();
String igfsName = igfsCtx.igfs().name();
topic = F.isEmpty(igfsName) ? TOPIC_IGFS : TOPIC_IGFS.topic(igfsName);
assert meta != null;
assert data != null;
log = igfsCtx.kernalContext().log(IgfsDeleteWorker.class);
}
/**
* Starts activity.
*
* @throws IgniteInterruptedCheckedException If interrupted.
*/
public void init() throws IgniteInterruptedCheckedException {
if (isDone()) return;
if (init.compareAndSet(false, true)) {
if (isDone()) return;
try {
// Wait for event to occur to make sure that discovery
// will return corresponding nodes.
U.await(evtLatch);
assert discoEvt != null : this;
assert !dummy && !forcePreload : this;
ClusterNode oldest = CU.oldestAliveCacheServerNode(cctx, exchId.topologyVersion());
oldestNode.set(oldest);
startCaches();
// True if client node joined or failed.
boolean clientNodeEvt;
if (F.isEmpty(reqs)) {
int type = discoEvt.type();
assert type == EVT_NODE_JOINED || type == EVT_NODE_LEFT || type == EVT_NODE_FAILED
: discoEvt;
clientNodeEvt = CU.clientNode(discoEvt.eventNode());
} else {
assert discoEvt.type() == EVT_DISCOVERY_CUSTOM_EVT : discoEvt;
boolean clientOnlyStart = true;
for (DynamicCacheChangeRequest req : reqs) {
if (!req.clientStartOnly()) {
clientOnlyStart = false;
break;
}
}
clientNodeEvt = clientOnlyStart;
}
if (clientNodeEvt) {
ClusterNode node = discoEvt.eventNode();
// Client need to initialize affinity for local join event or for stated client caches.
if (!node.isLocal()) {
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
GridDhtPartitionTopology top = cacheCtx.topology();
top.updateTopologyVersion(exchId, this, -1, stopping(cacheCtx.cacheId()));
if (cacheCtx.affinity().affinityTopologyVersion() == AffinityTopologyVersion.NONE) {
initTopology(cacheCtx);
top.beforeExchange(this);
} else
cacheCtx.affinity().clientEventTopologyChange(discoEvt, exchId.topologyVersion());
}
if (exchId.isLeft())
cctx.mvcc().removeExplicitNodeLocks(exchId.nodeId(), exchId.topologyVersion());
onDone(exchId.topologyVersion());
skipPreload = cctx.kernalContext().clientNode();
return;
}
}
if (cctx.kernalContext().clientNode()) {
skipPreload = true;
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
GridDhtPartitionTopology top = cacheCtx.topology();
top.updateTopologyVersion(exchId, this, -1, stopping(cacheCtx.cacheId()));
}
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
initTopology(cacheCtx);
}
if (oldestNode.get() != null) {
rmtNodes =
new ConcurrentLinkedQueue<>(
CU.aliveRemoteServerNodesWithCaches(cctx, exchId.topologyVersion()));
rmtIds = Collections.unmodifiableSet(new HashSet<>(F.nodeIds(rmtNodes)));
ready.set(true);
initFut.onDone(true);
if (log.isDebugEnabled()) log.debug("Initialized future: " + this);
sendPartitions();
} else onDone(exchId.topologyVersion());
return;
}
assert oldestNode.get() != null;
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (isCacheAdded(cacheCtx.cacheId(), exchId.topologyVersion())) {
if (cacheCtx
.discovery()
.cacheAffinityNodes(cacheCtx.name(), topologyVersion())
.isEmpty())
U.quietAndWarn(log, "No server nodes found for cache client: " + cacheCtx.namex());
}
cacheCtx.preloader().onExchangeFutureAdded();
}
List<String> cachesWithoutNodes = null;
if (exchId.isLeft()) {
for (String name : cctx.cache().cacheNames()) {
if (cctx.discovery().cacheAffinityNodes(name, topologyVersion()).isEmpty()) {
if (cachesWithoutNodes == null) cachesWithoutNodes = new ArrayList<>();
cachesWithoutNodes.add(name);
// Fire event even if there is no client cache started.
if (cctx.gridEvents().isRecordable(EventType.EVT_CACHE_NODES_LEFT)) {
Event evt =
new CacheEvent(
name,
cctx.localNode(),
cctx.localNode(),
"All server nodes have left the cluster.",
EventType.EVT_CACHE_NODES_LEFT,
0,
false,
null,
null,
null,
null,
false,
null,
false,
null,
null,
null);
cctx.gridEvents().record(evt);
}
}
}
}
if (cachesWithoutNodes != null) {
StringBuilder sb =
new StringBuilder(
"All server nodes for the following caches have left the cluster: ");
for (int i = 0; i < cachesWithoutNodes.size(); i++) {
String cache = cachesWithoutNodes.get(i);
sb.append('\'').append(cache).append('\'');
if (i != cachesWithoutNodes.size() - 1) sb.append(", ");
}
U.quietAndWarn(log, sb.toString());
U.quietAndWarn(log, "Must have server nodes for caches to operate.");
}
assert discoEvt != null;
assert exchId.nodeId().equals(discoEvt.eventNode().id());
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
GridClientPartitionTopology clientTop =
cctx.exchange().clearClientTopology(cacheCtx.cacheId());
long updSeq = clientTop == null ? -1 : clientTop.lastUpdateSequence();
// Update before waiting for locks.
if (!cacheCtx.isLocal())
cacheCtx
.topology()
.updateTopologyVersion(exchId, this, updSeq, stopping(cacheCtx.cacheId()));
}
// Grab all alive remote nodes with order of equal or less than last joined node.
rmtNodes =
new ConcurrentLinkedQueue<>(
CU.aliveRemoteServerNodesWithCaches(cctx, exchId.topologyVersion()));
rmtIds = Collections.unmodifiableSet(new HashSet<>(F.nodeIds(rmtNodes)));
for (Map.Entry<UUID, GridDhtPartitionsSingleMessage> m : singleMsgs.entrySet())
// If received any messages, process them.
onReceive(m.getKey(), m.getValue());
for (Map.Entry<UUID, GridDhtPartitionsFullMessage> m : fullMsgs.entrySet())
// If received any messages, process them.
onReceive(m.getKey(), m.getValue());
AffinityTopologyVersion topVer = exchId.topologyVersion();
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
// Must initialize topology after we get discovery event.
initTopology(cacheCtx);
cacheCtx.preloader().updateLastExchangeFuture(this);
}
IgniteInternalFuture<?> partReleaseFut = cctx.partitionReleaseFuture(topVer);
// Assign to class variable so it will be included into toString() method.
this.partReleaseFut = partReleaseFut;
if (log.isDebugEnabled()) log.debug("Before waiting for partition release future: " + this);
while (true) {
try {
partReleaseFut.get(2 * cctx.gridConfig().getNetworkTimeout(), TimeUnit.MILLISECONDS);
break;
} catch (IgniteFutureTimeoutCheckedException ignored) {
// Print pending transactions and locks that might have led to hang.
dumpPendingObjects();
}
}
if (log.isDebugEnabled()) log.debug("After waiting for partition release future: " + this);
if (!F.isEmpty(reqs)) blockGateways();
if (exchId.isLeft())
cctx.mvcc().removeExplicitNodeLocks(exchId.nodeId(), exchId.topologyVersion());
IgniteInternalFuture<?> locksFut = cctx.mvcc().finishLocks(exchId.topologyVersion());
while (true) {
try {
locksFut.get(2 * cctx.gridConfig().getNetworkTimeout(), TimeUnit.MILLISECONDS);
break;
} catch (IgniteFutureTimeoutCheckedException ignored) {
U.warn(
log,
"Failed to wait for locks release future. "
+ "Dumping pending objects that might be the cause: "
+ cctx.localNodeId());
U.warn(log, "Locked entries:");
Map<IgniteTxKey, Collection<GridCacheMvccCandidate>> locks =
cctx.mvcc().unfinishedLocks(exchId.topologyVersion());
for (Map.Entry<IgniteTxKey, Collection<GridCacheMvccCandidate>> e : locks.entrySet())
U.warn(log, "Locked entry [key=" + e.getKey() + ", mvcc=" + e.getValue() + ']');
}
}
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
// Notify replication manager.
GridCacheContext drCacheCtx =
cacheCtx.isNear() ? cacheCtx.near().dht().context() : cacheCtx;
if (drCacheCtx.isDrEnabled()) drCacheCtx.dr().beforeExchange(topVer, exchId.isLeft());
// Partition release future is done so we can flush the write-behind store.
cacheCtx.store().forceFlush();
// Process queued undeploys prior to sending/spreading map.
cacheCtx.preloader().unwindUndeploys();
GridDhtPartitionTopology top = cacheCtx.topology();
assert topVer.equals(top.topologyVersion())
: "Topology version is updated only in this class instances inside single ExchangeWorker thread.";
top.beforeExchange(this);
}
for (GridClientPartitionTopology top : cctx.exchange().clientTopologies()) {
top.updateTopologyVersion(exchId, this, -1, stopping(top.cacheId()));
top.beforeExchange(this);
}
} catch (IgniteInterruptedCheckedException e) {
onDone(e);
throw e;
} catch (Throwable e) {
U.error(
log,
"Failed to reinitialize local partitions (preloading will be stopped): " + exchId,
e);
onDone(e);
if (e instanceof Error) throw (Error) e;
return;
}
if (F.isEmpty(rmtIds)) {
onDone(exchId.topologyVersion());
return;
}
ready.set(true);
initFut.onDone(true);
if (log.isDebugEnabled()) log.debug("Initialized future: " + this);
// If this node is not oldest.
if (!oldestNode.get().id().equals(cctx.localNodeId())) sendPartitions();
else {
boolean allReceived = allReceived();
if (allReceived && replied.compareAndSet(false, true)) {
if (spreadPartitions()) onDone(exchId.topologyVersion());
}
}
scheduleRecheck();
} else assert false : "Skipped init future: " + this;
}
/** @return Remaining node IDs. */
Collection<UUID> remaining() {
if (rmtIds == null) return Collections.emptyList();
return F.lose(rmtIds, true, rcvdIds);
}