/** @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);
}
}
/** @return Nodes to execute on. */
private Collection<GridNode> nodes() {
GridCacheMode 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) return nodes(cctx, prj);
GridCacheDistributionMode mode = cctx.config().getDistributionMode();
return mode == PARTITIONED_ONLY || mode == NEAR_PARTITIONED
? Collections.singletonList(cctx.localNode())
: Collections.singletonList(F.rand(nodes(cctx, null)));
case PARTITIONED:
return nodes(cctx, prj);
default:
throw new IllegalStateException("Unknown cache distribution mode: " + cacheMode);
}
}
/** {@inheritDoc} */
@Override
public GridDhtPartitionFullMap partitionMap(boolean onlyActive) {
lock.readLock().lock();
try {
assert node2part != null && node2part.valid()
: "Invalid node2part [node2part: "
+ node2part
+ ", cache="
+ cctx.name()
+ ", started="
+ cctx.started()
+ ", stopping="
+ stopping
+ ", locNodeId="
+ cctx.localNode().id()
+ ", locName="
+ cctx.gridName()
+ ']';
GridDhtPartitionFullMap m = node2part;
return new GridDhtPartitionFullMap(
m.nodeId(), m.nodeOrder(), m.updateSequence(), m, onlyActive);
} finally {
lock.readLock().unlock();
}
}
/** {@inheritDoc} */
@Override
public boolean own(GridDhtLocalPartition part) {
ClusterNode loc = cctx.localNode();
lock.writeLock().lock();
try {
if (part.own()) {
updateLocal(part.id(), loc.id(), part.state(), updateSeq.incrementAndGet());
consistencyCheck();
return true;
}
consistencyCheck();
return false;
} finally {
lock.writeLock().unlock();
}
}
/** @param nodeId Node to remove. */
private void removeNode(UUID nodeId) {
assert nodeId != null;
assert lock.writeLock().isHeldByCurrentThread();
ClusterNode oldest = CU.oldestAliveCacheServerNode(cctx.shared(), topVer);
assert oldest != null;
ClusterNode loc = cctx.localNode();
if (node2part != null) {
if (oldest.equals(loc) && !node2part.nodeId().equals(loc.id())) {
updateSeq.setIfGreater(node2part.updateSequence());
node2part =
new GridDhtPartitionFullMap(
loc.id(), loc.order(), updateSeq.incrementAndGet(), node2part, false);
} else node2part = new GridDhtPartitionFullMap(node2part, node2part.updateSequence());
part2node = new HashMap<>(part2node);
GridDhtPartitionMap parts = node2part.remove(nodeId);
if (parts != null) {
for (Integer p : parts.keySet()) {
Set<UUID> nodeIds = part2node.get(p);
if (nodeIds != null) {
nodeIds.remove(nodeId);
if (nodeIds.isEmpty()) part2node.remove(p);
}
}
}
consistencyCheck();
}
}
/** {@inheritDoc} */
@Override
public void execute(@Nullable GridProjection prj) throws GridException {
if (cb == null)
throw new IllegalStateException("Mandatory local callback is not set for the query: " + this);
if (prj == null) prj = ctx.grid();
prj = prj.forCache(ctx.name());
if (prj.nodes().isEmpty())
throw new GridTopologyException("Failed to execute query (projection is empty): " + this);
GridCacheMode mode = ctx.config().getCacheMode();
if (mode == LOCAL || mode == REPLICATED) {
Collection<GridNode> nodes = prj.nodes();
GridNode node = nodes.contains(ctx.localNode()) ? ctx.localNode() : F.rand(nodes);
assert node != null;
if (nodes.size() > 1 && !ctx.cache().isDrSystemCache()) {
if (node.id().equals(ctx.localNodeId()))
U.warn(
log,
"Continuous query for "
+ mode
+ " cache can be run only on local node. "
+ "Will execute query locally: "
+ this);
else
U.warn(
log,
"Continuous query for "
+ mode
+ " cache can be run only on single node. "
+ "Will execute query on remote node [qry="
+ this
+ ", node="
+ node
+ ']');
}
prj = prj.forNode(node);
}
closeLock.lock();
try {
if (routineId != null)
throw new IllegalStateException("Continuous query can't be executed twice.");
guard.block();
GridContinuousHandler hnd =
new GridCacheContinuousQueryHandler<>(ctx.name(), topic, cb, filter, prjPred);
routineId =
ctx.kernalContext()
.continuous()
.startRoutine(hnd, bufSize, timeInterval, autoUnsubscribe, prj.predicate())
.get();
} finally {
closeLock.unlock();
}
}
/**
* @param topVer Topology version.
* @return {@code True} if local node is primary for this partition.
*/
public boolean primary(AffinityTopologyVersion topVer) {
return cctx.affinity().primary(cctx.localNode(), id, topVer);
}
/**
* Gets next near lock mapping and either acquires dht locks locally or sends near lock request to
* remote primary node.
*
* @param mappings Queue of mappings.
* @throws GridException If mapping can not be completed.
*/
private void proceedMapping(final ConcurrentLinkedDeque8<GridNearLockMapping<K, V>> mappings)
throws GridException {
GridNearLockMapping<K, V> map = mappings.poll();
// If there are no more mappings to process, complete the future.
if (map == null) return;
final GridNearLockRequest<K, V> req = map.request();
final Collection<K> mappedKeys = map.distributedKeys();
final GridNode node = map.node();
if (filter != null && filter.length != 0) req.filter(filter, cctx);
if (node.isLocal()) {
req.miniId(GridUuid.randomUuid());
if (log.isDebugEnabled()) log.debug("Before locally locking near request: " + req);
GridFuture<GridNearLockResponse<K, V>> fut;
if (CU.DHT_ENABLED) fut = dht().lockAllAsync(cctx.localNode(), req, filter);
else {
// Create dummy values for testing.
GridNearLockResponse<K, V> res =
new GridNearLockResponse<>(lockVer, futId, null, false, 1, null);
res.addValueBytes(null, null, true, lockVer, lockVer, cctx);
fut = new GridFinishedFuture<>(ctx, res);
}
// Add new future.
add(
new GridEmbeddedFuture<>(
cctx.kernalContext(),
fut,
new C2<GridNearLockResponse<K, V>, Exception, Boolean>() {
@Override
public Boolean apply(GridNearLockResponse<K, V> res, Exception e) {
if (CU.isLockTimeoutOrCancelled(e)
|| (res != null && CU.isLockTimeoutOrCancelled(res.error()))) return false;
if (e != null) {
onError(e);
return false;
}
if (res == null) {
onError(new GridException("Lock response is null for future: " + this));
return false;
}
if (res.error() != null) {
onError(res.error());
return false;
}
if (log.isDebugEnabled())
log.debug(
"Acquired lock for local DHT mapping [locId="
+ cctx.nodeId()
+ ", mappedKeys="
+ mappedKeys
+ ", fut="
+ GridNearLockFuture.this
+ ']');
try {
int i = 0;
for (K k : mappedKeys) {
while (true) {
GridNearCacheEntry<K, V> entry =
cctx.near().entryExx(k, req.topologyVersion());
try {
GridTuple3<GridCacheVersion, V, byte[]> oldValTup =
valMap.get(entry.key());
boolean hasBytes = entry.hasValue();
V oldVal = entry.rawGet();
V newVal = res.value(i);
byte[] newBytes = res.valueBytes(i);
GridCacheVersion dhtVer = res.dhtVersion(i);
GridCacheVersion mappedVer = res.mappedVersion(i);
// On local node don't record twice if DHT cache already recorded.
boolean record =
retval && oldValTup != null && oldValTup.get1().equals(dhtVer);
if (newVal == null) {
if (oldValTup != null) {
if (oldValTup.get1().equals(dhtVer)) {
newVal = oldValTup.get2();
newBytes = oldValTup.get3();
}
oldVal = oldValTup.get2();
}
}
// Lock is held at this point, so we can set the
// returned value if any.
entry.resetFromPrimary(newVal, newBytes, lockVer, dhtVer, node.id());
entry.readyNearLock(
lockVer,
mappedVer,
res.committedVersions(),
res.rolledbackVersions(),
res.pending());
if (inTx() && implicitTx() && tx.onePhaseCommit()) {
boolean pass = res.filterResult(i);
tx.entry(k).filters(pass ? CU.<K, V>empty() : CU.<K, V>alwaysFalse());
}
if (record) {
if (cctx.events().isRecordable(EVT_CACHE_OBJECT_READ))
cctx.events()
.addEvent(
entry.partition(),
entry.key(),
tx,
null,
EVT_CACHE_OBJECT_READ,
newVal,
newVal != null,
oldVal,
hasBytes,
CU.subjectId(tx, cctx));
cctx.cache().metrics0().onRead(oldVal != null);
}
if (log.isDebugEnabled())
log.debug(
"Processed response for entry [res="
+ res
+ ", entry="
+ entry
+ ']');
break; // Inner while loop.
} catch (GridCacheEntryRemovedException ignored) {
if (log.isDebugEnabled())
log.debug(
"Failed to add candidates because entry was "
+ "removed (will renew).");
// Replace old entry with new one.
entries.set(
i,
(GridDistributedCacheEntry<K, V>) cctx.cache().entryEx(entry.key()));
}
}
i++; // Increment outside of while loop.
}
// Proceed and add new future (if any) before completing embedded future.
proceedMapping(mappings);
} catch (GridException ex) {
onError(ex);
return false;
}
return true;
}
}));
} else {
final MiniFuture fut = new MiniFuture(node, mappedKeys, mappings);
req.miniId(fut.futureId());
add(fut); // Append new future.
GridFuture<?> txSync = null;
if (inTx()) txSync = cctx.tm().awaitFinishAckAsync(node.id(), tx.threadId());
if (txSync == null || txSync.isDone()) {
try {
if (log.isDebugEnabled())
log.debug("Sending near lock request [node=" + node.id() + ", req=" + req + ']');
cctx.io().send(node, req);
} catch (GridTopologyException ex) {
assert fut != null;
fut.onResult(ex);
}
} else {
txSync.listenAsync(
new CI1<GridFuture<?>>() {
@Override
public void apply(GridFuture<?> t) {
try {
if (log.isDebugEnabled())
log.debug(
"Sending near lock request [node=" + node.id() + ", req=" + req + ']');
cctx.io().send(node, req);
} catch (GridTopologyException ex) {
assert fut != null;
fut.onResult(ex);
} catch (GridException e) {
onError(e);
}
}
});
}
}
}
/** {@inheritDoc} */
@Override
public boolean afterExchange(GridDhtPartitionsExchangeFuture exchFut)
throws IgniteCheckedException {
boolean changed = waitForRent();
ClusterNode loc = cctx.localNode();
int num = cctx.affinity().partitions();
AffinityTopologyVersion topVer = exchFut.topologyVersion();
lock.writeLock().lock();
try {
if (stopping) return false;
assert topVer.equals(exchFut.topologyVersion())
: "Invalid topology version [topVer=" + topVer + ", exchId=" + exchFut.exchangeId() + ']';
if (log.isDebugEnabled())
log.debug(
"Partition map before afterExchange [exchId="
+ exchFut.exchangeId()
+ ", fullMap="
+ fullMapString()
+ ']');
long updateSeq = this.updateSeq.incrementAndGet();
for (int p = 0; p < num; p++) {
GridDhtLocalPartition locPart = localPartition(p, topVer, false, false);
if (cctx.affinity().localNode(p, topVer)) {
// This partition will be created during next topology event,
// which obviously has not happened at this point.
if (locPart == null) {
if (log.isDebugEnabled())
log.debug("Skipping local partition afterExchange (will not create): " + p);
continue;
}
GridDhtPartitionState state = locPart.state();
if (state == MOVING) {
if (cctx.rebalanceEnabled()) {
Collection<ClusterNode> owners = owners(p);
// If there are no other owners, then become an owner.
if (F.isEmpty(owners)) {
boolean owned = locPart.own();
assert owned
: "Failed to own partition [cacheName"
+ cctx.name()
+ ", locPart="
+ locPart
+ ']';
updateLocal(p, loc.id(), locPart.state(), updateSeq);
changed = true;
if (cctx.events().isRecordable(EVT_CACHE_REBALANCE_PART_DATA_LOST)) {
DiscoveryEvent discoEvt = exchFut.discoveryEvent();
cctx.events()
.addPreloadEvent(
p,
EVT_CACHE_REBALANCE_PART_DATA_LOST,
discoEvt.eventNode(),
discoEvt.type(),
discoEvt.timestamp());
}
if (log.isDebugEnabled()) log.debug("Owned partition: " + locPart);
} else if (log.isDebugEnabled())
log.debug(
"Will not own partition (there are owners to rebalance from) [locPart="
+ locPart
+ ", owners = "
+ owners
+ ']');
} else updateLocal(p, loc.id(), locPart.state(), updateSeq);
}
} else {
if (locPart != null) {
GridDhtPartitionState state = locPart.state();
if (state == MOVING) {
locPart.rent(false);
updateLocal(p, loc.id(), locPart.state(), updateSeq);
changed = true;
if (log.isDebugEnabled())
log.debug("Evicting moving partition (it does not belong to affinity): " + locPart);
}
}
}
}
consistencyCheck();
} finally {
lock.writeLock().unlock();
}
return changed;
}
/** {@inheritDoc} */
@Override
public void beforeExchange(GridDhtPartitionsExchangeFuture exchFut)
throws IgniteCheckedException {
waitForRent();
ClusterNode loc = cctx.localNode();
int num = cctx.affinity().partitions();
lock.writeLock().lock();
try {
GridDhtPartitionExchangeId exchId = exchFut.exchangeId();
if (stopping) return;
assert topVer.equals(exchId.topologyVersion())
: "Invalid topology version [topVer=" + topVer + ", exchId=" + exchId + ']';
if (exchId.isLeft()) removeNode(exchId.nodeId());
// In case if node joins, get topology at the time of joining node.
ClusterNode oldest = CU.oldestAliveCacheServerNode(cctx.shared(), topVer);
assert oldest != null;
if (log.isDebugEnabled())
log.debug(
"Partition map beforeExchange [exchId="
+ exchId
+ ", fullMap="
+ fullMapString()
+ ']');
long updateSeq = this.updateSeq.incrementAndGet();
// If this is the oldest node.
if (oldest.id().equals(loc.id())
|| exchFut.isCacheAdded(cctx.cacheId(), exchId.topologyVersion())) {
if (node2part == null) {
node2part = new GridDhtPartitionFullMap(oldest.id(), oldest.order(), updateSeq);
if (log.isDebugEnabled())
log.debug(
"Created brand new full topology map on oldest node [exchId="
+ exchId
+ ", fullMap="
+ fullMapString()
+ ']');
} else if (!node2part.valid()) {
node2part =
new GridDhtPartitionFullMap(oldest.id(), oldest.order(), updateSeq, node2part, false);
if (log.isDebugEnabled())
log.debug(
"Created new full topology map on oldest node [exchId="
+ exchId
+ ", fullMap="
+ node2part
+ ']');
} else if (!node2part.nodeId().equals(loc.id())) {
node2part =
new GridDhtPartitionFullMap(oldest.id(), oldest.order(), updateSeq, node2part, false);
if (log.isDebugEnabled())
log.debug(
"Copied old map into new map on oldest node (previous oldest node left) [exchId="
+ exchId
+ ", fullMap="
+ fullMapString()
+ ']');
}
}
if (cctx.rebalanceEnabled()) {
for (int p = 0; p < num; p++) {
// If this is the first node in grid.
boolean added = exchFut.isCacheAdded(cctx.cacheId(), exchId.topologyVersion());
if ((oldest.id().equals(loc.id())
&& oldest.id().equals(exchId.nodeId())
&& exchId.isJoined())
|| added) {
assert exchId.isJoined() || added;
try {
GridDhtLocalPartition locPart = localPartition(p, topVer, true, false);
assert locPart != null;
boolean owned = locPart.own();
assert owned
: "Failed to own partition for oldest node [cacheName"
+ cctx.name()
+ ", part="
+ locPart
+ ']';
if (log.isDebugEnabled()) log.debug("Owned partition for oldest node: " + locPart);
updateLocal(p, loc.id(), locPart.state(), updateSeq);
} catch (GridDhtInvalidPartitionException e) {
if (log.isDebugEnabled())
log.debug(
"Ignoring invalid partition on oldest node (no need to create a partition "
+ "if it no longer belongs to local node: "
+ e.partition());
}
}
// If this is not the first node in grid.
else {
if (node2part != null && node2part.valid()) {
if (cctx.affinity().localNode(p, topVer)) {
try {
// This will make sure that all non-existing partitions
// will be created in MOVING state.
GridDhtLocalPartition locPart = localPartition(p, topVer, true, false);
updateLocal(p, loc.id(), locPart.state(), updateSeq);
} catch (GridDhtInvalidPartitionException e) {
if (log.isDebugEnabled())
log.debug(
"Ignoring invalid partition (no need to create a partition if it "
+ "no longer belongs to local node: "
+ e.partition());
}
}
}
// If this node's map is empty, we pre-create local partitions,
// so local map will be sent correctly during exchange.
else if (cctx.affinity().localNode(p, topVer)) {
try {
localPartition(p, topVer, true, false);
} catch (GridDhtInvalidPartitionException e) {
if (log.isDebugEnabled())
log.debug(
"Ignoring invalid partition (no need to pre-create a partition if it "
+ "no longer belongs to local node: "
+ e.partition());
}
}
}
}
} else {
// If preloader is disabled, then we simply clear out
// the partitions this node is not responsible for.
for (int p = 0; p < num; p++) {
GridDhtLocalPartition locPart = localPartition(p, topVer, false, false);
boolean belongs = cctx.affinity().localNode(p, topVer);
if (locPart != null) {
if (!belongs) {
GridDhtPartitionState state = locPart.state();
if (state.active()) {
locPart.rent(false);
updateLocal(p, loc.id(), locPart.state(), updateSeq);
if (log.isDebugEnabled())
log.debug(
"Evicting partition with rebalancing disabled "
+ "(it does not belong to affinity): "
+ locPart);
}
}
} else if (belongs) {
try {
// Pre-create partitions.
localPartition(p, topVer, true, false);
} catch (GridDhtInvalidPartitionException e) {
if (log.isDebugEnabled())
log.debug(
"Ignoring invalid partition with disabled rebalancer (no need to "
+ "pre-create a partition if it no longer belongs to local node: "
+ e.partition());
}
}
}
}
if (node2part != null && node2part.valid()) checkEvictions(updateSeq);
consistencyCheck();
if (log.isDebugEnabled())
log.debug(
"Partition map after beforeExchange [exchId="
+ exchId
+ ", fullMap="
+ fullMapString()
+ ']');
} finally {
lock.writeLock().unlock();
}
// Wait for evictions.
waitForRent();
}
/**
* @param updateSeq Update sequence.
* @return Checks if any of the local partitions need to be evicted.
*/
private boolean checkEvictions(long updateSeq) {
assert lock.isWriteLockedByCurrentThread();
boolean changed = false;
UUID locId = cctx.nodeId();
for (GridDhtLocalPartition part : locParts.values()) {
GridDhtPartitionState state = part.state();
if (state.active()) {
int p = part.id();
List<ClusterNode> affNodes = cctx.affinity().nodes(p, topVer);
if (!affNodes.contains(cctx.localNode())) {
Collection<UUID> nodeIds = F.nodeIds(nodes(p, topVer, OWNING));
// If all affinity nodes are owners, then evict partition from local node.
if (nodeIds.containsAll(F.nodeIds(affNodes))) {
part.rent(false);
updateLocal(part.id(), locId, part.state(), updateSeq);
changed = true;
if (log.isDebugEnabled())
log.debug("Evicted local partition (all affinity nodes are owners): " + part);
} else {
int ownerCnt = nodeIds.size();
int affCnt = affNodes.size();
if (ownerCnt > affCnt) {
List<ClusterNode> sorted = new ArrayList<>(cctx.discovery().nodes(nodeIds));
// Sort by node orders in ascending order.
Collections.sort(sorted, CU.nodeComparator(true));
int diff = sorted.size() - affCnt;
for (int i = 0; i < diff; i++) {
ClusterNode n = sorted.get(i);
if (locId.equals(n.id())) {
part.rent(false);
updateLocal(part.id(), locId, part.state(), updateSeq);
changed = true;
if (log.isDebugEnabled())
log.debug(
"Evicted local partition (this node is oldest non-affinity node): " + part);
break;
}
}
}
}
}
}
}
return changed;
}