/**
* @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));
}
});
}
/**
* @param cacheCtx Cache context.
* @return {@code True} if local node can calculate affinity on it's own for this partition map
* exchange.
*/
private boolean canCalculateAffinity(GridCacheContext cacheCtx) {
AffinityFunction affFunc = cacheCtx.config().getAffinity();
// Do not request affinity from remote nodes if affinity function is not centralized.
if (!U.hasAnnotation(affFunc, AffinityCentralizedFunction.class)) return true;
// If local node did not initiate exchange or local node is the only cache node in grid.
Collection<ClusterNode> affNodes = CU.affinityNodes(cacheCtx, exchId.topologyVersion());
return !exchId.nodeId().equals(cctx.localNodeId())
|| (affNodes.size() == 1 && affNodes.contains(cctx.localNode()));
}
/**
* @param cacheCtx Cache context.
* @throws IgniteCheckedException If failed.
*/
private void initTopology(GridCacheContext cacheCtx) throws IgniteCheckedException {
if (stopping(cacheCtx.cacheId())) return;
if (canCalculateAffinity(cacheCtx)) {
if (log.isDebugEnabled())
log.debug(
"Will recalculate affinity [locNodeId="
+ cctx.localNodeId()
+ ", exchId="
+ exchId
+ ']');
cacheCtx.affinity().calculateAffinity(exchId.topologyVersion(), discoEvt);
} else {
if (log.isDebugEnabled())
log.debug(
"Will request affinity from remote node [locNodeId="
+ cctx.localNodeId()
+ ", exchId="
+ exchId
+ ']');
// Fetch affinity assignment from remote node.
GridDhtAssignmentFetchFuture fetchFut =
new GridDhtAssignmentFetchFuture(
cacheCtx,
exchId.topologyVersion(),
CU.affinityNodes(cacheCtx, exchId.topologyVersion()));
fetchFut.init();
List<List<ClusterNode>> affAssignment = fetchFut.get();
if (log.isDebugEnabled())
log.debug(
"Fetched affinity from remote node, initializing affinity assignment [locNodeId="
+ cctx.localNodeId()
+ ", topVer="
+ exchId.topologyVersion()
+ ']');
if (affAssignment == null) {
affAssignment = new ArrayList<>(cacheCtx.affinity().partitions());
List<ClusterNode> empty = Collections.emptyList();
for (int i = 0; i < cacheCtx.affinity().partitions(); i++) affAssignment.add(empty);
}
cacheCtx.affinity().initializeAffinity(exchId.topologyVersion(), affAssignment);
}
}
/**
* @param p Partition.
* @param topVer Topology version ({@code -1} for all nodes).
* @param state Partition state.
* @param states Additional partition states.
* @return List of nodes for the partition.
*/
private List<ClusterNode> nodes(
int p,
AffinityTopologyVersion topVer,
GridDhtPartitionState state,
GridDhtPartitionState... states) {
Collection<UUID> allIds =
topVer.topologyVersion() > 0 ? F.nodeIds(CU.affinityNodes(cctx, topVer)) : null;
lock.readLock().lock();
try {
assert node2part != null && node2part.valid()
: "Invalid node-to-partitions map [topVer="
+ topVer
+ ", allIds="
+ allIds
+ ", node2part="
+ node2part
+ ", cache="
+ cctx.name()
+ ']';
Collection<UUID> nodeIds = part2node.get(p);
// Node IDs can be null if both, primary and backup, nodes disappear.
int size = nodeIds == null ? 0 : nodeIds.size();
if (size == 0) return Collections.emptyList();
List<ClusterNode> nodes = new ArrayList<>(size);
for (UUID id : nodeIds) {
if (topVer.topologyVersion() > 0 && !allIds.contains(id)) continue;
if (hasState(p, id, state, states)) {
ClusterNode n = cctx.discovery().node(id);
if (n != null && (topVer.topologyVersion() < 0 || n.order() <= topVer.topologyVersion()))
nodes.add(n);
}
}
return nodes;
} finally {
lock.readLock().unlock();
}
}
/**
* Maps keys to nodes. Note that we can not simply group keys by nodes and send lock request as
* such approach does not preserve order of lock acquisition. Instead, keys are split in
* continuous groups belonging to one primary node and locks for these groups are acquired
* sequentially.
*
* @param keys Keys.
*/
private void map(Iterable<? extends K> keys) {
try {
GridDiscoveryTopologySnapshot snapshot = topSnapshot.get();
assert snapshot != null;
long topVer = snapshot.topologyVersion();
assert topVer > 0;
if (CU.affinityNodes(cctx, topVer).isEmpty()) {
onDone(
new GridTopologyException(
"Failed to map keys for near-only cache (all "
+ "partition nodes left the grid)."));
return;
}
ConcurrentLinkedDeque8<GridNearLockMapping<K, V>> mappings = new ConcurrentLinkedDeque8<>();
// Assign keys to primary nodes.
GridNearLockMapping<K, V> map = null;
for (K key : keys) {
GridNearLockMapping<K, V> updated = map(key, map, topVer);
// If new mapping was created, add to collection.
if (updated != map) mappings.add(updated);
map = updated;
}
if (isDone()) {
if (log.isDebugEnabled()) log.debug("Abandoning (re)map because future is done: " + this);
return;
}
if (log.isDebugEnabled())
log.debug("Starting (re)map for mappings [mappings=" + mappings + ", fut=" + this + ']');
// Create mini futures.
for (Iterator<GridNearLockMapping<K, V>> iter = mappings.iterator(); iter.hasNext(); ) {
GridNearLockMapping<K, V> mapping = iter.next();
GridNode node = mapping.node();
Collection<K> mappedKeys = mapping.mappedKeys();
assert !mappedKeys.isEmpty();
GridNearLockRequest<K, V> req = null;
Collection<K> distributedKeys = new ArrayList<>(mappedKeys.size());
boolean explicit = false;
for (K key : mappedKeys) {
while (true) {
GridNearCacheEntry<K, V> entry = null;
try {
entry = cctx.near().entryExx(key, topVer);
if (!cctx.isAll(entry.wrap(false), filter)) {
if (log.isDebugEnabled())
log.debug("Entry being locked did not pass filter (will not lock): " + entry);
onComplete(false, false);
return;
}
// Removed exception may be thrown here.
GridCacheMvccCandidate<K> cand = addEntry(topVer, entry, node.id());
if (isDone()) {
if (log.isDebugEnabled())
log.debug(
"Abandoning (re)map because future is done after addEntry attempt "
+ "[fut="
+ this
+ ", entry="
+ entry
+ ']');
return;
}
if (cand != null) {
if (tx == null && !cand.reentry())
cctx.mvcc().addExplicitLock(threadId, cand, snapshot);
GridTuple3<GridCacheVersion, V, byte[]> val = entry.versionedValue();
if (val == null) {
GridDhtCacheEntry<K, V> dhtEntry = dht().peekExx(key);
try {
if (dhtEntry != null) val = dhtEntry.versionedValue(topVer);
} catch (GridCacheEntryRemovedException ignored) {
assert dhtEntry.obsolete()
: " Got removed exception for non-obsolete entry: " + dhtEntry;
if (log.isDebugEnabled())
log.debug(
"Got removed exception for DHT entry in map (will ignore): " + dhtEntry);
}
}
GridCacheVersion dhtVer = null;
if (val != null) {
dhtVer = val.get1();
valMap.put(key, val);
}
if (!cand.reentry()) {
if (req == null) {
req =
new GridNearLockRequest<>(
topVer,
cctx.nodeId(),
threadId,
futId,
lockVer,
inTx(),
implicitTx(),
implicitSingleTx(),
read,
isolation(),
isInvalidate(),
timeout,
syncCommit(),
syncRollback(),
mappedKeys.size(),
inTx() ? tx.size() : mappedKeys.size(),
inTx() ? tx.groupLockKey() : null,
inTx() && tx.partitionLock(),
inTx() ? tx.subjectId() : null);
mapping.request(req);
}
distributedKeys.add(key);
GridCacheTxEntry<K, V> writeEntry = tx != null ? tx.writeMap().get(key) : null;
if (tx != null) tx.addKeyMapping(key, mapping.node());
req.addKeyBytes(
key,
node.isLocal() ? null : entry.getOrMarshalKeyBytes(),
retval && dhtVer == null,
dhtVer, // Include DHT version to match remote DHT entry.
writeEntry,
inTx() ? tx.entry(key).drVersion() : null,
cctx);
// Clear transfer required flag since we are sending message.
if (writeEntry != null) writeEntry.transferRequired(false);
}
if (cand.reentry())
explicit = tx != null && !entry.hasLockCandidate(tx.xidVersion());
} else
// Ignore reentries within transactions.
explicit = tx != null && !entry.hasLockCandidate(tx.xidVersion());
if (explicit) tx.addKeyMapping(key, mapping.node());
break;
} catch (GridCacheEntryRemovedException ignored) {
assert entry.obsolete() : "Got removed exception on non-obsolete entry: " + entry;
if (log.isDebugEnabled())
log.debug("Got removed entry in lockAsync(..) method (will retry): " + entry);
}
}
// Mark mapping explicit lock flag.
if (explicit) {
boolean marked = tx != null && tx.markExplicit(node.id());
assert tx == null || marked;
}
}
if (!distributedKeys.isEmpty()) mapping.distributedKeys(distributedKeys);
else {
assert mapping.request() == null;
iter.remove();
}
}
cctx.mvcc().recheckPendingLocks();
proceedMapping(mappings);
} catch (GridException ex) {
onError(ex);
}
}