use of org.apache.ignite.cache.affinity.AffinityFunction in project ignite by apache.
the class GridCacheAffinityApiSelfTest method testMapPartitionsToNodeCollection.
/**
* JUnit.
*
* @throws Exception If failed.
*/
@Test
public void testMapPartitionsToNodeCollection() throws Exception {
Collection<Integer> parts = new LinkedList<>();
for (int p = 0; p < affinity().partitions(); p++) parts.add(p);
Map<Integer, ClusterNode> map = grid(0).affinity(DEFAULT_CACHE_NAME).mapPartitionsToNodes(parts);
AffinityFunctionContext ctx = new GridAffinityFunctionContextImpl(new ArrayList<>(grid(0).cluster().nodes()), null, null, new AffinityTopologyVersion(1), 1);
AffinityFunction aff = affinity();
List<List<ClusterNode>> assignment = aff.assignPartitions(ctx);
for (Map.Entry<Integer, ClusterNode> e : map.entrySet()) assert F.eqNodes(F.first(nodes(assignment, aff, e.getKey())), e.getValue());
}
use of org.apache.ignite.cache.affinity.AffinityFunction in project ignite by apache.
the class CacheLateAffinityAssignmentTest method testSinglePartitionCacheNodeLeft.
/**
* Since we have only 1 partition, at each node left it will be lost (no rebalance needed) or it will be still
* located at second node (thanks to special affinity) (no rebalance neeeded). So, LAA should never happen.
*
* @param ownerLeft Kill owner flag.
*/
private void testSinglePartitionCacheNodeLeft(boolean ownerLeft) throws Exception {
String cacheName = "single-partitioned";
cacheC = new IgniteClosure<String, CacheConfiguration[]>() {
@Override
public CacheConfiguration[] apply(String igniteInstanceName) {
CacheConfiguration ccfg = new CacheConfiguration();
AffinityFunction aff;
ccfg.setName(cacheName);
ccfg.setWriteSynchronizationMode(FULL_SYNC);
ccfg.setBackups(0);
aff = ownerLeft ? affinityFunction(1) : new MapSinglePartitionToSecondNodeAffinityFunction();
ccfg.setAffinity(aff);
return new CacheConfiguration[] { ccfg };
}
};
int top = 0;
int nodes = 0;
startServer(nodes++, ++top);
checkAffinity(nodes, topVer(top, 0), true);
// Checks LAA is absent on initial topology.
checkNoExchange(nodes, topVer(top, 1));
for (int i = 0; i < 10; i++) startServer(nodes++, ++top);
awaitPartitionMapExchange();
Ignite primary = primaryNode(0, cacheName);
boolean laaOnJoin = primary.cluster().localNode().order() != 1;
boolean leftHappen = false;
while (nodes > 1) {
Map<String, List<List<ClusterNode>>> aff = checkAffinity(nodes, topVer(top, leftHappen ? 0 : (laaOnJoin ? 1 : 0)), true);
ClusterNode owner = aff.get(cacheName).get(/*part*/
0).get(/*primary*/
0);
Ignite actualOwner = primaryNode(0, cacheName);
assertEquals(actualOwner.cluster().localNode().order(), owner.order());
for (Ignite node : G.allGrids()) {
ClusterNode locNode = node.cluster().localNode();
boolean equals = locNode.order() == owner.order();
if (equals == ownerLeft) {
if (!ownerLeft)
assertNotSame(locNode.order(), 2);
grid(locNode).close();
calculateAffinity(++top);
leftHappen = true;
break;
}
}
checkAffinity(--nodes, topVer(top, 0), true);
checkNoExchange(nodes, topVer(top, 1));
}
}
use of org.apache.ignite.cache.affinity.AffinityFunction in project ignite by apache.
the class CacheLateAffinityAssignmentTest method calculateAffinity.
/**
* @param topVer Topology version.
* @param filterByRcvd If {@code true} filters caches by 'receivedFrom' property.
* @param cur Optional current affinity.
* @throws Exception If failed.
* @return {@code True} if some primary node changed comparing to given affinity.
*/
private boolean calculateAffinity(long topVer, boolean filterByRcvd, @Nullable Map<String, List<List<ClusterNode>>> cur) throws Exception {
List<Ignite> all = G.allGrids();
IgniteKernal ignite = (IgniteKernal) Collections.min(all, new Comparator<Ignite>() {
@Override
public int compare(Ignite n1, Ignite n2) {
return Long.compare(n1.cluster().localNode().order(), n2.cluster().localNode().order());
}
});
assert !all.isEmpty();
Map<Integer, List<List<ClusterNode>>> assignments = idealAff.get(topVer);
if (assignments == null)
idealAff.put(topVer, assignments = new HashMap<>());
GridKernalContext ctx = ignite.context();
GridCacheSharedContext cctx = ctx.cache().context();
AffinityTopologyVersion topVer0 = new AffinityTopologyVersion(topVer);
cctx.discovery().topologyFuture(topVer).get();
List<GridDhtPartitionsExchangeFuture> futs = cctx.exchange().exchangeFutures();
DiscoveryEvent evt = null;
long stopTime = System.currentTimeMillis() + 10_000;
boolean primaryChanged = false;
do {
for (int i = futs.size() - 1; i >= 0; i--) {
GridDhtPartitionsExchangeFuture fut = futs.get(i);
if (fut.initialVersion().equals(topVer0)) {
evt = fut.firstEvent();
break;
}
}
if (evt == null) {
U.sleep(500);
futs = cctx.exchange().exchangeFutures();
} else
break;
} while (System.currentTimeMillis() < stopTime);
assertNotNull("Failed to find exchange future:", evt);
Collection<ClusterNode> allNodes = ctx.discovery().serverNodes(topVer0);
for (DynamicCacheDescriptor cacheDesc : ctx.cache().cacheDescriptors().values()) {
if (assignments.get(cacheDesc.cacheId()) != null)
continue;
if (filterByRcvd && cacheDesc.receivedFrom() != null && ctx.discovery().node(topVer0, cacheDesc.receivedFrom()) == null)
continue;
AffinityFunction func = cacheDesc.cacheConfiguration().getAffinity();
func = cctx.cache().clone(func);
cctx.kernalContext().resource().injectGeneric(func);
List<ClusterNode> affNodes = new ArrayList<>();
IgnitePredicate<ClusterNode> filter = cacheDesc.cacheConfiguration().getNodeFilter();
for (ClusterNode n : allNodes) {
if (!n.isClient() && (filter == null || filter.apply(n)))
affNodes.add(n);
}
Collections.sort(affNodes, NodeOrderComparator.getInstance());
AffinityFunctionContext affCtx = new GridAffinityFunctionContextImpl(affNodes, previousAssignment(topVer, cacheDesc.cacheId()), evt, topVer0, cacheDesc.cacheConfiguration().getBackups());
List<List<ClusterNode>> assignment = func.assignPartitions(affCtx);
if (cur != null) {
List<List<ClusterNode>> prev = cur.get(cacheDesc.cacheConfiguration().getName());
assertEquals(prev.size(), assignment.size());
if (!primaryChanged) {
for (int p = 0; p < prev.size(); p++) {
List<ClusterNode> nodes0 = prev.get(p);
List<ClusterNode> nodes1 = assignment.get(p);
if (!nodes0.isEmpty() && !nodes1.isEmpty()) {
ClusterNode p0 = nodes0.get(0);
ClusterNode p1 = nodes1.get(0);
if (allNodes.contains(p0) && !p0.equals(p1)) {
primaryChanged = true;
log.info("Primary changed [cache=" + cacheDesc.cacheConfiguration().getName() + ", part=" + p + ", prev=" + F.nodeIds(nodes0) + ", new=" + F.nodeIds(nodes1) + ']');
break;
}
}
}
}
}
assignments.put(cacheDesc.cacheId(), assignment);
}
return primaryChanged;
}
use of org.apache.ignite.cache.affinity.AffinityFunction in project ignite by apache.
the class IgnitePdsSingleNodeWithIndexingAndGroupPutGetPersistenceSelfTest method configure.
/**
* {@inheritDoc}
*/
@Override
protected void configure(IgniteConfiguration cfg) {
super.configure(cfg);
for (CacheConfiguration ccfg : cfg.getCacheConfiguration()) {
AffinityFunction aff = ccfg.getAffinity();
int parts = aff != null ? aff.partitions() : RendezvousAffinityFunction.DFLT_PARTITION_COUNT;
ccfg.setGroupName("testGroup-parts" + parts);
}
}
use of org.apache.ignite.cache.affinity.AffinityFunction in project ignite by apache.
the class GridCommonAbstractTest method awaitPartitionMapExchange.
/**
* @param waitEvicts If {@code true} will wait for evictions finished.
* @param waitNode2PartUpdate If {@code true} will wait for nodes node2part info update finished.
* @param nodes Optional nodes. If {@code null} method will wait for all nodes, for non null collection nodes will
* be filtered
* @param printPartState If {@code true} will print partition state if evictions not happened.
* @param cacheNames Wait for specific caches.
* @throws InterruptedException If interrupted.
*/
@SuppressWarnings("BusyWait")
protected void awaitPartitionMapExchange(boolean waitEvicts, boolean waitNode2PartUpdate, @Nullable Collection<ClusterNode> nodes, boolean printPartState, @Nullable Set<String> cacheNames) throws InterruptedException {
long timeout = getPartitionMapExchangeTimeout();
long startTime = -1;
Set<String> names = new HashSet<>();
Ignite crd = null;
for (Ignite g : G.allGrids()) {
ClusterNode node = g.cluster().localNode();
if (crd == null || node.order() < crd.cluster().localNode().order()) {
crd = g;
if (node.order() == 1)
break;
}
}
if (crd == null)
return;
AffinityTopologyVersion waitTopVer = ((IgniteKernal) crd).context().discovery().topologyVersionEx();
if (waitTopVer.topologyVersion() <= 0)
waitTopVer = new AffinityTopologyVersion(1, 0);
for (Ignite g : G.allGrids()) {
if (nodes != null && !nodes.contains(g.cluster().localNode()))
continue;
IgniteKernal g0 = (IgniteKernal) g;
names.add(g0.configuration().getIgniteInstanceName());
if (startTime != -1) {
if (startTime != g0.context().discovery().gridStartTime())
fail("Found nodes from different clusters, probable some test does not stop nodes " + "[allNodes=" + names + ']');
} else
startTime = g0.context().discovery().gridStartTime();
if (g.cluster().localNode().isDaemon())
continue;
IgniteInternalFuture<?> exchFut = g0.context().cache().context().exchange().affinityReadyFuture(waitTopVer);
if (exchFut != null && !exchFut.isDone()) {
try {
exchFut.get(timeout);
} catch (IgniteCheckedException e) {
log.error("Failed to wait for exchange [topVer=" + waitTopVer + ", node=" + g0.name() + ']', e);
}
}
for (IgniteCacheProxy<?, ?> c : g0.context().cache().jcaches()) {
CacheConfiguration cfg = c.context().config();
if (cfg == null || cacheNames != null && !cacheNames.contains(cfg.getName()))
continue;
if (cfg.getCacheMode() != LOCAL && cfg.getRebalanceMode() != NONE && g.cluster().nodes().size() > 1) {
AffinityFunction aff = cfg.getAffinity();
GridDhtCacheAdapter<?, ?> dht = dht(c);
GridDhtPartitionTopology top = dht.topology();
for (int p = 0; p < aff.partitions(); p++) {
long start = 0;
for (int i = 0; ; i++) {
boolean match = false;
GridCachePartitionExchangeManager<?, ?> exchMgr = dht.context().shared().exchange();
AffinityTopologyVersion readyVer = exchMgr.readyAffinityVersion();
// Otherwise, there may be an assertion when printing top.readyTopologyVersion().
try {
IgniteInternalFuture<?> fut = exchMgr.affinityReadyFuture(readyVer);
if (fut != null)
fut.get();
} catch (IgniteCheckedException e) {
throw new IgniteException(e);
}
if (readyVer.topologyVersion() > 0 && c.context().started()) {
// Must map on updated version of topology.
List<ClusterNode> affNodes = dht.context().affinity().assignment(readyVer).idealAssignment().get(p);
int affNodesCnt = affNodes.size();
GridDhtTopologyFuture topFut = top.topologyVersionFuture();
Collection<ClusterNode> owners = (topFut != null && topFut.isDone()) ? top.owners(p, AffinityTopologyVersion.NONE) : Collections.<ClusterNode>emptyList();
int ownerNodesCnt = owners.size();
GridDhtLocalPartition loc = top.localPartition(p, readyVer, false);
boolean notPrimary = !affNodes.isEmpty() && !affNodes.get(0).equals(dht.context().affinity().primaryByPartition(p, readyVer));
if (affNodesCnt != ownerNodesCnt || !affNodes.containsAll(owners) || (waitEvicts && loc != null && loc.state() != OWNING) || notPrimary) {
if (i % 50 == 0)
LT.warn(log(), "Waiting for topology map update [" + "igniteInstanceName=" + g.name() + ", cache=" + cfg.getName() + ", cacheId=" + dht.context().cacheId() + ", topVer=" + top.readyTopologyVersion() + ", p=" + p + ", affNodesCnt=" + affNodesCnt + ", ownersCnt=" + ownerNodesCnt + ", affNodes=" + F.nodeIds(affNodes) + ", owners=" + F.nodeIds(owners) + ", topFut=" + topFut + ", locNode=" + g.cluster().localNode() + ']');
} else
match = true;
} else {
if (i % 50 == 0)
LT.warn(log(), "Waiting for topology map update [" + "igniteInstanceName=" + g.name() + ", cache=" + cfg.getName() + ", cacheId=" + dht.context().cacheId() + ", topVer=" + top.readyTopologyVersion() + ", started=" + dht.context().started() + ", p=" + p + ", readVer=" + readyVer + ", locNode=" + g.cluster().localNode() + ']');
}
if (!match) {
if (i == 0)
start = System.currentTimeMillis();
if (System.currentTimeMillis() - start > timeout) {
U.dumpThreads(log);
if (printPartState)
printPartitionState(c);
throw new IgniteException("Timeout of waiting for topology map update [" + "igniteInstanceName=" + g.name() + ", cache=" + cfg.getName() + ", cacheId=" + dht.context().cacheId() + ", topVer=" + top.readyTopologyVersion() + ", p=" + p + ", readVer=" + readyVer + ", locNode=" + g.cluster().localNode() + ']');
}
// Busy wait.
Thread.sleep(20);
continue;
}
if (i > 0)
log().warning("Finished waiting for topology map update [igniteInstanceName=" + g.name() + ", p=" + p + ", duration=" + (System.currentTimeMillis() - start) + "ms]");
break;
}
}
if (waitNode2PartUpdate) {
long start = System.currentTimeMillis();
boolean failed = true;
while (failed) {
failed = false;
for (GridDhtPartitionMap pMap : top.partitionMap(true).values()) {
if (failed)
break;
for (Map.Entry entry : pMap.entrySet()) {
if (System.currentTimeMillis() - start > timeout) {
U.dumpThreads(log);
throw new IgniteException("Timeout of waiting for partition state update [" + "igniteInstanceName=" + g.name() + ", cache=" + cfg.getName() + ", cacheId=" + dht.context().cacheId() + ", topVer=" + top.readyTopologyVersion() + ", locNode=" + g.cluster().localNode() + ']');
}
if (entry.getValue() != OWNING) {
LT.warn(log(), "Waiting for correct partition state part=" + entry.getKey() + ", should be OWNING [state=" + entry.getValue() + "], node=" + g.name() + ", cache=" + c.getName());
// Busy wait.
Thread.sleep(200);
failed = true;
break;
}
}
}
}
}
}
}
}
log.info("awaitPartitionMapExchange finished");
}
Aggregations