Examples with WithSystemProperty org.apache.ignite.testframework.junits.WithSystemProperty used on opensource projects

Example 51 with WithSystemProperty

use of org.apache.ignite.testframework.junits.WithSystemProperty in project ignite by apache.

the class FullHistRebalanceOnClientStopTest method testFullRebalanceNotTriggeredWhenClientNodeStopsDuringPme.

/**
 * @throws Exception if failed.
 */
@Test
@WithSystemProperty(key = IGNITE_PDS_WAL_REBALANCE_THRESHOLD, value = "0")
public void testFullRebalanceNotTriggeredWhenClientNodeStopsDuringPme() throws Exception {
    startGrids(2);
    IgniteEx ig0 = grid(0);
    ig0.cluster().active(true);
    IgniteCache<Object, Object> cache = ig0.cache(CACHE_NAME);
    startClientGrid(5);
    final int entryCnt = PARTS_CNT * 1000;
    final int preloadEntryCnt = PARTS_CNT * 1001;
    for (int i = 0; i < preloadEntryCnt; i++) cache.put(i, i);
    forceCheckpoint();
    stopGrid(1);
    for (int i = 0; i < entryCnt; i++) cache.put(i, i + 100);
    forceCheckpoint();
    final CountDownLatch exchangeLatch = new CountDownLatch(1);
    final CountDownLatch hangingPmeStartedLatch = new CountDownLatch(1);
    ig0.context().cache().context().exchange().registerExchangeAwareComponent(new PartitionsExchangeAware() {

        @Override
        public void onInitAfterTopologyLock(GridDhtPartitionsExchangeFuture fut) {
            try {
                hangingPmeStartedLatch.countDown();
                exchangeLatch.await();
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
        }
    });
    IgniteInternalFuture fut = GridTestUtils.runAsync(() -> {
        try {
            startGrid(1);
            awaitPartitionMapExchange();
        } catch (Exception e) {
            e.printStackTrace();
        }
    });
    IgniteInternalFuture clientStopFut = GridTestUtils.runAsync(() -> {
        try {
            hangingPmeStartedLatch.await();
            stopGrid(5);
        } catch (Exception e) {
            e.printStackTrace();
        }
    }, "client-starter");
    assertFalse(fut.isDone());
    exchangeLatch.countDown();
    clientStopFut.get();
    fut.get();
    awaitPartitionMapExchange();
    boolean histRebalanceInvoked = RebalanceCheckingCommunicationSpi.histRebalances();
    boolean fullRebalanceInvoked = RebalanceCheckingCommunicationSpi.fullRebalances();
    RebalanceCheckingCommunicationSpi.cleanup();
    assertTrue("Historical rebalance hasn't been invoked.", histRebalanceInvoked);
    assertFalse("Full rebalance has been invoked.", fullRebalanceInvoked);
}

Example 52 with WithSystemProperty

use of org.apache.ignite.testframework.junits.WithSystemProperty in project ignite by apache.

the class DeadLockOnNodeLeftExchangeTest method test.

/**
 */
@Test
@WithSystemProperty(key = ExchangeContext.IGNITE_EXCHANGE_COMPATIBILITY_VER_1, value = "true")
public void test() throws Exception {
    startGrids(4);
    IgniteClusterEx cluster = grid(0).cluster();
    cluster.baselineAutoAdjustEnabled(false);
    cluster.active(true);
    TestRecordingCommunicationSpi spi = TestRecordingCommunicationSpi.spi(grid(3));
    spi.blockMessages(GridDhtPartitionsSingleMessage.class, getTestIgniteInstanceName(0));
    stopGrid(1);
    spi.waitForBlocked();
    IgniteInternalFuture setBaselineTopFut = GridTestUtils.runAsync(() -> {
        cluster.setBaselineTopology(cluster.topologyVersion());
    });
    assertTrue(GridTestUtils.waitForCondition(() -> grid(0).context().state().clusterState().transition(), 5_000L));
    stopGrid(2);
    spi.stopBlock();
    setBaselineTopFut.get(30_000L);
}

Example 53 with WithSystemProperty

use of org.apache.ignite.testframework.junits.WithSystemProperty in project ignite by apache.

the class PartitionsExchangeCoordinatorFailoverTest method testChangeCoordinatorToLocallyJoiningNode.

/**
 * Test checks that changing coordinator to a node that joining to cluster at the moment works correctly
 * in case of exchanges merge and completed exchange on other joining nodes.
 */
@Test
@WithSystemProperty(key = IgniteSystemProperties.IGNITE_SKIP_CONFIGURATION_CONSISTENCY_CHECK, value = "true")
public void testChangeCoordinatorToLocallyJoiningNode() throws Exception {
    newCaches = false;
    spiFactory = TestRecordingCommunicationSpi::new;
    IgniteEx crd = startGrid(CRD_NONE);
    final int newCrdNodeIdx = 1;
    // A full message shouldn't be send to new coordinator.
    blockSendingFullMessage(crd, node -> node.consistentId().equals(getTestIgniteInstanceName(newCrdNodeIdx)));
    CountDownLatch joiningNodeSentSingleMsg = new CountDownLatch(1);
    // For next joining node delay sending single message to emulate exchanges merge.
    spiFactory = () -> new DynamicDelayingCommunicationSpi(msg -> {
        final int delay = 5_000;
        if (msg instanceof GridDhtPartitionsSingleMessage) {
            GridDhtPartitionsSingleMessage singleMsg = (GridDhtPartitionsSingleMessage) msg;
            if (singleMsg.exchangeId() != null) {
                joiningNodeSentSingleMsg.countDown();
                return delay;
            }
        }
        return 0;
    });
    IgniteInternalFuture<?> newCrdJoinFut = GridTestUtils.runAsync(() -> startGrid(newCrdNodeIdx));
    // Wait till new coordinator node sent single message.
    joiningNodeSentSingleMsg.await();
    spiFactory = TcpCommunicationSpi::new;
    // Additionally start 2 new nodes. Their exchange should be merged with exchange on join new coordinator node.
    startGridsMultiThreaded(2, 2);
    Assert.assertFalse("New coordinator join shouldn't be happened before stopping old coordinator.", newCrdJoinFut.isDone());
    // Stop coordinator.
    stopGrid(CRD_NONE);
    // New coordinator join process should succeed after that.
    newCrdJoinFut.get();
    awaitPartitionMapExchange();
    // Check that affinity are equal on all nodes.
    AffinityTopologyVersion affVer = ((IgniteEx) ignite(1)).cachex(CACHE_NAME).context().shared().exchange().readyAffinityVersion();
    List<List<ClusterNode>> expAssignment = null;
    IgniteEx expAssignmentNode = null;
    for (Ignite node : G.allGrids()) {
        IgniteEx nodeEx = (IgniteEx) node;
        List<List<ClusterNode>> assignment = nodeEx.cachex(CACHE_NAME).context().affinity().assignments(affVer);
        if (expAssignment == null) {
            expAssignment = assignment;
            expAssignmentNode = nodeEx;
        } else
            Assert.assertEquals("Affinity assignments are different " + "[expectedNode=" + expAssignmentNode + ", actualNode=" + nodeEx + "]", expAssignment, assignment);
    }
}

Example 54 with WithSystemProperty

use of org.apache.ignite.testframework.junits.WithSystemProperty in project ignite by apache.

the class IgniteWalRebalanceTest method testMultipleNodesFailHistoricalRebalance.

/**
 * Tests that demander switches to full rebalance if the previously chosen two of three of suppliers
 * for a group have failed to perform historical rebalance due to an unexpected error.
 *
 * @throws Exception If failed
 */
@Test
@WithSystemProperty(key = "IGNITE_DISABLE_WAL_DURING_REBALANCING", value = "true")
public void testMultipleNodesFailHistoricalRebalance() throws Exception {
    backups = 1;
    int node_cnt = 4;
    int demanderId = node_cnt - 1;
    // Start a new cluster with 3 suppliers.
    startGrids(node_cnt - 1);
    // Start demander node.
    userAttrs.put("TEST_ATTR", "TEST_ATTR");
    startGrid(node_cnt - 1);
    grid(0).cluster().state(ACTIVE);
    // Create a new cache that places a full set of partitions on demander node.
    RendezvousAffinityFunction aff = new RendezvousAffinityFunction(false, PARTS_CNT);
    aff.setAffinityBackupFilter(new ClusterNodeAttributeAffinityBackupFilter("TEST_ATTR"));
    String cacheName = "test-cache-1";
    IgniteCache<Integer, IndexedObject> cache0 = grid(0).getOrCreateCache(new CacheConfiguration<Integer, IndexedObject>(cacheName).setBackups(backups).setAffinity(aff).setWriteSynchronizationMode(CacheWriteSynchronizationMode.FULL_SYNC));
    // Fill initial data and force checkpoint.
    final int entryCnt = PARTS_CNT * 200;
    final int preloadEntryCnt = PARTS_CNT * 201;
    int val = 0;
    for (int k = 0; k < preloadEntryCnt; k++) cache0.put(k, new IndexedObject(val++));
    forceCheckpoint();
    // Stop demander node.
    stopGrid(demanderId);
    // Rewrite data to trigger further rebalance.
    for (int k = 0; k < entryCnt; k++) {
        // even though the corresponding RebalanceFuture will be cancelled.
        if (grid(0).affinity(cacheName).partition(k) != 12)
            cache0.put(k, new IndexedObject(val++));
    }
    // Upload additional data to a particular partition (primary partition belongs to coordinator, for instance)
    // in order to trigger full rebalance for that partition instead of historical one.
    int[] primaries0 = grid(0).affinity(cacheName).primaryPartitions(grid(0).localNode());
    for (int i = 0; i < preloadEntryCnt; ++i) cache0.put(primaries0[0], new IndexedObject(val++));
    forceCheckpoint();
    // Delay rebalance process for specified group.
    blockMsgPred = (node, msg) -> {
        if (msg instanceof GridDhtPartitionDemandMessage) {
            GridDhtPartitionDemandMessage msg0 = (GridDhtPartitionDemandMessage) msg;
            return msg0.groupId() == CU.cacheId(cacheName);
        }
        return false;
    };
    Queue<RecordedDemandMessage> recorderedMsgs = new ConcurrentLinkedQueue<>();
    // Record demand messages for specified group.
    recordMsgPred = (node, msg) -> {
        if (msg instanceof GridDhtPartitionDemandMessage) {
            GridDhtPartitionDemandMessage msg0 = (GridDhtPartitionDemandMessage) msg;
            if (msg0.groupId() == CU.cacheId(cacheName)) {
                recorderedMsgs.add(new RecordedDemandMessage(node.id(), msg0.groupId(), msg0.partitions().hasFull(), msg0.partitions().hasHistorical()));
            }
        }
        return false;
    };
    // Corrupt WAL on suppliers, except the one.
    injectFailingIOFactory(grid(0));
    injectFailingIOFactory(grid(1));
    // Trigger rebalance process from suppliers.
    IgniteEx restartedDemander = startGrid(node_cnt - 1);
    TestRecordingCommunicationSpi demanderSpi = TestRecordingCommunicationSpi.spi(restartedDemander);
    // Wait until demander starts historical rebalancning.
    demanderSpi.waitForBlocked();
    final IgniteInternalFuture<Boolean> preloadFut = restartedDemander.cachex(cacheName).context().group().preloader().rebalanceFuture();
    // Unblock messages and start tracking demand and supply messages.
    demanderSpi.stopBlock();
    // Wait until rebalancing will be cancelled for both suppliers.
    assertTrue("Rebalance future was not cancelled [fut=" + preloadFut + ']', GridTestUtils.waitForCondition(preloadFut::isDone, getTestTimeout()));
    Assert.assertEquals("Rebalance should be cancelled on demander node: " + preloadFut, false, preloadFut.get());
    awaitPartitionMapExchange(true, true, null);
    // Check data consistency.
    assertPartitionsSame(idleVerify(restartedDemander, cacheName));
    // Check that historical rebalance switched to full for supplier 1 & 2 and it was historical for supplier3.
    IgnitePredicate<RecordedDemandMessage> histPred = msg -> msg.hasHistorical() && !msg.hasFull();
    IgnitePredicate<RecordedDemandMessage> fullPred = msg -> !msg.hasHistorical() && msg.hasFull();
    IgnitePredicate<RecordedDemandMessage> mixedPred = msg -> msg.hasHistorical() && msg.hasFull();
    IgniteBiInClosure<UUID, Boolean> supplierChecker = (supplierId, mixed) -> {
        List<RecordedDemandMessage> demandMsgsForSupplier = recorderedMsgs.stream().filter(msg -> msg.supplierId().equals(supplierId)).filter(msg -> msg.groupId() == CU.cacheId(cacheName)).filter(msg -> msg.hasFull() || msg.hasHistorical()).collect(toList());
        assertEquals("There should only two demand messages [supplierId=" + supplierId + ']', 2, demandMsgsForSupplier.size());
        assertTrue("The first message should require " + (mixed ? "mixed" : "historical") + " rebalance [msg=" + demandMsgsForSupplier.get(0) + ']', (mixed ? mixedPred.apply(demandMsgsForSupplier.get(0)) : histPred.apply(demandMsgsForSupplier.get(0))));
        assertTrue("The second message should require full rebalance [msg=" + demandMsgsForSupplier.get(0) + ']', fullPred.apply(demandMsgsForSupplier.get(1)));
    };
    supplierChecker.apply(grid(0).cluster().localNode().id(), true);
    supplierChecker.apply(grid(1).cluster().localNode().id(), false);
    // Check supplier3
    List<RecordedDemandMessage> demandMsgsForSupplier = recorderedMsgs.stream().filter(msg -> msg.supplierId().equals(grid(2).cluster().localNode().id())).filter(msg -> msg.groupId() == CU.cacheId(cacheName)).filter(msg -> msg.hasFull() || msg.hasHistorical()).collect(toList());
    assertEquals("There should only one demand message.", 1, demandMsgsForSupplier.size());
    assertTrue("The first message should require historical rebalance [msg=" + demandMsgsForSupplier.get(0) + ']', histPred.apply(demandMsgsForSupplier.get(0)));
}

Example 55 with WithSystemProperty

use of org.apache.ignite.testframework.junits.WithSystemProperty in project ignite by apache.

the class CheckpointFreeListTest method testRestoreFreeListCorrectlyAfterRandomStop.

/**
 * Note: Test assumes that PDS size didn't change between the first checkpoint and after several node stops.
 * It's not true anymore with free-list caching since the only final free-list state is persisted on checkpoint.
 * Some changed, but currently empty buckets are not persisted and PDS size is smaller after the first checkpoint.
 * Test makes sense only with disabled caching.
 *
 * @throws Exception if fail.
 */
@Test
@WithSystemProperty(key = IgniteSystemProperties.IGNITE_PAGES_LIST_DISABLE_ONHEAP_CACHING, value = "true")
public void testRestoreFreeListCorrectlyAfterRandomStop() throws Exception {
    IgniteEx ignite0 = startGrid(0);
    ignite0.cluster().active(true);
    Random random = new Random();
    List<T2<Integer, byte[]>> cachedEntry = new ArrayList<>();
    IgniteCache<Integer, Object> cache = ignite0.cache(CACHE_NAME);
    for (int j = 0; j < CACHE_SIZE; j++) {
        byte[] val = new byte[random.nextInt(SF.apply(3072))];
        cache.put(j, val);
        cachedEntry.add(new T2<>(j, val));
    }
    Collections.shuffle(cachedEntry);
    // Remove half of entries.
    Collection<T2<Integer, byte[]>> entriesToRemove = cachedEntry.stream().limit(cachedEntry.size() / 2).collect(Collectors.toCollection(ConcurrentLinkedQueue::new));
    entriesToRemove.forEach(t2 -> cache.remove(t2.get1()));
    // During removing of entries free list grab a lot of free pages to itself
    // so will do put/remove again for stabilization of free pages.
    entriesToRemove.forEach(t2 -> cache.put(t2.get1(), t2.get2()));
    entriesToRemove.forEach(t2 -> cache.remove(t2.get1()));
    forceCheckpoint();
    Path cacheFolder = Paths.get(U.defaultWorkDirectory(), DFLT_STORE_DIR, ignite0.name().replaceAll("\\.", "_"), CACHE_DIR_PREFIX + CACHE_NAME);
    Optional<Long> totalPartSizeBeforeStop = totalPartitionsSize(cacheFolder);
    CyclicBarrier nodeStartBarrier = new CyclicBarrier(2);
    int approximateIterationCount = SF.applyLB(10, 6);
    // Approximate count of entries to put per one iteration.
    int iterationDataCount = entriesToRemove.size() / approximateIterationCount;
    startAsyncPutThread(entriesToRemove, nodeStartBarrier);
    // Will stop node during put data several times.
    while (true) {
        stopGrid(0, true);
        ignite0 = startGrid(0);
        ignite0.cluster().active(true);
        if (entriesToRemove.isEmpty())
            break;
        // Notify put thread that node successfully started.
        nodeStartBarrier.await();
        nodeStartBarrier.reset();
        int awaitSize = entriesToRemove.size() - iterationDataCount;
        waitForCondition(() -> entriesToRemove.size() < awaitSize || entriesToRemove.size() == 0, 20000);
    }
    forceCheckpoint();
    Optional<Long> totalPartSizeAfterRestore = totalPartitionsSize(cacheFolder);
    // It allow that size after repeated put operations should be not more than on 15%(heuristic value) greater than before operations.
    // In fact, it should not be multiplied in twice every time.
    long correctedRestoreSize = totalPartSizeAfterRestore.get() - (long) (totalPartSizeBeforeStop.get() * 0.15);
    assertTrue("Size after repeated put operations should be not more than on 15% greater. " + "Size before = " + totalPartSizeBeforeStop.get() + ", Size after = " + totalPartSizeAfterRestore.get(), totalPartSizeBeforeStop.get() > correctedRestoreSize);
}