@Test
public void testUpdate() throws Exception {
ZooKeeper zk = getClient(0);
ZooKeeper zk2 = getClient(0);
MailboxTracker tracker = new MailboxTracker(zk, handler);
MailboxPublisher publisher = new MailboxPublisher(VoltZK.mailboxes + "/1");
VoltZK.createPersistentZKNodes(zk);
publisher.registerMailbox(MailboxType.ExecutionSite, new MailboxNodeContent(1L, 0));
publisher.publish(zk2);
publisher = new MailboxPublisher(VoltZK.mailboxes + "/2");
publisher.registerMailbox(MailboxType.ExecutionSite, new MailboxNodeContent(2L, 1));
publisher.publish(zk);
tracker.start();
// The ephemaral node just created will disappear and we should get an update
zk2.close();
while (handler.m_handleCount.get() < 2) {
Thread.sleep(1);
}
Map<MailboxType, List<MailboxNodeContent>> value = handler.m_mailboxes;
assertTrue(value.containsKey(MailboxType.ExecutionSite));
List<MailboxNodeContent> list = value.get(MailboxType.ExecutionSite);
assertEquals(1, list.size());
assertEquals(2, list.get(0).HSId.longValue());
assertEquals(1, list.get(0).partitionId.intValue());
tracker.shutdown();
}
/**
* Rebuild the point-in-time snapshot of the children objects and set watches on new
* children. @Param event may be null on the first initialization.
*/
private void processParentEvent(WatchedEvent event) throws Exception {
// get current children snapshot and reset this watch.
Set<String> children = new TreeSet<String>(m_zk.getChildren(m_rootNode, m_parentWatch));
// intersect to get newChildren and update m_lastChildren to the current set.
Set<String> newChildren = new HashSet<String>(children);
newChildren.removeAll(m_lastChildren);
m_lastChildren = children;
List<ByteArrayCallback> callbacks = new ArrayList<ByteArrayCallback>();
for (String child : children) {
ByteArrayCallback cb = new ByteArrayCallback();
// set watches on new children.
if (newChildren.contains(child)) {
m_zk.getData(ZKUtil.joinZKPath(m_rootNode, child), m_childWatch, cb, null);
} else {
m_zk.getData(ZKUtil.joinZKPath(m_rootNode, child), false, cb, null);
}
callbacks.add(cb);
}
HashMap<String, JSONObject> cache = new HashMap<String, JSONObject>();
for (ByteArrayCallback callback : callbacks) {
try {
byte payload[] = callback.getData();
JSONObject jsObj = new JSONObject(new String(payload, "UTF-8"));
cache.put(callback.getPath(), jsObj);
} catch (KeeperException.NoNodeException e) {
// child may have been deleted between the parent trigger and getData.
}
}
m_publicCache.set(ImmutableMap.copyOf(cache));
}
/*
* Inherit the per partition txnid from the long since gone
* partition that existed in the past
*/
private long[] fetchPerPartitionTxnId() {
ZooKeeper zk = VoltDB.instance().getHostMessenger().getZK();
byte partitionTxnIdsBytes[] = null;
try {
partitionTxnIdsBytes = zk.getData(VoltZK.perPartitionTxnIds, false, null);
} catch (KeeperException.NoNodeException e) {
return null;
} // Can be no node if the cluster was never restored
catch (Exception e) {
VoltDB.crashLocalVoltDB("Error retrieving per partition txn ids", true, e);
}
ByteBuffer buf = ByteBuffer.wrap(partitionTxnIdsBytes);
int count = buf.getInt();
Long partitionTxnId = null;
long partitionTxnIds[] = new long[count];
for (int ii = 0; ii < count; ii++) {
long txnId = buf.getLong();
partitionTxnIds[ii] = txnId;
int partitionId = TxnEgo.getPartitionId(txnId);
if (partitionId == m_partitionId) {
partitionTxnId = txnId;
continue;
}
}
if (partitionTxnId != null) {
return partitionTxnIds;
}
return null;
}
@Test
public void testDeleteChildWithCallback() throws Exception {
ZooKeeper zk = getClient(0);
configure("/cache02", zk);
TestCallback cb = new TestCallback();
LeaderCache dut = new LeaderCache(zk, "/cache02", cb);
dut.start(true);
Map<Integer, Long> cache = cb.m_cache;
assertEquals("3 items cached.", 3, cache.size());
zk.delete("/cache02/1", -1);
while (true) {
cache = cb.m_cache;
if (cache.size() == 3) {
Thread.sleep(1);
} else {
break;
}
}
assertEquals("Item removed", 2, cache.size());
assertEquals(null, cache.get(1));
assertEquals(12345678, cache.get(0).longValue());
assertEquals(11223344, cache.get(2).longValue());
dut.shutdown();
zk.close();
}
@Test
public void testModifyChildWithCallback() throws Exception {
ZooKeeper zk = getClient(0);
configure("/cache03", zk);
TestCallback cb = new TestCallback();
LeaderCache dut = new LeaderCache(zk, "/cache03", cb);
dut.start(true);
Map<Integer, Long> cache = cb.m_cache;
assertEquals("3 items cached.", 3, cache.size());
assertEquals(12345678, cache.get(0).longValue());
dut.put(0, 23456789);
while (true) {
cache = cb.m_cache;
if (cache.get(0) == 23456789) {
break;
}
}
cache = cb.m_cache;
assertEquals("3 items cached.", 3, cache.size());
assertEquals(23456789, cache.get(0).longValue());
assertEquals(87654321, cache.get(1).longValue());
assertEquals(11223344, cache.get(2).longValue());
dut.shutdown();
zk.close();
}
@Test
public void testModifyChild() throws Exception {
ZooKeeper zk = getClient(0);
configure("/cache03", zk);
LeaderCache dut = new LeaderCache(zk, "/cache03");
dut.start(true);
Map<Integer, Long> cache = dut.pointInTimeCache();
assertEquals("3 items cached.", 3, cache.size());
assertEquals(12345678, dut.get(0).longValue());
zk.setData("/cache03/0", Long.toString(23456789).getBytes(), -1);
while (true) {
if (dut.get(0) == 23456789) {
break;
}
}
assertEquals("3 items cached.", 3, cache.size());
assertEquals(23456789L, dut.get(0).longValue());
assertEquals(87654321L, dut.get(1).longValue());
assertEquals(11223344L, dut.get(2).longValue());
dut.shutdown();
zk.close();
}
public static void deleteRecursively(ZooKeeper zk, String dir)
throws KeeperException, InterruptedException {
List<String> children = zk.getChildren(dir, false);
for (String child : children) {
deleteRecursively(zk, joinZKPath(dir, child));
}
zk.delete(dir, -1);
}
void configure(String root, ZooKeeper zk) throws Exception {
Long aa = 12345678L;
Long bb = 87654321L;
Long cc = 11223344L;
zk.create(root, new byte[] {}, Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
zk.create(root + "/0", aa.toString().getBytes(), Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
zk.create(root + "/1", bb.toString().getBytes(), Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
zk.create(root + "/2", cc.toString().getBytes(), Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
}
@Override
public void run() {
try {
JSONStringer js = new JSONStringer();
js.object();
js.key("role").value(m_config.m_replicationRole.ordinal());
js.key("active").value(m_rvdb.getReplicationActive());
js.endObject();
ZooKeeper zk = m_rvdb.getHostMessenger().getZK();
// rejoining nodes figure out the replication role from other nodes
if (!m_isRejoin) {
try {
zk.create(
VoltZK.replicationconfig,
js.toString().getBytes("UTF-8"),
Ids.OPEN_ACL_UNSAFE,
CreateMode.PERSISTENT);
} catch (KeeperException.NodeExistsException e) {
}
String discoveredReplicationConfig =
new String(zk.getData(VoltZK.replicationconfig, false, null), "UTF-8");
JSONObject discoveredjsObj = new JSONObject(discoveredReplicationConfig);
ReplicationRole discoveredRole =
ReplicationRole.get((byte) discoveredjsObj.getLong("role"));
if (!discoveredRole.equals(m_config.m_replicationRole)) {
VoltDB.crashGlobalVoltDB(
"Discovered replication role "
+ discoveredRole
+ " doesn't match locally specified replication role "
+ m_config.m_replicationRole,
true,
null);
}
// See if we should bring the server up in WAN replication mode
m_rvdb.setReplicationRole(discoveredRole);
} else {
String discoveredReplicationConfig =
new String(zk.getData(VoltZK.replicationconfig, false, null), "UTF-8");
JSONObject discoveredjsObj = new JSONObject(discoveredReplicationConfig);
ReplicationRole discoveredRole =
ReplicationRole.get((byte) discoveredjsObj.getLong("role"));
boolean replicationActive = discoveredjsObj.getBoolean("active");
// See if we should bring the server up in WAN replication mode
m_rvdb.setReplicationRole(discoveredRole);
m_rvdb.setReplicationActive(replicationActive);
}
} catch (Exception e) {
VoltDB.crashGlobalVoltDB("Error discovering replication role", false, e);
}
}
public static ZKUtil.StringCallback asyncMkdirs(ZooKeeper zk, String dirDN, byte payload[]) {
Preconditions.checkArgument(
dirDN != null && !dirDN.trim().isEmpty() && !"/".equals(dirDN) && dirDN.startsWith("/"));
StringBuilder dsb = new StringBuilder(128);
ZKUtil.StringCallback lastCallback = null;
try {
String dirPortions[] = dirDN.substring(1).split("/");
for (int ii = 0; ii < dirPortions.length; ii++) {
String dirPortion = dirPortions[ii];
lastCallback = new ZKUtil.StringCallback();
dsb.append('/').append(dirPortion);
zk.create(
dsb.toString(),
ii == dirPortions.length - 1 ? payload : null,
Ids.OPEN_ACL_UNSAFE,
CreateMode.PERSISTENT,
lastCallback,
null);
}
} catch (Throwable t) {
Throwables.propagate(t);
}
return lastCallback;
}
/**
* Given a set of partition IDs, return a map of partition to a list of HSIDs of all the sites
* with copies of each partition
*/
public Map<Integer, List<Long>> getReplicasForPartitions(Collection<Integer> partitions) {
Map<Integer, List<Long>> retval = new HashMap<Integer, List<Long>>();
List<Pair<Integer, ZKUtil.ChildrenCallback>> callbacks =
new ArrayList<Pair<Integer, ZKUtil.ChildrenCallback>>();
for (Integer partition : partitions) {
String zkpath = LeaderElector.electionDirForPartition(partition);
ZKUtil.ChildrenCallback cb = new ZKUtil.ChildrenCallback();
callbacks.add(Pair.of(partition, cb));
m_zk.getChildren(zkpath, false, cb, null);
}
for (Pair<Integer, ZKUtil.ChildrenCallback> p : callbacks) {
final Integer partition = p.getFirst();
try {
List<String> children = p.getSecond().getChildren();
List<Long> sites = new ArrayList<Long>();
for (String child : children) {
sites.add(Long.valueOf(child.split("_")[0]));
}
retval.put(partition, sites);
} catch (KeeperException ke) {
org.voltdb.VoltDB.crashLocalVoltDB(
"KeeperException getting replicas for partition: " + partition, true, ke);
} catch (InterruptedException ie) {
org.voltdb.VoltDB.crashLocalVoltDB(
"InterruptedException getting replicas for partition: " + partition, true, ie);
}
}
return retval;
}
@Test
public void testInitialCacheWithCallback() throws Exception {
ZooKeeper zk = getClient(0);
configure("/cache01", zk);
TestCallback cb = new TestCallback();
LeaderCache dut = new LeaderCache(zk, "/cache01", cb);
dut.start(true);
assertEquals("3 items cached.", 3, cb.m_cache.size());
assertEquals(12345678, cb.m_cache.get(0).longValue());
assertEquals(87654321, cb.m_cache.get(1).longValue());
assertEquals(11223344, cb.m_cache.get(2).longValue());
dut.shutdown();
zk.close();
}
@Test
public void testInitialCache() throws Exception {
ZooKeeper zk = getClient(0);
configure("/cache01", zk);
LeaderCache dut = new LeaderCache(zk, "/cache01");
dut.start(true);
Map<Integer, Long> cache = dut.pointInTimeCache();
assertEquals("3 items cached.", 3, cache.size());
assertEquals(12345678L, dut.get(0).longValue());
assertEquals(87654321L, dut.get(1).longValue());
assertEquals(11223344L, dut.get(2).longValue());
dut.shutdown();
zk.close();
}
/**
* Once participating host count is set, SnapshotCompletionMonitor can check this ZK node to
* determine whether the snapshot has finished or not.
*
* <p>This should only be called when all participants have responded. It is possible that some
* hosts finish taking snapshot before the coordinator logs the participating host count. In this
* case, the host count would have been decremented multiple times already. To make sure finished
* hosts are logged correctly, this method adds participating host count + 1 to the current host
* count.
*
* @param txnId The snapshot txnId
* @param participantCount The number of hosts participating in this snapshot
*/
public static void logParticipatingHostCount(long txnId, int participantCount) {
ZooKeeper zk = VoltDB.instance().getHostMessenger().getZK();
final String snapshotPath = VoltZK.completed_snapshots + "/" + txnId;
boolean success = false;
while (!success) {
Stat stat = new Stat();
byte data[] = null;
try {
data = zk.getData(snapshotPath, false, stat);
} catch (KeeperException e) {
if (e.code() == KeeperException.Code.NONODE) {
// If snapshot creation failed for some reason, the node won't exist. ignore
return;
}
VoltDB.crashLocalVoltDB("Failed to get snapshot completion node", true, e);
} catch (InterruptedException e) {
VoltDB.crashLocalVoltDB("Interrupted getting snapshot completion node", true, e);
}
if (data == null) {
VoltDB.crashLocalVoltDB("Data should not be null if the node exists", false, null);
}
try {
JSONObject jsonObj = new JSONObject(new String(data, Charsets.UTF_8));
if (jsonObj.getLong("txnId") != txnId) {
VoltDB.crashLocalVoltDB("TxnId should match", false, null);
}
int hostCount = jsonObj.getInt("hostCount");
// +1 because hostCount was initialized to -1
jsonObj.put("hostCount", hostCount + participantCount + 1);
zk.setData(snapshotPath, jsonObj.toString(4).getBytes(Charsets.UTF_8), stat.getVersion());
} catch (KeeperException.BadVersionException e) {
continue;
} catch (Exception e) {
VoltDB.crashLocalVoltDB("This ZK call should never fail", true, e);
}
success = true;
}
}
public static void uploadBytesAsChunks(
ZooKeeper zk, String node, byte payload[], boolean ephemeral) throws Exception {
ByteBuffer buffer = ByteBuffer.wrap(compressBytes(payload));
while (buffer.hasRemaining()) {
int nextChunkSize = Math.min(1024 * 1024, buffer.remaining());
ByteBuffer nextChunk = ByteBuffer.allocate(nextChunkSize);
buffer.limit(buffer.position() + nextChunkSize);
nextChunk.put(buffer);
buffer.limit(buffer.capacity());
zk.create(
node,
nextChunk.array(),
Ids.OPEN_ACL_UNSAFE,
ephemeral ? CreateMode.EPHEMERAL_SEQUENTIAL : CreateMode.PERSISTENT_SEQUENTIAL);
}
zk.create(
node + "_complete",
null,
Ids.OPEN_ACL_UNSAFE,
ephemeral ? CreateMode.EPHEMERAL : CreateMode.PERSISTENT);
}
@Test
public void testAddChildWithPutWithCallback() throws Exception {
ZooKeeper zk = getClient(0);
configure("/cache04", zk);
TestCallback cb = new TestCallback();
LeaderCache dut = new LeaderCache(zk, "/cache04", cb);
dut.start(true);
Map<Integer, Long> cache = cb.m_cache;
dut.put(3, 88776655);
while (true) {
cache = cb.m_cache;
if (cache.size() == 3) {
Thread.sleep(1);
} else {
break;
}
}
assertEquals("Item added", 4, cache.size());
assertEquals(12345678, cache.get(0).longValue());
assertEquals(87654321, cache.get(1).longValue());
assertEquals(11223344, cache.get(2).longValue());
assertEquals(88776655, cache.get(3).longValue());
// modify the new child and make sure it has a watch set.
dut.put(3, 99887766);
while (true) {
cache = cb.m_cache;
if (cache.get(3) == 99887766) {
break;
}
}
assertEquals("Items accounted for.", 4, cache.size());
assertEquals(99887766, cache.get(3).longValue());
dut.shutdown();
zk.close();
}
private void writeKnownLiveNodes(List<Integer> liveNodes) {
try {
if (m_zk.exists(VoltZK.lastKnownLiveNodes, null) == null) {
// VoltZK.createPersistentZKNodes should have done this
m_zk.create(VoltZK.lastKnownLiveNodes, null, Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
}
JSONStringer stringer = new JSONStringer();
stringer.object();
stringer.key("liveNodes").array();
for (Integer node : liveNodes) {
stringer.value(node);
}
stringer.endArray();
stringer.endObject();
JSONObject obj = new JSONObject(stringer.toString());
tmLog.debug("Writing live nodes to ZK: " + obj.toString(4));
m_zk.setData(VoltZK.lastKnownLiveNodes, obj.toString(4).getBytes("UTF-8"), -1);
} catch (Exception e) {
VoltDB.crashLocalVoltDB(
"Unable to update known live nodes at ZK path: " + VoltZK.lastKnownLiveNodes, true, e);
}
}
/**
* Update a modified child and republish a new snapshot. This may indicate a deleted child or a
* child with modified data.
*/
private void processChildEvent(WatchedEvent event) throws Exception {
HashMap<String, JSONObject> cacheCopy = new HashMap<String, JSONObject>(m_publicCache.get());
ByteArrayCallback cb = new ByteArrayCallback();
m_zk.getData(event.getPath(), m_childWatch, cb, null);
try {
byte payload[] = cb.getData();
JSONObject jsObj = new JSONObject(new String(payload, "UTF-8"));
cacheCopy.put(cb.getPath(), jsObj);
} catch (KeeperException.NoNodeException e) {
cacheCopy.remove(event.getPath());
}
m_publicCache.set(ImmutableMap.copyOf(cacheCopy));
}
/**
* Returns the IDs of the partitions currently in the cluster.
*
* @return A list of partition IDs
*/
public static List<Integer> getPartitions(ZooKeeper zk) {
List<Integer> partitions = new ArrayList<Integer>();
try {
List<String> children = zk.getChildren(VoltZK.leaders_initiators, null);
for (String child : children) {
partitions.add(LeaderElector.getPartitionFromElectionDir(child));
}
} catch (KeeperException e) {
VoltDB.crashLocalVoltDB("Failed to get partition IDs from ZK", true, e);
} catch (InterruptedException e) {
VoltDB.crashLocalVoltDB("Failed to get partition IDs from ZK", true, e);
}
return partitions;
}
public static Pair<byte[], Integer> retrieveChunksAsBytes(
ZooKeeper zk, String path, String prefix, boolean getCRC) throws Exception {
TreeSet<String> chunks = new TreeSet<String>();
while (true) {
boolean allUploadsComplete = true;
if (!chunks.contains(path + "/" + prefix + "_complete")) {
allUploadsComplete = false;
}
if (allUploadsComplete) {
break;
}
chunks = new TreeSet<String>(zk.getChildren(path, false));
for (String chunk : chunks) {
for (int ii = 0; ii < chunks.size(); ii++) {
if (chunk.startsWith(path + "/" + prefix)) {
chunks.add(chunk);
}
}
}
}
byte resultBuffers[][] = new byte[chunks.size() - 1][];
int ii = 0;
PureJavaCrc32 crc = getCRC ? new PureJavaCrc32() : null;
for (String chunk : chunks) {
if (chunk.endsWith("_complete")) continue;
resultBuffers[ii] = zk.getData(chunk, false, null);
if (crc != null) {
crc.update(resultBuffers[ii]);
}
ii++;
}
return Pair.of(decompressBytes(resultBuffers), crc != null ? (int) crc.getValue() : null);
}
private boolean isClusterKSafe() {
boolean retval = true;
for (int i = 0; i < m_partitionCount; i++) {
String dir = LeaderElector.electionDirForPartition(i);
try {
List<String> replicas = m_zk.getChildren(dir, null, null);
if (replicas.isEmpty()) {
tmLog.fatal("K-Safety violation: No replicas found for partition: " + i);
retval = false;
}
} catch (Exception e) {
VoltDB.crashLocalVoltDB("Unable to read replicas in ZK dir: " + dir, true, e);
}
}
return retval;
}
/** Given a partition ID, return a list of HSIDs of all the sites with copies of that partition */
public List<Long> getReplicasForPartition(int partition) {
String zkpath = LeaderElector.electionDirForPartition(partition);
List<Long> retval = new ArrayList<Long>();
try {
List<String> children = m_zk.getChildren(zkpath, null);
for (String child : children) {
retval.add(Long.valueOf(child.split("_")[0]));
}
} catch (KeeperException ke) {
org.voltdb.VoltDB.crashLocalVoltDB(
"KeeperException getting replicas for partition: " + partition, true, ke);
} catch (InterruptedException ie) {
org.voltdb.VoltDB.crashLocalVoltDB(
"InterruptedException getting replicas for partition: " + partition, true, ie);
}
return retval;
}
private Set<Integer> readPriorKnownLiveNodes() {
Set<Integer> nodes = new HashSet<Integer>();
try {
byte[] data = m_zk.getData(VoltZK.lastKnownLiveNodes, false, null);
String jsonString = new String(data, "UTF-8");
tmLog.debug("Read prior known live nodes: " + jsonString);
JSONObject jsObj = new JSONObject(jsonString);
JSONArray jsonNodes = jsObj.getJSONArray("liveNodes");
for (int ii = 0; ii < jsonNodes.length(); ii++) {
nodes.add(jsonNodes.getInt(ii));
}
} catch (Exception e) {
VoltDB.crashLocalVoltDB(
"Unable to read prior known live nodes at ZK path: " + VoltZK.lastKnownLiveNodes,
true,
e);
}
return nodes;
}
@Override
public void acceptPromotion() throws InterruptedException, ExecutionException, KeeperException {
// Crank up the leader caches. Use blocking startup so that we'll have valid point-in-time
// caches later.
m_iv2appointees.start(true);
m_iv2masters.start(true);
// Figure out what conditions we assumed leadership under.
if (m_iv2appointees.pointInTimeCache().size() == 0) {
tmLog.debug("LeaderAppointer in startup");
m_state.set(AppointerState.CLUSTER_START);
} else if ((m_iv2appointees.pointInTimeCache().size() != m_partitionCount)
|| (m_iv2masters.pointInTimeCache().size() != m_partitionCount)) {
// If we are promoted and the appointees or masters set is partial, the previous appointer
// failed
// during startup (at least for now, until we add add/remove a partition on the fly).
VoltDB.crashGlobalVoltDB("Detected failure during startup, unable to start", false, null);
} else {
tmLog.debug("LeaderAppointer in repair");
m_state.set(AppointerState.DONE);
}
if (m_state.get() == AppointerState.CLUSTER_START) {
// Need to block the return of acceptPromotion until after the MPI is promoted. Wait for this
// latch
// to countdown after appointing all the partition leaders. The
// LeaderCache callback will count it down once it has seen all the
// appointed leaders publish themselves as the actual leaders.
m_startupLatch = new CountDownLatch(1);
writeKnownLiveNodes(m_hostMessenger.getLiveHostIds());
for (int i = 0; i < m_partitionCount; i++) {
String dir = LeaderElector.electionDirForPartition(i);
// Race along with all of the replicas for this partition to create the ZK parent node
try {
m_zk.create(dir, null, Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
} catch (KeeperException.NodeExistsException e) {
// expected on all nodes that don't start() first.
}
m_callbacks[i] = new PartitionCallback(i);
Pair<BabySitter, List<String>> sitterstuff =
BabySitter.blockingFactory(m_zk, dir, m_callbacks[i], m_es);
m_partitionWatchers[i] = sitterstuff.getFirst();
}
m_startupLatch.await();
} else {
// If we're taking over for a failed LeaderAppointer, we know when
// we get here that every partition had a leader at some point in
// time. We'll seed each of the PartitionCallbacks for each
// partition with the HSID of the last published leader. The
// blocking startup of the BabySitter watching that partition will
// call our callback, get the current full set of replicas, and
// appoint a new leader if the seeded one has actually failed
Map<Integer, Long> masters = m_iv2masters.pointInTimeCache();
tmLog.info("LeaderAppointer repairing with master set: " + masters);
for (Entry<Integer, Long> master : masters.entrySet()) {
int partId = master.getKey();
String dir = LeaderElector.electionDirForPartition(partId);
m_callbacks[partId] = new PartitionCallback(partId, master.getValue());
Pair<BabySitter, List<String>> sitterstuff =
BabySitter.blockingFactory(m_zk, dir, m_callbacks[partId], m_es);
m_partitionWatchers[partId] = sitterstuff.getFirst();
}
// just go ahead and promote our MPI
m_MPI.acceptPromotion();
}
}
private void handleChildUpdate(final WatchedEvent event) {
m_zk.getChildren(
event.getPath(), constructMailboxChildWatcher(), constructChildRetrievalCallback(), null);
}
private void createAndRegisterAckMailboxes(
final Set<Integer> localPartitions, HostMessenger messenger) {
m_zk = messenger.getZK();
m_mailboxesZKPath = VoltZK.exportGenerations + "/" + m_timestamp + "/" + "mailboxes";
m_mbox =
new LocalMailbox(messenger) {
@Override
public void deliver(VoltMessage message) {
if (message instanceof BinaryPayloadMessage) {
BinaryPayloadMessage bpm = (BinaryPayloadMessage) message;
ByteBuffer buf = ByteBuffer.wrap(bpm.m_payload);
final int partition = buf.getInt();
final int length = buf.getInt();
byte stringBytes[] = new byte[length];
buf.get(stringBytes);
String signature = new String(stringBytes, Constants.UTF8ENCODING);
final long ackUSO = buf.getLong();
final HashMap<String, ExportDataSource> partitionSources =
m_dataSourcesByPartition.get(partition);
if (partitionSources == null) {
exportLog.error(
"Received an export ack for partition "
+ partition
+ " which does not exist on this node");
return;
}
final ExportDataSource eds = partitionSources.get(signature);
if (eds == null) {
exportLog.error(
"Received an export ack for partition "
+ partition
+ " source signature "
+ signature
+ " which does not exist on this node");
return;
}
try {
eds.ack(ackUSO);
} catch (RejectedExecutionException ignoreIt) {
// ignore it: as it is already shutdown
}
} else {
exportLog.error("Receive unexpected message " + message + " in export subsystem");
}
}
};
messenger.createMailbox(null, m_mbox);
for (Integer partition : localPartitions) {
final String partitionDN = m_mailboxesZKPath + "/" + partition;
ZKUtil.asyncMkdirs(m_zk, partitionDN);
ZKUtil.StringCallback cb = new ZKUtil.StringCallback();
m_zk.create(
partitionDN + "/" + m_mbox.getHSId(),
null,
Ids.OPEN_ACL_UNSAFE,
CreateMode.EPHEMERAL,
cb,
null);
}
ListenableFuture<?> fut =
m_childUpdatingThread.submit(
new Runnable() {
@Override
public void run() {
List<Pair<Integer, ZKUtil.ChildrenCallback>> callbacks =
new ArrayList<Pair<Integer, ZKUtil.ChildrenCallback>>();
for (Integer partition : localPartitions) {
ZKUtil.ChildrenCallback callback = new ZKUtil.ChildrenCallback();
m_zk.getChildren(
m_mailboxesZKPath + "/" + partition,
constructMailboxChildWatcher(),
callback,
null);
callbacks.add(Pair.of(partition, callback));
}
for (Pair<Integer, ZKUtil.ChildrenCallback> p : callbacks) {
final Integer partition = p.getFirst();
List<String> children = null;
try {
children = p.getSecond().getChildren();
} catch (InterruptedException e) {
Throwables.propagate(e);
} catch (KeeperException e) {
Throwables.propagate(e);
}
ImmutableList.Builder<Long> mailboxes = ImmutableList.builder();
for (String child : children) {
if (child.equals(Long.toString(m_mbox.getHSId()))) continue;
mailboxes.add(Long.valueOf(child));
}
ImmutableList<Long> mailboxHsids = mailboxes.build();
for (ExportDataSource eds : m_dataSourcesByPartition.get(partition).values()) {
eds.updateAckMailboxes(Pair.of(m_mbox, mailboxHsids));
}
}
}
});
try {
fut.get();
} catch (Throwable t) {
Throwables.propagate(t);
}
}
