/**
* Creates or force updates an unassigned node to the OFFLINE state for the specified region.
*
* <p>Attempts to create the node but if it exists will force it to transition to and OFFLINE
* state.
*
* <p>Sets a watcher on the unassigned region node if the method is successful.
*
* <p>This method should be used when assigning a region.
*
* @param zkw zk reference
* @param region region to be created as offline
* @param serverName server event originates from
* @throws KeeperException if unexpected zookeeper exception
* @throws KeeperException.NodeExistsException if node already exists
*/
public static boolean createOrForceNodeOffline(
ZooKeeperWatcher zkw, HRegionInfo region, String serverName) throws KeeperException {
LOG.debug(
zkw.prefix(
"Creating (or updating) unassigned node for "
+ region.getEncodedName()
+ " with OFFLINE state"));
RegionTransitionData data =
new RegionTransitionData(EventType.M_ZK_REGION_OFFLINE, region.getRegionName(), serverName);
synchronized (zkw.getNodes()) {
String node = getNodeName(zkw, region.getEncodedName());
zkw.sync(node);
zkw.getNodes().add(node);
int version = ZKUtil.checkExists(zkw, node);
if (version == -1) {
ZKUtil.createAndWatch(zkw, node, data.getBytes());
} else {
if (!ZKUtil.setData(zkw, node, data.getBytes(), version)) {
return false;
} else {
// We successfully forced to OFFLINE, reset watch and handle if
// the state changed in between our set and the watch
RegionTransitionData curData = ZKAssign.getData(zkw, region.getEncodedName());
if (curData.getEventType() != data.getEventType()) {
// state changed, need to process
return false;
}
}
}
}
return true;
}
public static void createNodeOffline(
ZooKeeperWatcher zkw, HRegionInfo region, String serverName, final EventType event)
throws KeeperException, KeeperException.NodeExistsException {
LOG.debug(
zkw.prefix(
"Creating unassigned node for " + region.getEncodedName() + " in OFFLINE state"));
RegionTransitionData data = new RegionTransitionData(event, region.getRegionName(), serverName);
synchronized (zkw.getNodes()) {
String node = getNodeName(zkw, region.getEncodedName());
zkw.getNodes().add(node);
ZKUtil.createAndWatch(zkw, node, data.getBytes());
}
}
/**
* Forces an existing unassigned node to the OFFLINE state for the specified region.
*
* <p>Does not create a new node. If a node does not already exist for this region, a {@link
* NoNodeException} will be thrown.
*
* <p>Sets a watcher on the unassigned region node if the method is successful.
*
* <p>This method should only be used during recovery of regionserver failure.
*
* @param zkw zk reference
* @param region region to be forced as offline
* @param serverName server event originates from
* @throws KeeperException if unexpected zookeeper exception
* @throws KeeperException.NoNodeException if node does not exist
*/
public static void forceNodeOffline(ZooKeeperWatcher zkw, HRegionInfo region, String serverName)
throws KeeperException, KeeperException.NoNodeException {
LOG.debug(
zkw.prefix(
"Forcing existing unassigned node for "
+ region.getEncodedName()
+ " to OFFLINE state"));
RegionTransitionData data =
new RegionTransitionData(EventType.M_ZK_REGION_OFFLINE, region.getRegionName(), serverName);
synchronized (zkw.getNodes()) {
String node = getNodeName(zkw, region.getEncodedName());
zkw.getNodes().add(node);
ZKUtil.setData(zkw, node, data.getBytes());
}
}
/**
* Creates a new unassigned node in the CLOSING state for the specified region.
*
* <p>Does not transition nodes from any states. If a node already exists for this region, a
* {@link NodeExistsException} will be thrown.
*
* <p>If creation is successful, returns the version number of the CLOSING node created.
*
* <p>Does not set any watches.
*
* <p>This method should only be used by a RegionServer when initiating a close of a region after
* receiving a CLOSE RPC from the Master.
*
* @param zkw zk reference
* @param region region to be created as closing
* @param serverName server event originates from
* @return version of node after transition, -1 if unsuccessful transition
* @throws KeeperException if unexpected zookeeper exception
* @throws KeeperException.NodeExistsException if node already exists
*/
public static int createNodeClosing(ZooKeeperWatcher zkw, HRegionInfo region, String serverName)
throws KeeperException, KeeperException.NodeExistsException {
LOG.debug(
zkw.prefix(
"Creating unassigned node for " + region.getEncodedName() + " in a CLOSING state"));
RegionTransitionData data =
new RegionTransitionData(
EventType.RS_ZK_REGION_CLOSING, region.getRegionName(), serverName);
synchronized (zkw.getNodes()) {
String node = getNodeName(zkw, region.getEncodedName());
zkw.getNodes().add(node);
return ZKUtil.createAndWatch(zkw, node, data.getBytes());
}
}
/**
* Deletes an existing unassigned node that is in the specified state for the specified region.
*
* <p>If a node does not already exist for this region, a {@link NoNodeException} will be thrown.
*
* <p>No watcher is set whether this succeeds or not.
*
* <p>Returns false if the node was not in the proper state but did exist.
*
* <p>This method is used during table disables when a region finishes successfully closing. This
* is the Master acknowledging completion of the specified regions transition to being closed.
*
* @param zkw zk reference
* @param regionName region to be deleted from zk
* @param expectedState state region must be in for delete to complete
* @param expectedVersion of the znode that is to be deleted. If expectedVersion need not be
* compared while deleting the znode pass -1
* @throws KeeperException if unexpected zookeeper exception
* @throws KeeperException.NoNodeException if node does not exist
*/
public static boolean deleteNode(
ZooKeeperWatcher zkw, String regionName, EventType expectedState, int expectedVersion)
throws KeeperException, KeeperException.NoNodeException {
LOG.debug(
zkw.prefix(
"Deleting existing unassigned "
+ "node for "
+ regionName
+ " that is in expected state "
+ expectedState));
String node = getNodeName(zkw, regionName);
zkw.sync(node);
Stat stat = new Stat();
byte[] bytes = ZKUtil.getDataNoWatch(zkw, node, stat);
if (bytes == null) {
// If it came back null, node does not exist.
throw KeeperException.create(Code.NONODE);
}
RegionTransitionData data = RegionTransitionData.fromBytes(bytes);
if (!data.getEventType().equals(expectedState)) {
LOG.warn(
zkw.prefix(
"Attempting to delete unassigned "
+ "node "
+ regionName
+ " in "
+ expectedState
+ " state but node is in "
+ data.getEventType()
+ " state"));
return false;
}
if (expectedVersion != -1 && stat.getVersion() != expectedVersion) {
LOG.warn(
"The node we are trying to delete is not the expected one. " + "Got a version mismatch");
return false;
}
synchronized (zkw.getNodes()) {
// TODO: Does this go here or only if we successfully delete node?
zkw.getNodes().remove(node);
if (!ZKUtil.deleteNode(zkw, node, stat.getVersion())) {
LOG.warn(
zkw.prefix(
"Attempting to delete "
+ "unassigned node in "
+ expectedState
+ " state but "
+ "after verifying it was in OPENED state, we got a version mismatch"));
return false;
}
LOG.debug(
zkw.prefix(
"Successfully deleted unassigned node for region "
+ regionName
+ " in expected state "
+ expectedState));
return true;
}
}
/**
* Creates an unassigned node in the OFFLINE state for the specified region.
*
* <p>Runs asynchronously. Depends on no pre-existing znode.
*
* <p>Sets a watcher on the unassigned region node.
*
* @param zkw zk reference
* @param region region to be created as offline
* @param serverName server event originates from
* @param cb
* @param ctx
* @throws KeeperException if unexpected zookeeper exception
* @throws KeeperException.NodeExistsException if node already exists
*/
public static void asyncCreateNodeOffline(
ZooKeeperWatcher zkw,
HRegionInfo region,
String serverName,
final AsyncCallback.StringCallback cb,
final Object ctx)
throws KeeperException {
LOG.debug(
zkw.prefix(
"Async create of unassigned node for "
+ region.getEncodedName()
+ " with OFFLINE state"));
RegionTransitionData data =
new RegionTransitionData(EventType.M_ZK_REGION_OFFLINE, region.getRegionName(), serverName);
synchronized (zkw.getNodes()) {
String node = getNodeName(zkw, region.getEncodedName());
zkw.getNodes().add(node);
ZKUtil.asyncCreate(zkw, node, data.getBytes(), cb, ctx);
}
}
public void deleteMetaLocation(ZooKeeperWatcher zookeeper, int replicaId) throws KeeperException {
if (replicaId == HRegionInfo.DEFAULT_REPLICA_ID) {
LOG.info("Deleting hbase:meta region location in ZooKeeper");
} else {
LOG.info("Deleting hbase:meta for " + replicaId + " region location in ZooKeeper");
}
try {
// Just delete the node. Don't need any watches.
ZKUtil.deleteNode(zookeeper, zookeeper.getZNodeForReplica(replicaId));
} catch (KeeperException.NoNodeException nne) {
// Has already been deleted
}
}
/**
* Sets the location of <code>hbase:meta</code> in ZooKeeper to the specified server address.
*
* @param zookeeper
* @param serverName
* @param replicaId
* @param state
* @throws KeeperException
*/
public static void setMetaLocation(
ZooKeeperWatcher zookeeper, ServerName serverName, int replicaId, RegionState.State state)
throws KeeperException {
LOG.info("Setting hbase:meta region location in ZooKeeper as " + serverName);
// Make the MetaRegionServer pb and then get its bytes and save this as
// the znode content.
MetaRegionServer pbrsr =
MetaRegionServer.newBuilder()
.setServer(ProtobufUtil.toServerName(serverName))
.setRpcVersion(HConstants.RPC_CURRENT_VERSION)
.setState(state.convert())
.build();
byte[] data = ProtobufUtil.prependPBMagic(pbrsr.toByteArray());
try {
ZKUtil.setData(zookeeper, zookeeper.getZNodeForReplica(replicaId), data);
} catch (KeeperException.NoNodeException nne) {
if (replicaId == HRegionInfo.DEFAULT_REPLICA_ID) {
LOG.debug("META region location doesn't exist, create it");
} else {
LOG.debug("META region location doesn't exist for replicaId " + replicaId + ", create it");
}
ZKUtil.createAndWatch(zookeeper, zookeeper.getZNodeForReplica(replicaId), data);
}
}
/**
* Wait until the primary meta region is available. Get the secondary locations as well but don't
* block for those.
*
* @param zkw
* @param timeout
* @param conf
* @return ServerName or null if we timed out.
* @throws InterruptedException
*/
public List<ServerName> blockUntilAvailable(
final ZooKeeperWatcher zkw, final long timeout, Configuration conf)
throws InterruptedException {
int numReplicasConfigured = 1;
try {
List<String> metaReplicaNodes = zkw.getMetaReplicaNodes();
numReplicasConfigured = metaReplicaNodes.size();
} catch (KeeperException e) {
LOG.warn("Got ZK exception " + e);
}
List<ServerName> servers = new ArrayList<ServerName>(numReplicasConfigured);
ServerName server = blockUntilAvailable(zkw, timeout);
if (server == null) return null;
servers.add(server);
for (int replicaId = 1; replicaId < numReplicasConfigured; replicaId++) {
// return all replica locations for the meta
servers.add(getMetaRegionLocation(zkw, replicaId));
}
return servers;
}
/**
* Verifies that the specified region is in the specified state in ZooKeeper.
*
* <p>Returns true if region is in transition and in the specified state in ZooKeeper. Returns
* false if the region does not exist in ZK or is in a different state.
*
* <p>Method synchronizes() with ZK so will yield an up-to-date result but is a slow read.
*
* @param zkw
* @param region
* @param expectedState
* @return true if region exists and is in expected state
*/
public static boolean verifyRegionState(
ZooKeeperWatcher zkw, HRegionInfo region, EventType expectedState) throws KeeperException {
String encoded = region.getEncodedName();
String node = getNodeName(zkw, encoded);
zkw.sync(node);
// Read existing data of the node
byte[] existingBytes = null;
try {
existingBytes = ZKUtil.getDataAndWatch(zkw, node);
} catch (KeeperException.NoNodeException nne) {
return false;
} catch (KeeperException e) {
throw e;
}
if (existingBytes == null) return false;
RegionTransitionData existingData = RegionTransitionData.fromBytes(existingBytes);
if (existingData.getEventType() == expectedState) {
return true;
}
return false;
}
/**
* Load the meta region state from the meta server ZNode.
*
* @param zkw
* @param replicaId
* @return regionstate
* @throws KeeperException
*/
public static RegionState getMetaRegionState(ZooKeeperWatcher zkw, int replicaId)
throws KeeperException {
RegionState.State state = RegionState.State.OPEN;
ServerName serverName = null;
try {
byte[] data = ZKUtil.getData(zkw, zkw.getZNodeForReplica(replicaId));
if (data != null && data.length > 0 && ProtobufUtil.isPBMagicPrefix(data)) {
try {
int prefixLen = ProtobufUtil.lengthOfPBMagic();
ZooKeeperProtos.MetaRegionServer rl =
ZooKeeperProtos.MetaRegionServer.PARSER.parseFrom(
data, prefixLen, data.length - prefixLen);
if (rl.hasState()) {
state = RegionState.State.convert(rl.getState());
}
HBaseProtos.ServerName sn = rl.getServer();
serverName = ServerName.valueOf(sn.getHostName(), sn.getPort(), sn.getStartCode());
} catch (InvalidProtocolBufferException e) {
throw new DeserializationException("Unable to parse meta region location");
}
} else {
// old style of meta region location?
serverName = ServerName.parseFrom(data);
}
} catch (DeserializationException e) {
throw ZKUtil.convert(e);
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
}
if (serverName == null) {
state = RegionState.State.OFFLINE;
}
return new RegionState(
RegionReplicaUtil.getRegionInfoForReplica(HRegionInfo.FIRST_META_REGIONINFO, replicaId),
state,
serverName);
}
/**
* Private method that actually performs unassigned node transitions.
*
* <p>Attempts to transition the unassigned node for the specified region from the expected state
* to the state in the specified transition data.
*
* <p>Method first reads existing data and verifies it is in the expected state. If the node does
* not exist or the node is not in the expected state, the method returns -1. If the transition is
* successful, the version number of the node following the transition is returned.
*
* <p>If the read state is what is expected, it attempts to write the new state and data into the
* node. When doing this, it includes the expected version (determined when the existing state was
* verified) to ensure that only one transition is successful. If there is a version mismatch, the
* method returns -1.
*
* <p>If the write is successful, no watch is set and the method returns true.
*
* @param zkw zk reference
* @param region region to be transitioned to opened
* @param serverName server event originates from
* @param endState state to transition node to if all checks pass
* @param beginState state the node must currently be in to do transition
* @param expectedVersion expected version of data before modification, or -1
* @return version of node after transition, -1 if unsuccessful transition
* @throws KeeperException if unexpected zookeeper exception
*/
public static int transitionNode(
ZooKeeperWatcher zkw,
HRegionInfo region,
String serverName,
EventType beginState,
EventType endState,
int expectedVersion)
throws KeeperException {
String encoded = region.getEncodedName();
if (LOG.isDebugEnabled()) {
LOG.debug(
zkw.prefix(
"Attempting to transition node "
+ HRegionInfo.prettyPrint(encoded)
+ " from "
+ beginState.toString()
+ " to "
+ endState.toString()));
}
String node = getNodeName(zkw, encoded);
zkw.sync(node);
// Read existing data of the node
Stat stat = new Stat();
byte[] existingBytes = ZKUtil.getDataNoWatch(zkw, node, stat);
if (existingBytes == null) {
// Node no longer exists. Return -1. It means unsuccessful transition.
return -1;
}
RegionTransitionData existingData = RegionTransitionData.fromBytes(existingBytes);
// Verify it is the expected version
if (expectedVersion != -1 && stat.getVersion() != expectedVersion) {
LOG.warn(
zkw.prefix(
"Attempt to transition the "
+ "unassigned node for "
+ encoded
+ " from "
+ beginState
+ " to "
+ endState
+ " failed, "
+ "the node existed but was version "
+ stat.getVersion()
+ " not the expected version "
+ expectedVersion));
return -1;
}
// Verify it is in expected state
if (!existingData.getEventType().equals(beginState)) {
LOG.warn(
zkw.prefix(
"Attempt to transition the "
+ "unassigned node for "
+ encoded
+ " from "
+ beginState
+ " to "
+ endState
+ " failed, "
+ "the node existed but was in the state "
+ existingData.getEventType()
+ " set by the server "
+ existingData.getServerName()));
return -1;
}
// Write new data, ensuring data has not changed since we last read it
try {
RegionTransitionData data =
new RegionTransitionData(endState, region.getRegionName(), serverName);
if (!ZKUtil.setData(zkw, node, data.getBytes(), stat.getVersion())) {
LOG.warn(
zkw.prefix(
"Attempt to transition the "
+ "unassigned node for "
+ encoded
+ " from "
+ beginState
+ " to "
+ endState
+ " failed, "
+ "the node existed and was in the expected state but then when "
+ "setting data we got a version mismatch"));
return -1;
}
if (LOG.isDebugEnabled()) {
LOG.debug(
zkw.prefix(
"Successfully transitioned node "
+ encoded
+ " from "
+ beginState
+ " to "
+ endState));
}
return stat.getVersion() + 1;
} catch (KeeperException.NoNodeException nne) {
LOG.warn(
zkw.prefix(
"Attempt to transition the "
+ "unassigned node for "
+ encoded
+ " from "
+ beginState
+ " to "
+ endState
+ " failed, "
+ "the node existed and was in the expected state but then when "
+ "setting data it no longer existed"));
return -1;
}
}
/**
* Deletes all unassigned nodes regardless of their state.
*
* <p>No watchers are set.
*
* <p>This method is used by the Master during cluster startup to clear out any existing state
* from other cluster runs.
*
* @param zkw zk reference
* @throws KeeperException if unexpected zookeeper exception
*/
public static void deleteAllNodes(ZooKeeperWatcher zkw) throws KeeperException {
LOG.debug(zkw.prefix("Deleting any existing unassigned nodes"));
ZKUtil.deleteChildrenRecursively(zkw, zkw.assignmentZNode);
}
