Class RecoverableZooKeeper
To handle recoverable errors, developers need to realize that there are two classes of requests: idempotent and non-idempotent requests. Read requests and unconditional sets and deletes are examples of idempotent requests, they can be reissued with the same results.
(Although, the delete may throw a NoNodeException on reissue its effect on the ZooKeeper state is the same.) Non-idempotent requests need special handling, application and library writers need to keep in mind that they may need to encode information in the data or name of znodes to detect retries. A simple example is a create that uses a sequence flag. If a process issues a create("/x-", ..., SEQUENCE) and gets a connection loss exception, that process will reissue another create("/x-", ..., SEQUENCE) and get back x-111. When the process does a getChildren("/"), it sees x-1,x-30,x-109,x-110,x-111, now it could be that x-109 was the result of the previous create, so the process actually owns both x-109 and x-111. An easy way around this is to use "x-process id-" when doing the create. If the process is using an id of 352, before reissuing the create it will do a getChildren("/") and see "x-222-1", "x-542-30", "x-352-109", x-333-110". The process will know that the original create succeeded an the znode it created is "x-352-109".
- See Also:
-
- "http://wiki.apache.org/hadoop/ZooKeeper/ErrorHandling"
-
Field Summary
Modifier and TypeFieldDescriptionprivate final byte[]
private final String
private static final org.slf4j.Logger
private final int
private final String
private final RetryCounterFactory
private final int
private final org.apache.zookeeper.Watcher
private org.apache.zookeeper.ZooKeeper
private final org.apache.zookeeper.client.ZKClientConfig
-
Constructor Summary
ConstructorDescriptionRecoverableZooKeeper
(String quorumServers, int sessionTimeout, org.apache.zookeeper.Watcher watcher, int maxRetries, int retryIntervalMillis, int maxSleepTime, String identifier, int maxMultiSize, org.apache.zookeeper.client.ZKClientConfig zkClientConfig) -
Method Summary
Modifier and TypeMethodDescriptionprivate org.apache.zookeeper.ZooKeeper
checkZk()
Try to create a ZooKeeper connection.void
close()
static RecoverableZooKeeper
connect
(org.apache.hadoop.conf.Configuration conf, String ensemble, org.apache.zookeeper.Watcher watcher, String identifier, org.apache.zookeeper.client.ZKClientConfig zkClientConfig) Creates a new connection to ZooKeeper, pulling settings and ensemble config from the specified configuration object using methods fromZKConfig
.static RecoverableZooKeeper
connect
(org.apache.hadoop.conf.Configuration conf, org.apache.zookeeper.Watcher watcher) create
(String path, byte[] data, List<org.apache.zookeeper.data.ACL> acl, org.apache.zookeeper.CreateMode createMode) NONSEQUENTIAL create is idempotent operation.private String
createNonSequential
(String path, byte[] data, List<org.apache.zookeeper.data.ACL> acl, org.apache.zookeeper.CreateMode createMode) private String
createSequential
(String path, byte[] data, List<org.apache.zookeeper.data.ACL> acl, org.apache.zookeeper.CreateMode createMode) void
delete is an idempotent operation.org.apache.zookeeper.data.Stat
exists is an idempotent operation.org.apache.zookeeper.data.Stat
exists is an idempotent operation.filterByPrefix
(List<String> nodes, String... prefixes) Filters the given node list by the given prefixes.private String
List<org.apache.zookeeper.data.ACL>
getAcl is an idempotent operation.getChildren
(String path, boolean watch) getChildren is an idempotent operation.getChildren
(String path, org.apache.zookeeper.Watcher watcher) getChildren is an idempotent operation.byte[]
getData is an idempotent operation.byte[]
getData is an idempotent operation.int
Returns the maximum size (in bytes) that should be included in any single multi() call.long
byte[]
org.apache.zookeeper.ZooKeeper.States
getState()
org.apache.zookeeper.ZooKeeper
List<org.apache.zookeeper.OpResult>
Run multiple operations in a transactional manner.private Iterable<org.apache.zookeeper.Op>
prepareZKMulti
(Iterable<org.apache.zookeeper.Op> ops) Convert Iterable ofOp
we got into the ZooKeeper.Op instances to actually pass to multi (need to do this in order to appendMetaData).void
private void
retryOrThrow
(RetryCounter retryCounter, org.apache.zookeeper.KeeperException e, String opName) org.apache.zookeeper.data.Stat
setAcl is an idempotent operation.org.apache.zookeeper.data.Stat
setData is NOT an idempotent operation.void
-
Field Details
-
LOG
-
zk
-
retryCounterFactory
-
identifier
-
id
-
watcher
-
sessionTimeout
-
quorumServers
-
maxMultiSize
-
zkClientConfig
-
-
Constructor Details
-
RecoverableZooKeeper
RecoverableZooKeeper(String quorumServers, int sessionTimeout, org.apache.zookeeper.Watcher watcher, int maxRetries, int retryIntervalMillis, int maxSleepTime, String identifier, int maxMultiSize, org.apache.zookeeper.client.ZKClientConfig zkClientConfig) throws IOException - Throws:
IOException
-
-
Method Details
-
connect
public static RecoverableZooKeeper connect(org.apache.hadoop.conf.Configuration conf, org.apache.zookeeper.Watcher watcher) throws IOException - Throws:
IOException
-
connect
public static RecoverableZooKeeper connect(org.apache.hadoop.conf.Configuration conf, String ensemble, org.apache.zookeeper.Watcher watcher, String identifier, org.apache.zookeeper.client.ZKClientConfig zkClientConfig) throws IOException Creates a new connection to ZooKeeper, pulling settings and ensemble config from the specified configuration object using methods fromZKConfig
. Sets the connection status monitoring watcher to the specified watcher.- Parameters:
conf
- configuration to pull ensemble and other settings fromwatcher
- watcher to monitor connection changesensemble
- ZooKeeper servers quorum stringidentifier
- value used to identify this client instance.zkClientConfig
- client specific configurations for this instance- Returns:
- connection to zookeeper
- Throws:
IOException
- if unable to connect to zk or config problem
-
getMaxMultiSizeLimit
Returns the maximum size (in bytes) that should be included in any single multi() call. NB: This is an approximation, so there may be variance in the msg actually sent over the wire. Please be sure to set this approximately, with respect to your ZK server configuration for jute.maxbuffer. -
checkZk
Try to create a ZooKeeper connection. Turns any exception encountered into a KeeperException.OperationTimeoutException so it can retried.- Returns:
- The created ZooKeeper connection object
- Throws:
org.apache.zookeeper.KeeperException
- if a ZooKeeper operation fails
-
reconnectAfterExpiration
public void reconnectAfterExpiration() throws IOException, org.apache.zookeeper.KeeperException, InterruptedException- Throws:
IOException
org.apache.zookeeper.KeeperException
InterruptedException
-
delete
public void delete(String path, int version) throws InterruptedException, org.apache.zookeeper.KeeperException delete is an idempotent operation. Retry before throwing exception. This function will not throw NoNodeException if the path does not exist.- Throws:
InterruptedException
org.apache.zookeeper.KeeperException
-
exists
public org.apache.zookeeper.data.Stat exists(String path, org.apache.zookeeper.Watcher watcher) throws org.apache.zookeeper.KeeperException, InterruptedException exists is an idempotent operation. Retry before throwing exception- Returns:
- A Stat instance
- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
exists
public org.apache.zookeeper.data.Stat exists(String path, boolean watch) throws org.apache.zookeeper.KeeperException, InterruptedException exists is an idempotent operation. Retry before throwing exception- Returns:
- A Stat instance
- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
retryOrThrow
private void retryOrThrow(RetryCounter retryCounter, org.apache.zookeeper.KeeperException e, String opName) throws org.apache.zookeeper.KeeperException - Throws:
org.apache.zookeeper.KeeperException
-
getChildren
public List<String> getChildren(String path, org.apache.zookeeper.Watcher watcher) throws org.apache.zookeeper.KeeperException, InterruptedException getChildren is an idempotent operation. Retry before throwing exception- Returns:
- List of children znodes
- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
getChildren
public List<String> getChildren(String path, boolean watch) throws org.apache.zookeeper.KeeperException, InterruptedException getChildren is an idempotent operation. Retry before throwing exception- Returns:
- List of children znodes
- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
getData
public byte[] getData(String path, org.apache.zookeeper.Watcher watcher, org.apache.zookeeper.data.Stat stat) throws org.apache.zookeeper.KeeperException, InterruptedException getData is an idempotent operation. Retry before throwing exception- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
getData
public byte[] getData(String path, boolean watch, org.apache.zookeeper.data.Stat stat) throws org.apache.zookeeper.KeeperException, InterruptedException getData is an idempotent operation. Retry before throwing exception- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
setData
public org.apache.zookeeper.data.Stat setData(String path, byte[] data, int version) throws org.apache.zookeeper.KeeperException, InterruptedException setData is NOT an idempotent operation. Retry may cause BadVersion Exception Adding an identifier field into the data to check whether badversion is caused by the result of previous correctly setData- Returns:
- Stat instance
- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
getAcl
public List<org.apache.zookeeper.data.ACL> getAcl(String path, org.apache.zookeeper.data.Stat stat) throws org.apache.zookeeper.KeeperException, InterruptedException getAcl is an idempotent operation. Retry before throwing exception- Returns:
- list of ACLs
- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
setAcl
public org.apache.zookeeper.data.Stat setAcl(String path, List<org.apache.zookeeper.data.ACL> acls, int version) throws org.apache.zookeeper.KeeperException, InterruptedException setAcl is an idempotent operation. Retry before throwing exception- Returns:
- list of ACLs
- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
create
public String create(String path, byte[] data, List<org.apache.zookeeper.data.ACL> acl, org.apache.zookeeper.CreateMode createMode) throws org.apache.zookeeper.KeeperException, InterruptedException NONSEQUENTIAL create is idempotent operation. Retry before throwing exceptions. But this function will not throw the NodeExist exception back to the application.
But SEQUENTIAL is NOT idempotent operation. It is necessary to add identifier to the path to verify, whether the previous one is successful or not.
- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
createNonSequential
private String createNonSequential(String path, byte[] data, List<org.apache.zookeeper.data.ACL> acl, org.apache.zookeeper.CreateMode createMode) throws org.apache.zookeeper.KeeperException, InterruptedException - Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
createSequential
private String createSequential(String path, byte[] data, List<org.apache.zookeeper.data.ACL> acl, org.apache.zookeeper.CreateMode createMode) throws org.apache.zookeeper.KeeperException, InterruptedException - Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
prepareZKMulti
private Iterable<org.apache.zookeeper.Op> prepareZKMulti(Iterable<org.apache.zookeeper.Op> ops) throws UnsupportedOperationException Convert Iterable ofOp
we got into the ZooKeeper.Op instances to actually pass to multi (need to do this in order to appendMetaData).- Throws:
UnsupportedOperationException
-
multi
public List<org.apache.zookeeper.OpResult> multi(Iterable<org.apache.zookeeper.Op> ops) throws org.apache.zookeeper.KeeperException, InterruptedException Run multiple operations in a transactional manner. Retry before throwing exception- Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
findPreviousSequentialNode
private String findPreviousSequentialNode(String path) throws org.apache.zookeeper.KeeperException, InterruptedException - Throws:
org.apache.zookeeper.KeeperException
InterruptedException
-
getSessionId
-
close
- Throws:
InterruptedException
-
getState
-
getZooKeeper
-
getSessionPasswd
-
sync
public void sync(String path, org.apache.zookeeper.AsyncCallback.VoidCallback cb, Object ctx) throws org.apache.zookeeper.KeeperException - Throws:
org.apache.zookeeper.KeeperException
-
filterByPrefix
Filters the given node list by the given prefixes. This method is all-inclusive--if any element in the node list starts with any of the given prefixes, then it is included in the result.- Parameters:
nodes
- the nodes to filterprefixes
- the prefixes to include in the result- Returns:
- list of every element that starts with one of the prefixes
-
getIdentifier
-