hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pankaj Misra <pankaj.mi...@impetus.co.in>
Subject RE: HBase NonBlocking and Async Thrift
Date Tue, 20 Nov 2012 07:06:09 GMT
Dear All,

Requesting for help on the async thrift protocol for non-blocking streaming mode, would greatly
appreciate any input on the issue outlined in the thread below. Putting a wait/delay defeats
the purpose of using the async capability. Please help.

Thanks & Regards
Pankaj Misra

-----Original Message-----
From: Pankaj Misra
Sent: Monday, November 19, 2012 5:47 PM
To: user@hbase.apache.org
Subject: HBase NonBlocking and Async Thrift

Dear All,

I am currently using Hadoop 0.23.1 with HBase 0.94.1 in a pseudo-distributed mode. I am trying
to use HBase Thrift API (not using Thrift2 yet) in a nonblocking and async mode to insert
a bulk of records. I am sharing the set of steps for everyone's information and setting the
context to my problem

Please find below the code that I am using for initializing the async client

TBinaryProtocol.Factory binProtoFactory=new TBinaryProtocol.Factory();

TAsyncClientManager clientManager=null;
TNonblockingSocket nonBlockingSocket=null;

try {
   clientManager=new TAsyncClientManager(); } catch (IOException e) {
  throw new RuntimeException(e);
}
try {
  nonBlockingSocket=new TNonblockingSocket(HOST_NAME,PORT_NUMBER);
} catch (IOException e) {
throw new RuntimeException(e);
}


And, I am initializing the client as shown below

Hbase.AsyncClient client=new Hbase.AsyncClient(binProtoFactory, clientManager, nonBlockingSocket);



I could see two ways of using the client, i.e. one client for all the records to be inserted
or separate instance of client for every record. I thought since this is a non-blocking channel,
it would make sense to initialize 1 client for all the requests, since all the requests would
be streamed using a framed transport.


// 1 async client for all the requests
Hbase.AsyncClient client=new Hbase.AsyncClient(binProtoFactory, clientManager, nonBlockingSocket);

// mutate rows called in a loop to insert multiple records, using the same client client.mutateRow(table,
ByteBuffer.wrap(key), mutations,mutationAttributes,new HBaseInsertAsyncHandler());


But soon I found that I was wrong as I got back the following error.

java.lang.IllegalStateException: Client is currently executing another method: org.apache.hadoop.hbase.thrift.generated.Hbase$AsyncClient$mutateRow_call
    at org.apache.thrift.async.TAsyncClient.checkReady(TAsyncClient.java:78)
    at org.apache.hadoop.hbase.thrift.generated.Hbase$AsyncClient.mutateRow(Hbase.java:2714)
Reading through the following JIRA educated me a bit more on this

https://issues.apache.org/jira/browse/THRIFT-945

So, I changed my code to initialize the client per record to be inserted.


//called both these statements for every record to be inserted in a loop Hbase.AsyncClient
client=new Hbase.AsyncClient(binProtoFactory, clientManager, nonBlockingSocket); client.mutateRow(table,
ByteBuffer.wrap(key), mutations,mutationAttributes,new HBaseInsertAsyncHandler());

Even this failed with the following error
2012-11-19 17:25:15,275 WARN  [TAsyncClientManager#SelectorThread 9] async.TAsyncClientManager
(TAsyncClientManager.java:startPendingMethods(177)) - Caught exception in TAsyncClientManager!
java.nio.channels.ClosedChannelException
    at java.nio.channels.spi.AbstractSelectableChannel.register(AbstractSelectableChannel.java:167)
    at java.nio.channels.SelectableChannel.register(SelectableChannel.java:254)
    at org.apache.thrift.transport.TNonblockingSocket.registerSelector(TNonblockingSocket.java:99)
    at org.apache.thrift.async.TAsyncMethodCall.start(TAsyncMethodCall.java:141)
    at org.apache.thrift.async.TAsyncClientManager$SelectThread.startPendingMethods(TAsyncClientManager.java:169)
    at org.apache.thrift.async.TAsyncClientManager$SelectThread.run(TAsyncClientManager.java:114)

So, it looked to me as if the channel registration could not happen in time for it to get
initialized and since the records are getting inserted in a loop, it possibly needs a time
window for initialization to get complete and insert the record. So I had to introduce a delay
for every such insertion, which I do not prefer to do. The code changes for that are as shown
below

Hbase.AsyncClient client=new Hbase.AsyncClient(binProtoFactory, clientManager, nonBlockingSocket);
client.mutateRow(table, ByteBuffer.wrap(key), mutations,mutationAttributes,new HBaseInsertAsyncHandler());
synchronized (client.getProtocolFactory()) {
    client.getProtocolFactory().wait(20);
}
With the above change, I could see the records getting inserted into HBase using async thrift
client, but I think this is not the right solution and will look for some guidance from the
community to have a more consistent way to utilize the the async thrift capability without
any specific wait or sleep times, as putting a wait call, kills the async advantage and introduces
delays in the overall throughput. Looking forward for your help.

Thanks and Regards
Pankaj Misra

________________________________

Neustar VP and Impetus CEO to present on 'Innovative information services powered by Cloud
and Big Data technologies'at Cloud Expo - Santa Clara, Nov 6th. http://www.impetus.com/events#2.

Check out Impetus contribution to build Luminar - a new business unit at Entravision. http://lf1.me/MS/


NOTE: This message may contain information that is confidential, proprietary, privileged or
otherwise protected by law. The message is intended solely for the named addressee. If received
in error, please destroy and notify the sender. Any use of this email is prohibited when received
in error. Impetus does not represent, warrant and/or guarantee, that the integrity of this
communication has been maintained nor that the communication is free of errors, virus, interception
or interference.

________________________________

Neustar VP and Impetus CEO to present on ‘Innovative information services powered by Cloud
and Big Data technologies’at Cloud Expo - Santa Clara, Nov 6th. http://www.impetus.com/events#2.

Check out Impetus contribution to build Luminar - a new business unit at Entravision. http://lf1.me/MS/


NOTE: This message may contain information that is confidential, proprietary, privileged or
otherwise protected by law. The message is intended solely for the named addressee. If received
in error, please destroy and notify the sender. Any use of this email is prohibited when received
in error. Impetus does not represent, warrant and/or guarantee, that the integrity of this
communication has been maintained nor that the communication is free of errors, virus, interception
or interference.
Mime
View raw message