Error 73 when attempting a CREATE TABLE

[Atanu Mishra]

Encountered a TMF error 73 with hbase .98 build 20141004:

SQL>CREATE TABLE like_nopart LIKE mysalt

>>> SQL statement failed, [25000] [HP][HP ODBC Driver][HP Neoview Database] SQL ERROR:*** ERROR[8606] Transaction subsystem TMF returned error 73 on a commit transaction. [2014-10-06 13:31:46] (-8606) (SQLExecDirectW)

trafodion.dtm.log contained:

centos-mapr6: 2014-10-06 20:31:45,997 ERROR transactional.TransactionManager: Abort HasException true: true
centos-mapr6: 2014-10-06 20:31:45,997 ERROR transactional.TransactionManager: Abort HasException true: java.io.IOException: UnknownTransactionException
centos-mapr6: 2014-10-06 20:31:45,998 ERROR transactional.TransactionManager: exception in doAbortX (ignoring): java.lang.Exception: java.io.IOException: UnknownTransactionException

-----
TM will need to add a balancer workaround into the build until we can implement the code to handle a balance or a split.

Before your testing, can you please try disabling the load balancer through the hbase shell as shown below. Setting the balance_switch to false. The problem is the region is being moved while the old location is still in the client-side Transaction State.

Example:

>hbase shell

                        hbase(main):002:0> balance_switch false

true <-- Output will be the last setting of the balance_switch value

0 row(s) in 0.0080 seconds

A simple solution is not available yet since the Master process does the balancing, and TM code is mostly in the Region. In the 0.94-based version, the TM subclassed the RegionServer, and we had RS calls that were used to communicate whether a split or balance was allowed.

For 0.98 the call needs to be made through a coprocessor

[Atanu Mishra (atanu-mishra)]

We have resolved many of the UnknownTransactionException errors with fixes for other issues so the original problem may already be resolved, but I think this is a good placeholder bug for the split/balance work that needs to be done.

[Oliver Bucaojit (oliver-bucaojit)]

Checked-in split delay portion of the fix on Jan 8.

Commit Message:
Resubmitting split delay work, additional checks

Split delay portion of LP bug 1378544.
Added a check/put mechanism for putting objects into
shared map. Depends on which coprocessor is started
first according to what was entered in the HBase
configuration under 'hbase.coprocessor.region.classes'

[Oliver Bucaojit (oliver-bucaojit)]

We haven’t seen any more reports of error 73’s such as in the bug scenario.

This bug was mainly left as a placeholder for the balance/split work.

We are making progress on online-handling of region balance and spitting, it is planned to be enabled in release 1.3+.

Blueprint for the design can be found here:
https://blueprints.launchpad.net/trafodion/+spec/dtm-handling-rebalance-split