NdbTransaction.hpp has two settings added :

  MaxPendingBlobReadBytes
  MaxPendingBlobWriteBytes

When the volume of data being read/written to Blob column part tables in 
the transaction exceeds these values, the transaction will implicitly execute
the accumulated operations.

This avoids an excessive build up of pending data which can result in resource
exhaustion in the kernel.

By default these parameters are set to 0, which means no limit.

At the MySQLD level, two new session variables are added :

  ndb_blob_read_batch_bytes
  ndb_blob_write_batch_bytes

These control the setting of the per-transaction values above.  
They can be defaulted for all transactions on the MySQLD command line.

When no default is given, they both default to 64kB.

This patch should improve the stability of MySQL Cluster when large Blob 
values are being read and written.