Issue with inserting data

[Jonathon Coombes]

I am trying to insert data into an 8-node (2 physical machines) cluster and seem to have trouble no matter how it is done. It is one simple table defined as being one field called `number` which is of type bigint. The number of records in the table is 127M.

Here are some of the methods attempted:

1. Via SQL file containing insert statements inserting into NDB table.
2. Via LOAD DATA INFILE into NDB table.
3. Via LOAD DATA INFILE into MYISAM table and then ALTER TABLE tablename ENGINE=NDB
4. via LOAD DATA INFILE into NDB table with ndb_use_transactions=OFF set for that session.

Results I have seen so far:

1. Nodes crash during the loading stage. Can be anywhere from 2M records up to around 30M as the best to date.
2. Error shows QMGR error most of the time when using the simple methods (1,2 above)
3. With more recent tests, the number 3 nodes seems to be the one failing. It can be started again by simply rerunning the ndbd process.
4. Doing the load with transactions turned off seem to be slower than with them on? Seems contradictory to what is expected?
5. On doing the ALTER TABLE step, the load into MYISAM was very quick (less than a few minutes). However, when I did the ALTER TABLE, rather than taking at most 30 minutes (expected), it ran for a few hours then gave a BAD error. It showed a filesystem error that the journal I/O had failed, and turned the partition (which was also root) to a read-only partition which required repairing to fix.

Questions:

1. How are people loading large amounts of data into their clusters?
2. Any ideas of where the problem may exist for the issue mentioned above?
3. What is the largest table loaded using the ALTER TABLE method?