INSERT INTO ndbtable SELECT FROM table uses ridiculous amounts of memory in MySQL Cluster

Question

If I move 300,000 rows from a table (either innodb or ndb) into a ndb table like so:

INSERT INTO ndbtable2
SELECT a, b,IFNULL(c,UTC_TIMESTAMP()),CASE WHEN b = 'X' THEN e ELSE d END,f
FROM ndbtable1;

Query OK, 308372 rows affected (5 min 12.59 sec)
Records: 308372  Duplicates: 0  Warnings: 0

ndb uses an increasingly huge about of data memory until it completes. Before / peak / after as follows

ndb_mgm -e "all report memoryusage"
Connected to Management Server at: fl-prd-mysql1:1186
Node 1: Data usage is 2%(5752 32K pages of total 262144)
Node 1: Index usage is 0%(2428 8K pages of total 262176)
Node 2: Data usage is 2%(5752 32K pages of total 262144)
Node 2: Index usage is 0%(2428 8K pages of total 262176)

Connected to Management Server at: fl-prd-mysql1:1186
Node 1: Data usage is 62%(164013 32K pages of total 262144)
Node 1: Index usage is 1%(3136 8K pages of total 262176)
Node 2: Data usage is 62%(164013 32K pages of total 262144)
Node 2: Index usage is 1%(3136 8K pages of total 262176)

Connected to Management Server at: fl-prd-mysql1:1186
Node 1: Data usage is 3%(10293 32K pages of total 262144)
Node 1: Index usage is 1%(4590 8K pages of total 262176)
Node 2: Data usage is 3%(10293 32K pages of total 262144)
Node 2: Index usage is 1%(4590 8K pages of total 262176)

If my math is right I insert 10293 - 5752 = 4541 = 142 MB but memory spikes by 164013 - 5752 = 158261 = 4945 MB

Also if I limit the insert to 50,000 rows, with only a 3MB before/after difference, the spike was 780MB.

Obviously this will be problem when ndb isn't empty... what is happening here?!

Answer 1

In absence of an explanation the manual does acknowledge the limitation as follows

As noted elsewhere in this chapter, MySQL Cluster does not handle large transactions well; it is better to perform a number of small transactions with a few operations each than to attempt a single large transaction containing a great many operations. Among other considerations, large transactions require very large amounts of memory.

http://dev.mysql.com/doc/refman/5.5/en/mysql-cluster-limitations-transactions.html

"As noted elsewhere in this chapter" ... I have looked but yet to find anything specific.

FYI the alternatives are:

1. Loop through iterations with a row LIMIT (say 10,000)
2. If inserting into an empty table consider LOAD DATA or even using innodb and finally ALTER TABLE ... ENGINE = ndbcluster (neither are transactional)