Folks,

I have some jobs running 4 wide that open a cursor and then fetch rows from a single DB2 table. This table is partitioned by account number. Somehow, every so often, two of the jobs will collide with -911 deadlock errors, killing one of the jobs, even though they are accessing accounts in distinct partitions.

Has anyone heard of -911 deadlock errors when the data being accessed is in separate partitions?

The cursor does use account number in the WHERE clause.

Please excuse the general nature of this note; any insight is appreciated.

Hello! I am sorry for my poor English!

Details are necessary (OS, version of DB2, isolation level, sample code and etc.)!

Make sure you check the reason code. RC 2 is deadlock, and RC 68 is lock timeout. They are not the same thing.

If you are only doing selects, you cannot get lock contention since multiple share locks can coexist with eachother without any interference.

Check to see if you are getting lock escalation (db2diag.log) and it might be a good idea to run a snapshot for locks.

If those jobs are just reading and not updating the table, make sure that they access the table with the correct isolation level; if necessary, set the isolation level to CS (cursor stability) or even to UR (uncommitted read).

-- Peter.