Submitting large numbers of jobs can cause the negotiator to crash when it runs out of file descriptors. For 2.0 our solution is to update scale documentation with instructions to set MAX_FILE_DESCRIPTORS higher for large scale installations. See also: bug 603663 bug 691440
Content for "Large" subsection of the "MRG Grid Scale Requirements" section (chapter 3) (beginning with existing paragraph): ====================================== Large A large-scale grid is defined as a grid supporting 5000 Execute Nodes, or a similar magnitude. There are several considerations when implementing a large scale grid. Red Hat, Inc currently recommends that customers configure large-scale MRG Grid installations in cooperation with a Solutions Architect through Red Hat, Inc Consulting. When the job volume is expected to be high, the administrator should increase the number of file descriptors allowed for the Negotiator. A safe practice is to set this limit to twice the number of jobs expected during a negotiation cycle. For example, if the anticipated job volume is 1000 jobs per negotiation cycle, then increase the file descriptor limit to 2000: NEGOTIATOR.MAX_FILE_DESCRIPTORS = 2000 ======================================
This doc update may become moot if repro fails for bz 603663
(In reply to comment #2) > This doc update may become moot if repro fails for bz 603663 Can you advise whether this is still required as bug603663 has now been put into CLOSED-wontfix status? Thanks