DM application failure

If only the DM application fails, the session itself remains active, events remain pending, and client threads remain blocked waiting for a response. New events will continue to arrive at the session queue.

Note: GPFS is unable to detect that the DM application has failed.

The failed DM application must be recovered on the same node, and continue handling the events. Since no DMAPI resources are lost in this case, there is little purpose in moving the DM application to another node. Assuming an existing session on another node is not permitted in GPFS, except after session node failure.

If the DM application fails simultaneously with the session node, the gpfsready shell script can be used to restart the DM application on the failed node. See Initializing the Data Management application. In the case of simultaneous failures, the DM application can also be moved to another node and assume the failed session there. See Single-node failure.