When a competing lock request timed out, the attempted lock was sometimes unable to be fully released, which could eventually cause the node to fail.
Each DDL transaction acquires global catalog X lock to commit the transaction for few secs/milisec range and if the transaction fails to acquire that because of other transaction holding that lock for longer time then the database/nodes get down to avoid further corruption on the catalog. The message below from the vertica.log is a sample indication of this problem:
2014-11-24 17:02:17.157 DistCall Dispatch:0x2ab265884a40-a0000001977085
<PANIC> @v_data_node0003: VX001/5706: Failure to acquire global catalog X lock for commit of txn a0000001977085
LOCATION: doIt, /scratch_a/release/vbu
This issue has been resolved in Vertica 7.0.2-5 that accompanies the Infrastructure Management Release 2.4.1.
Upgrade to Infrastructure Management 2.4.1.