cinder-manage db purge can fail on large volumes tables (~500000 rows or more) because the time to execute the transaction to purge volumes exceeds the default 120 second innodb timeout at the database layer.

For deployments that have gotten to a huge table like this (perhaps because purging wasn't completing successfully due to other bugs) it would be helpful to either:

a) Add a parameter to the command line that limits the operation to X rows

b) Loop inside of the purge operation with size limits that won't exceed the db timeout