Comment 3 for bug 1100141

I still think it's important the FTWRL not start if we know a long running query is currently running. Many environments have hundreds of new connections per second coming in so we cannot afford for FTWRL to take longer than a second or so. For example if we know a query has been running for 3 hours we should not kick off FTWRL as it will be unclear when that query will finish and the databases will be completely locked during that time. So we either have two choices in this case:

1) wait for the query to finish for a period of time
2) immediately kill off the query that is preventing us from starting FTWRL

I suggest making it configurable.

It's not important that we avoid every race condition as we aren't trying to prevent every short running query. We are trying to prevent already running long-running queries from blocking FTWRL.

In addition once FTWRL has started the behavior I would like is to kill anything that is preventing that from going through quickly. This can be configurable but for customers with hundreds of new connections per second we want to ensure it's done in a timely manner and should kill anything that is causing a delay.