Some of our integration tests are a little flaky and have been causing CI builds to fail. See builds on #338 for some examples.

It's possible that some of this flakiness is caused by creating connectors and then immediately hitting the Connect cluster again with a request that fails because a rebalance is pending (rebalances are usually triggered twice when connectors are created: once to assign the Connector instance to a worker, and a second time to assign its Task instances to workers across the cluster).