v1.7.2

Summary Concurrency-limited jobs now give their slot back only to the semaphore they got it from. That fixes a job that ran past its duration letting an extra job run alongside the next holder ( #827 ). It also fixes jobs whose class goes missing during a deploy giving back a slot they never took, or keeping one they did take. Nothing needs migrating; see "Upgrading to version 1.7.2" for what to do before 2.0. What's Changed Give concurrency slots back only to the semaphore they came from by ...