Currently when wait_for_compactions is true, we block until pending compactions hit zero. However, very frequently all compactions will finish, but C* still calculated >0 pending compactions. What this means is that the cstar perf job will then hang forever. This means the entire job needs restarted.
If pending compactions >0, but no compactions are in progress, cstar should wait some time (5-30) minutes before accepting that compactions are finished and moving on with the job to the next task, and potentially logging a WARN.
Currently when wait_for_compactions is true, we block until pending compactions hit zero. However, very frequently all compactions will finish, but C* still calculated >0 pending compactions. What this means is that the cstar perf job will then hang forever. This means the entire job needs restarted.
If pending compactions >0, but no compactions are in progress, cstar should wait some time (5-30) minutes before accepting that compactions are finished and moving on with the job to the next task, and potentially logging a WARN.