Process indicators are collected over a period and written in batches.  This often results in it going down the path of copyFrom in the store.  CopyFrom batches them in batches of 10000.  However, copyfrom requires a lock because it does a delete and we want to avoid deadlocks and odd states of rows getting written  So we have a lock. This is also done under a single transaction. The problem is that under large batches of indicators we wind up holding a lock and a transaction for a long time.  This puts strain on central and central-db.  In benchmark testing I could not even successfully add groups of 500K indicators because the mutex would timeout.