When a query is temporarily prevented from starting new work (for example, because it's waiting for downloads from bucket storage), there is a limit to how much penalty credit it is allowed to "save up". This is to avoid LogScale overprioritizing such a query when it becomes ready to run.

As an example of why this matters is, if a query is waiting for bucket downloads for 20 seconds because bucket storage happens to be slow, then that query should not get exclusive use of the worker for 20 seconds once it becomes able to run again.

The default value is 1. This allows a query to "save up" up to one second of credit per mapper pool thread. For example, on a host with 20 mapper threads (with default settings, this might be a 40-core machine), this default setting allows a query to gain up to 20 seconds of execution time while it is paused. As the execution time is divided onto the mapper threads, this would likely result in the query getting to run uncontested for 1 second after unpausing.

The range of permitted settings for this value is 0 to 10, allowing queries to "save up" that many seconds per mapper pool thread.