It’s funny, I literally just finished writing a blogpost about SLOG stuff now. 
A maximum of three transaction groups can exist (one each in a state of OPEN, QUIESCING and SYNCING) - and yes, out of the box you’ll have the 4GB upper bound on the dirty_data_max value discussed lower down in the Resource thread Some Insights (and generalizations) on the OpenZFS Write Throttle
Obviously cranking the dirty_data_max value family up to match the whole drive size of 960GB would let you potentially fill up all of your RAM there if you had a write stall from your vdevs (e.g.: a failed/failing disk that becomes painfully slow)