I have been using BigQuery in the on-demand pricing model. For some complex queries I have seen from the Stackdriver report that BigQuery over allocates slots beyond the 2,000 mark, e.g. I have seen up to 7,000 slots allocated in the same project over several minutes. It did not go beyond this number. I am wondering how to interpret these figures.
Does BigQuery allocate additional (otherwise) idle slots if there is spare capacity in Google data centres?
The 2,000 slot limit does not seem to be a hard limit, correct?
When I hit 7,000 slots I saw a slowdown of queries. Is this when some sort of resource management kicks in that serialises the queries (FIFO) or FAIRly allocates resources.
I am curious if this over allocation also happen for flat rate pricing?