Change cluster client aggregation to test minimum clients

[jeremyprime]

Summary

This test frequently fails due to a race condition between measuring the cluster-wide client count and summing individual node counts as connections can be created/destroyed between the two measurements. Instead of verifying an exact count, verify the cluster-wide count is at least the minimum expected from individual nodes.

Testing

Ran CI tests a few times in a row without a single failure.

[ikolomi]

@jeremyprime

connections can be created/destroyed between the two measurements

how come?
this should not happen, we need to understand it because it might highlight a real problem

[jeremyprime]

@jeremyprime

connections can be created/destroyed between the two measurements

how come? this should not happen, we need to understand it because it might highlight a real problem

I believe this is due to the connection pooling. The connection pool maintains clients which have reconnects, keepalives, etc that continue to occur between the 2 getClientList() calls (plus the for loop is creating more connections on each getClientList() call) , causing the test to fail since the second count is larger than the first.