Infinity loop on pool max overflow

[alexandremcosta]

If we reach a pool max overflow limit, or if we configure pool_max_overflow with 0:

1. Client calls :eredis_cluster.q(["GET", "something"])
2. Which calls :eredis_cluster.query/3 on src/eredis_cluster.erl:180
3. Which calls :eredis_cluster_pool:transaction/2 on src/eredis_cluster.erl:186
4. Which has this code:

   transaction(PoolName, Transaction) ->
   try
       poolboy:transaction(PoolName, Transaction)
   catch
       exit:_ ->
           {error, no_connection}
   end.

5. And returns {error, no_connection} because we are out of poolboy workers
6. Then, :eredis_cluster.handle_transaction_result/2 patterns match the result
7. Which calls eredis_cluster_monitor:refresh_mapping(Version)
8. Which calls :eredis_cluster_monitor.reload_slots_map/1
9. Which calls :eredis_cluster_monitor.get_cluster_slots/1
10. Which calls :eredis:q(Connection, ["CLUSTER", "SLOTS"])
11. Which goes back to STEP 1, and starts the infinity loop

This triggers infinity error like:

(ErlangError) Erlang error: {:timeout, {:gen_server, :call, [:eredis_cluster_monitor, {:reload_slots_map, 1}]}}

Until we restart our application, which releases the pool of workers and our system is back...

[alexandremcosta]

Actually this is wrong, because the ["CLUSTER", "SLOTS"] call is done by eredis not eredis_cluster, so it does not trigger an infinity loop.