Baoyuantop commented on issue #12748:
URL: https://github.com/apache/apisix/issues/12748#issuecomment-3525576645

   Simply dividing the configured global rate by the number of running nodes 
will under-deliver when traffic for a given key is sticky to a single node 
(common with client keep-alive or session-affinity load balancers); the user 
would get rate / nodes instead of the advertised global limit.
   
   Node counts derived from /data_plane/server_info/* are soft-state with TTL 
refresh and lag; scaling events, network partitions, or an unhealthy worker 
could lead to stale limits and either tighten or loosen throttling unexpectedly.
   
   Achieving an accurate cluster-wide budget without an external counter still 
requires some consistent coordination layer; Redis (or another shared store) is 
currently the only supported option that solves this problem deterministically.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to