Topology refresh mechanisms

[SushmaReddyLoka]

In this issue @ #339 (comment) its mentioned that,

1. Adaptive refreshing only works in the scope of the data/nodes a client accesses. If you never experience a disconnect or a redirection, the client is not able to discover a change. That's ok for slot changes but in a case of a master/slave failover, that happens before the client opens a connection, there's currently no chance to discover that. I think lettuce could improve here by checking the cluster node role after connecting.
   Can you please elaborate on, failover case you are mentioning. Why lettuce has to determine role after connecting ?

2. Periodic refresh is a generic attempt to catch 'em all. It comes at the cost of scheduling and cluster utilization. The connection is established in a blocking fashion (again something lettuce could improve). Refreshing in a large cluster (say 200 nodes) previously opened up to 200 additional connections. Refreshing is now configurable to open only connections to the initially specified seed nodes. Because of the scheduling nature, it's not possible to obtain changes in real time, but they are seen always delayed – by the scheduling intervals.

• Can you please elaborate on cluster utilization, why does cluster utilization comes into picture.
• The connection is established in a blocking fashion (again something lettuce could improve).
  What does blocking mean here. are commands on the client blocked ?
  As you said there is a scope of improvement, was it done, if yes , is it available n 5.3.0 version.

[mp911de]

checking the cluster node role after connecting.

This is only a momentarily check that doesn't cover changes when the connection is already connected.

established in a blocking fashion

That's already changed.

again something lettuce could improve

Happy to review a pull request.

The connection is established in a blocking fashion (again something lettuce could improve).
What does blocking mean here. are commands on the client blocked ?

This means that a worker thread is blocked for the entire duration of a topology refresh run. This means, that other work (such as processing command timeouts) gets delayed until the topology refresh has completed. Lettuce 6 uses non-blocking refresh which improves on efficiency and removes the delay.

Let me know if this helps.

[mp911de]

Closing due to lack of requested feedback. If you would like us to look at this issue, please provide the requested information and we will re-open the issue.