Github user hanm commented on the issue:
https://github.com/apache/zookeeper/pull/580
We at Twitter manage large ZK clusters and we also have our own internal
metrics system (you can't really operate things like ZK without metrics at
large scale.). Our design is similar like this in terms of metrics collection
and code instrumentation, but we coupled the metrics collection with
publication. We are publishing metrics through Finagle (Twitter's RPC library)
and the tight coupling caused several issues, such as circular dependencies
between Finagle and ZooKeeper, where ZK depends on Finagle for metrics, and
Finagle depends on ZK for ServerSets (naming resolution).
I think the current design proposed in the community would solve this
problem. Basically we will have two systems:
* The ZooKeeper metrics system for metrics collection.
* The pluggable system @eolivelli is working on is the backends for the
metrics.
So IMO both work are sort of orthogonal and can be done in parallel. And I
agree with the design decisions @jtuple proposed and @eolivelli commented
previously.
Regarding use an external metrics type system: I think it would be more
flexible for us to define and own the metrics definitions so it's easy to add /
change the system without depends on third parties.
---
