Github user revans2 commented on a diff in the pull request:
https://github.com/apache/storm/pull/2319#discussion_r138659373
--- Diff: storm-core/src/clj/org/apache/storm/cluster.clj ---
@@ -244,15 +255,19 @@
;; Watches should be used for optimization. When ZK is reconnecting,
they're not guaranteed to be called.
(defnk mk-storm-cluster-state
- [cluster-state-spec :acls nil :context (ClusterStateContext.)]
- (let [[solo? cluster-state] (if (instance? ClusterState
cluster-state-spec)
- [false cluster-state-spec]
- [true (mk-distributed-cluster-state
cluster-state-spec :auth-conf cluster-state-spec :acls acls :context context)])
+ [conf :cluster-state nil :acls nil :context (ClusterStateContext.)
:backend nil]
+ (let [[solo? cluster-state] (if (and (not-nil? cluster-state) (instance?
ClusterState cluster-state))
+ [false cluster-state]
+ [true (mk-distributed-cluster-state conf
:auth-conf conf :acls acls :context context)])
+ assignments-backend (if (nil? backend)
+ (doto (InMemoryAssignmentBackend.) (.prepare
nil nil))
--- End diff --
Why do we call `(.prepare nil nil)` only on the InMemoryAssignmentBackend?
It feels like we might be making it difficult to extend it in the future. From
what I have seen the assignments dir is a subdirectory of storm-local. Could
we just make prepare take the conf, and then we can pass in the conf we already
have.
---
