[
https://issues.apache.org/jira/browse/SAMZA-2597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Bharath Kumarasubramanian updated SAMZA-2597:
---------------------------------------------
Priority: Minor (was: Blocker)
> Maintain traversal order for registered operators
> -------------------------------------------------
>
> Key: SAMZA-2597
> URL: https://issues.apache.org/jira/browse/SAMZA-2597
> Project: Samza
> Issue Type: Bug
> Reporter: Bharath Kumarasubramanian
> Assignee: Bharath Kumarasubramanian
> Priority: Minor
>
> *Problem*: Currently, we have a deterministic way of creating the operator
> DAG by having a LinkedHashMap so that during our runtime, we ensure the
> lifecycle of operators follow a deterministic order.
> While we use the same order to traverse the graph and create the DAG, we lose
> the order within the sub DAG as the registered operators is a HashSet. The
> implication is result of an operator is dispatched non-deterministically to
> its sub-DAG. i.e
> Op A --> Op B --> Op C
> |--> Op D --> Op E
> Output of Op A can be dispatched to Op B or Op D depending how we iterate the
> `registeredOperators` set of Op A.
> While this is not a guarantee Samza provides to applications, we want to be
> consistent with graph traversal order, DAG insertion order and DAG traversal
> order.
> *Change*: Use `LinkedHashSet` instead of `HashSet` to make it consistent.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)