Details
-
Sub-task
-
Status: Resolved
-
Minor
-
Resolution: Not A Problem
-
None
-
None
-
None
Description
Came up during the review of ACCUMULO-378:
ReplicationClient, MasterClient, and ThriftUtil are all classes with lots of static methods used to make executing RPC against a remote service much easier with nice constructs (like getting a typed return object).
The downside is that they have lots of implicit gotchas contained (implicit retries, network failure handling, etc), which make them easier to use by the invoking code, but very complex beasts.
It would be good to consolidate what we have into a common base, and then provide strong semantics on each operation that encapsulates how we want each to operate (master ops should retry quickly, replication ops should retry slowly and give up eventually, etc), and also allow for testing of these intended implementations.