DOYUNG YOON created S2GRAPH-68:
----------------------------------
Summary: Refactor write-write conflict resolving logic
Key: S2GRAPH-68
URL: https://issues.apache.org/jira/browse/S2GRAPH-68
Project: S2Graph
Issue Type: Improvement
Reporter: DOYUNG YOON
Assignee: DOYUNG YOON
Current implementation for resolving write-write conflict on snapshotEdge is
overwhelmingly complicated, so I am suggesting refactor this to make it easy to
understand.
Current master test cases failed with positive fail probability for
RPC(hbase.fail.prob=0.01).
I can reproduce it by running `sbt "project s2core"
-Dconfig.file=s2rest_play/conf/test.conf test`.
We can think this problem as critical section problem. Multiple thread that
want to mutate same SnapshotEdge can only proceed one-by-one to keep consistent
view on related IndexEdge. To achieve this, we are storing extra data on
SnapshotEdge.
Details are on
[gitbook|https://steamshon.gitbooks.io/s2graph-book/content/request_state_diagram.html].
I think following 3 things would be expected output after resolving this issue.
1. more comments and simplified code.
2. remove excessive re-fetch on every retry from partial failure. since it is
guaranteed that only one thread can proceed at a time, re-fetch is not
necessary.
3. remove Thread.sleep between retry from partial failure.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
