| Log Message: |
[jira] [HBASE-5744] Thrift server metrics should be long instead of int
Summary:
As we measure our Thrift call latencies in nanoseconds, we need to make
latencies long instead of int everywhere. There is a bug where we truncate a
nanosecond latency to int, which is a problem with RPCs that take more than
2.147483647 seconds to process.
This is a trunk diff. The 89-fb version of this diff is at D2679.
Test Plan:
TestThriftServer is updated to test for the failure case (an RPC is artificially
made to take 3 seconds). The new test case fails without the fix.
Re-run all unit tests.
Reviewers: stack, sc, Kannan, JIRA
Reviewed By: sc
Differential Revision: https://reviews.facebook.net/D2685
|
hbase/trunk/src/main/java/org/apache/hadoop/hbase/thrift/HbaseHandlerMetricsProxy.java