Xiao Liu created HBASE-30505:
--------------------------------
Summary: Incorrect BYTES_IN_REMOTE_RESULTS scan metric in async
scan
Key: HBASE-30505
URL: https://issues.apache.org/jira/browse/HBASE-30505
Project: HBase
Issue Type: Bug
Components: Client
Reporter: Xiao Liu
Assignee: Xiao Liu
Fix For: 4.0.0-alpha-1, 2.7.0, 3.1.0, 3.0.1, 2.6.8, 2.5.17
{\{AsyncScanSingleRegionRpcRetryingCaller#onComplete}} passes
\{{isHeartbeatMessage}} as the \{{isRegionServerRemote}} argument of
\{{ConnectionUtils#updateResultsMetrics}}:
{code:java}
updateResultsMetrics(scanMetrics, rawResults, isHeartbeatMessage);
{code}
So whether results are counted in \{{BYTES_IN_REMOTE_RESULTS}} depends on
whether the response is a heartbeat, not on where the region server is:
* Results from a remote region server in a normal response are not counted. The
metric stays at or near 0.
* Results in a heartbeat response from a local region server are counted as
remote.
{\{BYTES_IN_RESULTS}} is not affected.
The bug was introduced by HBASE-17691. \{{ScannerCallable}}, changed in the
same commit, passes the correct flag. On branch-2, \{{Table}} scans still use
\{{ScannerCallable}}, so only \{{AsyncTable}} scans are affected. On branch-3
and master, \{{Table}} is implemented on top of \{{AsyncTable}}, so all scans
are affected. The wrong value shows up in \{{ResultScanner#getScanMetrics}} and
in the MapReduce job counters.
The fix is to pass \{{regionServerRemote}}, which the RPC call and retry
metrics in the same class already use. HBASE-30495 fixed a similar issue for
\{{REMOTE_RPC_RETRIES}}.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
