Yang Jie created HDFS-18020:
-------------------------------
Summary: AddBlockFlag.NO_LOCAL_RACK is dropped over RPC because
AddBlockFlagProto has no value 3
Key: HDFS-18020
URL: https://issues.apache.org/jira/browse/HDFS-18020
Project: Hadoop HDFS
Issue Type: Bug
Components: hdfs-client
Reporter: Yang Jie
HDFS-13739 added {{CreateFlag.NO_LOCAL_RACK}} and
{{AddBlockFlag.NO_LOCAL_RACK}} (mode 3) together with the placement logic in
BlockPlacementPolicyDefault, but {{AddBlockFlagProto}} in
ClientNamenodeProtocol.proto only has NO_LOCAL_WRITE = 1 and
IGNORE_CLIENT_LOCALITY = 2. PBHelperClient#convertAddBlockFlags maps each flag
with {{AddBlockFlagProto.forNumber(flag.getMode())}} and skips nulls, so the
client silently drops NO_LOCAL_RACK and the NameNode never sees it.
{{HdfsDataOutputStreamBuilder#noLocalRack()}} has had no effect over RPC since
it was added in 3.3.0. The HDFS-13739 test calls the in-process
NameNodeRpcServer directly, which skips the protobuf conversion.
The fix adds {{NO_LOCAL_RACK = 3}} to the proto enum. An older NameNode or
Router keeps the unknown value in the unknown field set, which behaves as
today. Router-based federation uses the same converter on both hops, so it gets
the fix too.
The test now checks that every AddBlockFlag has a proto value with the same
number and name, so a future flag added without a proto value fails it. The
change also corrects the javadoc of AddBlockFlag.NO_LOCAL_RACK, which was
copied from NO_LOCAL_WRITE.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]