hive-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jason Dere <>
Subject Re: Review Request 60259: HIVE-16926 LlapTaskUmbilicalExternalClient should not start new umbilical server for every fragment request
Date Fri, 23 Jun 2017 23:08:19 GMT

This is an automatically generated e-mail. To reply, visit:

(Updated June 23, 2017, 11:08 p.m.)

Review request for hive, Sergey Shelukhin and Siddharth Seth.


- Umbilical token should be same for all fragments of the same request
- Minor restructuring of retried requests
- Minor renaming

Bugs: HIVE-16926

Repository: hive-git


Initial patch, restructured the LlapTaskUmbilicalExternalClient code a bit.
- Uses shared LLAP umbilical server rather than a new server per external client
- Retries rejected submissions (WorkSubmitter helper class)
- No more deferred cleanup (from HIVE-16652). One thing about this is that once clients are
closed/unregistered, communicator.stop() is called and it's removed from the registered list
of clients. So we might get a few warning messages about untracked taskAttemptIds coming in
during heartbeat() .. if this is undesirable we might be able to leave them in the registeredClients
list (but ignore heartbeats to them as they are tagged as closed), and remove them using the
HeartbeatCheckTask once they get too old.

Diffs (updated)

  llap-client/src/java/org/apache/hadoop/hive/llap/ 7fff147 
  llap-client/src/java/org/apache/hadoop/hive/llap/ 95b0ffc 
  llap-ext-client/src/java/org/apache/hadoop/hive/llap/ eb93241 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/ d4ec44e 





Jason Dere

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message