hadoop-yarn-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sangjin Lee <sj...@apache.org>
Subject [YARN-2928] working on rebase with trunk
Date Mon, 09 Nov 2015 19:36:04 GMT
I am working on rebasing the YARN-2928 branch (timeline service v.2) with
trunk. So far it seems fairly straightforward, but I noted a few things
that require some investigation and discussion.

(1) TestDistributedShell failure (and timing out)
It appears that after the rebase the TestDistributedShell test is unhappy.
If I isolate the v2 tests only, I have a sporadic failure (missing the
application created event). If I run the test in its entirety, it times out
after 15 minutes. I suspect the failure is caused by YARN-4129
<https://issues.apache.org/jira/browse/YARN-4129>. I haven't tracked down
the cause of the timeout yet, though. This may or may not be related with
YARN-2859 <https://issues.apache.org/jira/browse/YARN-2859> that changed
the port that the AHS binds to, and the subsequent TestDistributedShell
test change (YARN-4326 <https://issues.apache.org/jira/browse/YARN-4326>).
We should file a JIRA to look into this once the rebase is done.

(2) reconciling with YARN-4183
As Naga already pointed out on that JIRA, how we reconcile with the changes
made in YARN-4183 requires a discussion. It changed the config that drives
the RM's system metrics publisher from "timeline service enabled" to
"application history server enabled". I'm not entirely sure if that is the
right config to use for v.1, much less v.2. We should decide whether we
retain our version (continue to use "timeline service enabled"), and decide
how we enable v.1 and v.2 respectively. I think I'm going to preserve our
version in our branch for now, but we can certainly revisit it.

(3) reconciling with YARN-2556
YARN-2556 was merged into trunk that delivers the timeline service
performance tests. Since we already merged the v.2 version of the tests in
our branch, I'm going to spend some time to make sure we eliminate code
duplication. I'll file a JIRA to clean this up after the rebase.

Please let me know if you have other feedback. Thanks!


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