[
https://issues.apache.org/jira/browse/YARN-4676?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Daniel Zhi updated YARN-4676:
-----------------------------
Attachment: YARN-4676.021.patch
YARN-4676.021.patch address some of the most recent comments.
1. removed doc GracefulDecommission.md
2. The recent HostsFileReader change to use readLock and expose a copy of the
content to caller lead to inefficient code like isNodeValid() inside
NodesListManager, where a copy of the content is created for every single host
check. That said, getExcludedHostsWithTimeout() suppose to be removed given the
lock protected getHostDetails(). I have fixed NodesListManager.isNodeValid() to
be efficient as well.
3. done to support list of hosts in <name> tag
4. all the read methods inside HostFileReader throws IOException because
fileInputStream.close(); inside the finally block may throw IOException;
5. updated usage for -g|graceful.
6. updated comments hopefully it help. Basically a node without any running
container might had map container for an application that is still running. The
cost to track
"list of running applications that I was part of historically" exceeds the
benefit so currently such "idle" node will be DECOMMISSIONED and the affected
map task will be rescheduled. If such info in future can be efficient
tracked/obtained from RmNodeImpl, then DecommissioningNodesWatcher can leverage
it.
7. the 60 seconds delay removal is to prevent such node from suddenly disappear
from the status log (currently debug-mode only) but instead will appear as
DECOMMISSIONED before it is removed.
8. switched to use MonotonicClock;
9. 1) PollTimerTask is by default scheduled to run once every 20 seconds,
without initial delay. There is no tight loop;
2) normally NM heartbeat every second so the timeout tracking is every
second, the stale check logic guarantee a check for node missing heartbeat
update for 30 seconds. The logic only matters to terminated instance. That
said, I don't see downside to reduce 30 second to 5 second, as long as to avoid
node being regularly checked already.
3) I don't think I fully understand this point.
Note: I will be busy on new projects and won't be able to afford further
iterations on this JIRA unless it is an immediate bug and within next two weeks.
> Automatic and Asynchronous Decommissioning Nodes Status Tracking
> ----------------------------------------------------------------
>
> Key: YARN-4676
> URL: https://issues.apache.org/jira/browse/YARN-4676
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: resourcemanager
> Affects Versions: 2.8.0
> Reporter: Daniel Zhi
> Assignee: Daniel Zhi
> Labels: features
> Attachments: GracefulDecommissionYarnNode.pdf,
> GracefulDecommissionYarnNode.pdf, YARN-4676.004.patch, YARN-4676.005.patch,
> YARN-4676.006.patch, YARN-4676.007.patch, YARN-4676.008.patch,
> YARN-4676.009.patch, YARN-4676.010.patch, YARN-4676.011.patch,
> YARN-4676.012.patch, YARN-4676.013.patch, YARN-4676.014.patch,
> YARN-4676.015.patch, YARN-4676.016.patch, YARN-4676.017.patch,
> YARN-4676.018.patch, YARN-4676.019.patch, YARN-4676.020.patch,
> YARN-4676.021.patch
>
>
> YARN-4676 implements an automatic, asynchronous and flexible mechanism to
> graceful decommission
> YARN nodes. After user issues the refreshNodes request, ResourceManager
> automatically evaluates
> status of all affected nodes to kicks out decommission or recommission
> actions. RM asynchronously
> tracks container and application status related to DECOMMISSIONING nodes to
> decommission the
> nodes immediately after there are ready to be decommissioned. Decommissioning
> timeout at individual
> nodes granularity is supported and could be dynamically updated. The
> mechanism naturally supports multiple
> independent graceful decommissioning “sessions” where each one involves
> different sets of nodes with
> different timeout settings. Such support is ideal and necessary for graceful
> decommission request issued
> by external cluster management software instead of human.
> DecommissioningNodeWatcher inside ResourceTrackingService tracks
> DECOMMISSIONING nodes status automatically and asynchronously after
> client/admin made the graceful decommission request. It tracks
> DECOMMISSIONING nodes status to decide when, after all running containers on
> the node have completed, will be transitioned into DECOMMISSIONED state.
> NodesListManager detect and handle include and exclude list changes to kick
> out decommission or recommission as necessary.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
