Release looks good overall except for complete latency and capacity.
Verified the md5, sha and asc
Unzipped and untared the source and binary releases.
Built from source and ran daemons from storm-dist binary release after packing
from the same.
Submitted FastWordCount and RollingTopWords
Played around with UI and tried different operations like Deactivate, Activate,
Kill, LogViewer, Search log, etc, . Everything works ok.
However I noticed complete latency for FastWordCount to be more than 4 seconds
which seems higher. Also, capacity for RollingTopWords (was verifying for
Jungtaek) seemed to be very high.
Running 1.1.1 to compare the same numbers for both the topologies. Will update
with my results.
On 1/24/18, 12:03 PM, "Alexandre Vermeerbergen" <avermeerber...@gmail.com>
wrote:
Hello Storm developers,
One of my team member (Noureddine Chatti) found that the regression in
"Assigned Memory (MB)" columns for our topologies with Storm 1.2.0rc1,
where the 65 MB value is displayed regardless of our actual
topologies's memory setting:
We define our workers's max memory heap using this option (as
documented in Hortonworks documentation here:
https://docs.hortonworks.com/HDPDocuments/HDP2/HDP-2.6.2/bk_storm-component-guide/content/config-storm-settings.html)
worker.childopts: "-Xmx2048m"
Until Storm 1.1.0, this value is displayed in "Assigned Memory (MB)"
columns by Nimbus UI.
In Storm 1.2.0 rc1, the 65 MB value is displayed regardless of this setting.
He also found that using the following setting:
topology.worker.max.heap.size.mb: 2048.
then the displayed "Assigned Memory (MB)" columns by Nimbus UI are in
line with the value (i.e. they show 2048MB).
Questions:
1. This is change of behavior from Storm 1.1.0 to Storm 1.2.0rc1
international? any related JIRA who would help understanding the
rationale?
2. Is the maximum heap size defined using worker.childopts actually
used to setup Worker's max heap size?
3. What's the best practice for setting worker's memory consumption :
is topology.worker.max.heap.size.mb now mandatory, or is the use of
-Xmx in worker.childopts still supported?
4. Could be there other differences from Storm 1.1.0 and Storm
1.2.0rc1 which could explain why we get very weird statistics in
Nimbus UI for Capacity & Latency for our topologies?
Best regards,
Alexandre Vermeerbergen
