Fred Krone created GEODE-3968:
---------------------------------
Summary: Document how rebalance actually works
Key: GEODE-3968
URL: https://issues.apache.org/jira/browse/GEODE-3968
Project: Geode
Issue Type: Task
Components: regions
Reporter: Fred Krone
There are a lot of user questions around how rebalance works, configuration etc.
Example from Gideon: I still think we need to make important improvements to
the rebalancing documentation. One "big picture" item is to explain the
resource manager's role in rebalancing activity (RM isn't mentioned in the
doc's in this context).
We should also add more detail explaining how to optimize multi-threaded
rebalancing.
What exactly happens with multi-threaded rebalancing, and what are the limits
and/or consequences? For example, could too high a degree of parallelism
defeat the rebalancing algorithm (given the original design targets reaching
the right end-state by moving buckets one at-a-time)? Or, is the reverse true,
and we might bet better final results with more threads ? Is there some
guideline we can devise based on the number of nodes in the cluster, the number
of cores per server, and the configured number of buckets for a PR? Is
multi-threaded applied on each host individually, or are the extra threads only
running on the rebalance "coordinator" node?
I wasn't able to explain any of the nuances of rebalancing based on our docs
and cursory review of the Geode codebase (although for the latter I'm sure I
could eventually . . .). This actually screams for a dedicated section on the
Geode Wiki "Geode Internal Architecture" section . . . I would be happy to help
write this I can get get started with answers to the above questions.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
