[
https://issues.apache.org/jira/browse/USERGRID-405?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Todd Nine updated USERGRID-405:
-------------------------------
Description:
Currently, performing a continuous PUT in 2.0 under heavy load causes a hotspot
in our cassandra data.
Cause:
# Under load, entities can be PUT continuously
# Asynchronous cleanups run and delete previous versions
# These versions are retained in cassandra for long periods of time. This
causes severe row bloating before compaction occurs.
Solution:
For entity data, we only care about the current max version. We should change
this column family to store only the maximum data format. We will need to keep
the log of previous versions, so that we can bring ES into a consistent state
was:
Currently, performing a continuous PUT in 2.0 under heavy load causes a hotspot
in our cassandra data.
Cause:
# Under load, our post processing pool becomes saturated
# As a result the entity cleanup task is dropped since read-repair will correct
the issue
> Fix hotspot from continuous writes of entities
> ----------------------------------------------
>
> Key: USERGRID-405
> URL: https://issues.apache.org/jira/browse/USERGRID-405
> Project: Usergrid
> Issue Type: Story
> Reporter: Todd Nine
> Assignee: Todd Nine
>
> Currently, performing a continuous PUT in 2.0 under heavy load causes a
> hotspot in our cassandra data.
> Cause:
> # Under load, entities can be PUT continuously
> # Asynchronous cleanups run and delete previous versions
> # These versions are retained in cassandra for long periods of time. This
> causes severe row bloating before compaction occurs.
> Solution:
> For entity data, we only care about the current max version. We should
> change this column family to store only the maximum data format. We will
> need to keep the log of previous versions, so that we can bring ES into a
> consistent state
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
