openstack/openstack-manuals
Code Issues Proposed changes

Add telemetry best practices to the admin guide

Browse Source 
Closes-Bug: #1460824

Change-Id: I73f60610450eb08a61b7150685901ca93065b6ec
This commit is contained in:
Pradeep Kilambi 2015-06-01 16:49:09 -04:00
parent a112a0b5bc
commit 6a26fff4d3
2 changed files with 178 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
doc/admin-guide-cloud/ch_telemetry.xml
View File

@ -51,4 +51,5 @@
<xi:include href="telemetry/section_telemetry-measurements.xml"/>
<xi:include href="telemetry/section_telemetry-events.xml"/>
<xi:include href="telemetry/section_telemetry-troubleshooting-guide.xml"/>
<xi:include href="telemetry/section_telemetry-best-practices.xml"/>
</chapter>

177
doc/admin-guide-cloud/telemetry/section_telemetry-best-practices.xml Normal file
View File

@ -0,0 +1,177 @@
<?xml version="1.0" encoding="UTF-8"?>
<section xmlns="http://docbook.org/ns/docbook"
xmlns:xi="http://www.w3.org/2001/XInclude"
xmlns:xlink="http://www.w3.org/1999/xlink"
version="5.0"
xml:id="section_telemetry-best-practices">
<title>Telemetry best practices</title>
<para>
The following are some suggested best practices to follow when deploying
and configuring the Telemetry service. The best practices are divided into
data collection and storage.
</para>
<section xml:id="section_telemetry-data-collection-best-practices">
<title>Data collection</title>
<procedure>
<step>
<para>
The Telemetry module collects a continuously growing set of data.
Not all the data will be relevant for a cloud administrator to monitor.
</para>
<itemizedlist>
<listitem>
<para>Based on your needs, you can edit the <filename>pipeline.yaml</filename>
configuration file to include a selected number of meters while disregarding
the rest.</para>
</listitem>
<listitem>
<para>By default, Telemetry service polls the service APIs every 10 minutes.
You can change the polling interval on a per meter basis by editing the
<filename>pipeline.yaml</filename> configuration file.</para>
<warning>
<para>
If the polling interval is too short, it will likely cause increase of stored data
and the stress on the service APIs.
</para>
</warning>
</listitem>
<listitem>
<para>Expand the configuration to have greater control over different meter intervals.</para>
</listitem>
</itemizedlist>
<note>
<para>
For more information, see the <link xlink:href=
" http://docs.openstack.org/admin-guide-cloud/content/section_telemetry-pipeline-configuration.html">
Pipeline Configuration Options</link>.
</para>
</note>
</step>
<step>
<para>
If you are using the Kilo version of Telemetry, you can delay or adjust polling requests by enabling
the jitter support. This adds a random delay on how the polling agents send requests to the
service APIs. To enable jitter, set <option>shuffle_time_before_polling_task</option>
in the <filename>ceilometer.conf</filename> configuration file to an integer greater than 0.
</para>
</step>
<step>
<para>
If you are using Juno or later releases, based on the number of resources that will be polled,
you can add additional central and compute agents as necessary. The agents are designed to scale horizontally.
</para>
<note>
<para>
For more information see, <link xlink:href=
"http://docs.openstack.org/admin-guide-cloud/content/section_telemetry-cetral-compute-agent-ha.html">
HA deployment for Central and Compute Agents</link>.
</para>
</note>
</step>
<step>
<para>
If you are using Juno or later releases, use the <literal>notifier://</literal> publisher rather
than <literal>rpc://</literal> as there is a certain level of overhead that comes with RPC.
</para>
<note>
<para>
For more information on RPC overhead, see <link xlink:href=
"https://www.rabbitmq.com/tutorials/tutorial-six-python.html">
RPC overhead info</link>.
</para>
</note>
</step>
</procedure>
</section>
<section xml:id="section_telemetry-data-storage-best-practices">
<title>Data storage</title>
<procedure>
<step>
<para>
We recommend that you avoid open-ended queries. In order to get better performance you can use
reasonable time ranges and/or other query constraints for retrieving measurements.
</para>
<para>
For example, this open-ended query might return an unpredictable amount of data:
<screen><prompt>$</prompt><userinput>ceilometer sample-list --meter cpu -q 'resource_id=INSTANCE_ID_1'</userinput></screen>
Whereas, this well-formed query returns a more reasonable amount of data, hence better performance:
<screen><prompt>$</prompt><userinput>ceilometer sample-list --meter cpu -q 'resource_id=INSTANCE_ID_1;timestamp&gt;2015-05-01T00:00:00;timestamp&lt;2015-06-01T00:00:00'</userinput>
</screen>
</para>
</step>
<step>
<para>
You can install the API behind <literal>mod_wsgi</literal>, as it provides more settings to tweak,
like <literal>threads</literal> and <literal>processes</literal> in case of <literal>WSGIDaemon</literal>.
</para>
<note>
<para>
For more information on how to configure <literal>mod_wsgi</literal>, see the <link xlink:href=
"http://docs.openstack.org/developer/ceilometer/install/mod_wsgi.html">
Install Documentation</link>.
</para>
</note>
</step>
<step>
<para>
The collection service provided by the Telemetry project is not intended to be
an archival service. Set a Time to Live (TTL) value to expire data and minimize
the database size. If you would like to keep your data for longer time period,
you may consider storing it in a data warehouse outside of Telemetry.
</para>
<note>
<para>
For more information on how to set the TTL, see <link xlink:href=
"http://docs.openstack.org/admin-guide-cloud/content/section_telemetry-storing-data.html">
TTL support for various back ends</link>.
</para>
</note>
</step>
<step>
<para>
We recommend that you do not use SQLAlchemy back end prior to the Juno release, as it previously contained extraneous
relationships to handle deprecated data models. This resulted in extremely poor query
performance.
</para>
</step>
<step>
<para>
We recommend that you do not run MongoDB on the same node as the controller. Keep it on a separate
node optimized for fast storage for better performance. Also it is advisable for the MongoDB
node to have a lot of memory.
</para>
<note>
<para>
For more information on how much memory you need, see <link xlink:href=
"http://docs.mongodb.org/manual/faq/diagnostics/#how-do-i-calculate-how-much-ram-i-need-for-my-application">
MongoDB FAQ</link>.
</para>
</note>
</step>
<step>
<para>
Use replica sets in MongoDB. Replica sets provide high availability through
automatic failover. If your primary node fails, MongoDB will elect a secondary
node to replace the primary node, and your cluster will remain functional.
</para>
<para>
For more information on replica sets, see the <link xlink:href=
"http://docs.mongodb.org/manual/tutorial/deploy-replica-set/">
MongoDB replica sets docs</link>.
</para>
</step>
<step>
<para>
Use sharding in MongoDB. Sharding helps in storing data records across
multiple machines and is the MongoDB’s approach to meet the demands of data growth.
</para>
<para>
For more information on sharding, see the <link xlink:href=
"http://docs.mongodb.org/manual/sharding/">
MongoDB sharding docs</link>.
</para>
</step>
</procedure>
</section>
</section>