a7112fd30b
The main improvement comes from using the Influxdb Line Protocol. The
encoding methods in line_utils.py are like the ones used in the influxdb
client but optimized for our data
Additional improvement comes from avoiding calls to encode('utf8') as
the influxdb client already does that.
On my test system, these changes increased the number of measurements
processed from about 2200/second to about 3700/second. Measurement
processing time is now dominated by Kafka. Approximately, 35% of time
is spent reading from Kafka and approximately 22% of time is committing
offsets. Only 10% of the time is spent writing to Influxdb. About 30% of
the time is spent converting messages from the json string read from
Kafka into the Line Protocol format for Influxdb.
Once monasca-common is modified to use the faster kafka library,
performance should be even better.
I did try using ujson, but my tests showed it wasn't any faster than
the json package.
Change-Id: I2acf76d9a5f583c74a272e18350b9c0ad5883f95
|
||
|---|---|---|
| common | ||
| etc/monasca | ||
| java | ||
| monasca_persister | ||
| tools | ||
| .gitignore | ||
| .gitreview | ||
| .testr.conf | ||
| LICENSE | ||
| README.md | ||
| pom.xml | ||
| requirements.txt | ||
| run_maven.sh | ||
| setup.cfg | ||
| setup.py | ||
| test-requirements.txt | ||
| tox.ini | ||
README.md
Team and repository tags
monasca-persister
The Monitoring Persister consumes metrics and alarm state transitions from the Message Queue and stores them in the Metrics and Alarms database.
Although the Persister isn't primarily a Web service it uses DropWizard, https://dropwizard.github.io/dropwizard/, which provides a nice Web application framework to expose an http endpoint that provides an interface through which metrics about the Persister can be queried as well as health status.
The basic design of the Persister is to have one Kafka consumer publish to a Disruptor, https://github.com/LMAX-Exchange/disruptor, that has output processors. The output processors use prepared batch statements to write to the Metrics and Alarms database.
The number of output processors/threads in the Persister can be specified to scale to more messages. To horizontally scale and provide fault-tolerance any number of Persisters can be started as consumers from the Message Queue.
Build
Requires monasca-common from https://github.com/openstack/monasca-common. Download and build following instructions in its README.md. Then build monasca-persister by:
mvn clean package
Configuration
A sample configuration file is available in java/src/deb/etc/persister-config.yml-sample.
A second configuration file is provided in java/src/main/resources/persister-config.yml for use with the vagrant "mini-mon" development environment.
TODO
- Purge metrics on shutdown
- Add more robust offset management in Kafka. Currently, the offset is advanced as each message is read. If the Persister stops after the metric has been read and prior to it being committed to the Metrics and Alarms database, the metric will be lost.
- Add better handling of SQL exceptions.
- Complete health check.
- Specify and document the names of the metrics that are available for monitoring of the Persister.
- Document the yaml configuration parameters.
License
Copyright (c) 2014 Hewlett-Packard Development Company, L.P.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.