https://github.com/takuti/datadog-anomaly-detector
:dog: Anomaly detection system for Datadog multiple metrics
https://github.com/takuti/datadog-anomaly-detector
anomalydetection datadog fluentd norikra python
Last synced: about 2 months ago
JSON representation
:dog: Anomaly detection system for Datadog multiple metrics
- Host: GitHub
- URL: https://github.com/takuti/datadog-anomaly-detector
- Owner: takuti
- License: mit
- Created: 2016-08-03T01:34:43.000Z (almost 9 years ago)
- Default Branch: master
- Last Pushed: 2016-11-11T05:25:16.000Z (over 8 years ago)
- Last Synced: 2025-03-28T15:11:53.564Z (3 months ago)
- Topics: anomalydetection, datadog, fluentd, norikra, python
- Language: Python
- Homepage:
- Size: 332 KB
- Stars: 22
- Watchers: 4
- Forks: 5
- Open Issues: 7
-
Metadata Files:
- Readme: README.md
- Changelog: changefinder.sh
- License: LICENSE
Awesome Lists containing this project
README
Datadog Anomaly Detector
===
[](https://travis-ci.org/takuti/datadog-anomaly-detector)
Get Datadog metrics and pass anomaly scores to Datadog itself via Fluentd.
By integrating CEP engines such as [Esper](http://www.espertech.com/esper/) and [Norikra](http://norikra.github.io/), you can implement more practical applications as the following picture illustrates. We introduce it in **[doc/norikra.md](https://github.com/takuti/datadog-anomaly-detector/blob/master/doc/norikra.md)**.
## Minimal Requirements
### System
- Python 3.x (2.x is not supported)
- Fluentd 0.12.x
### Python packages
See **requirements.txt**
## Basic Installation and Usage
### 1. Setup Fluentd (td-agent)
Note: You can replace `td-agent` with `fluent` depending on your system environment.
Follow [Installation | Fluentd](http://docs.fluentd.org/categories/installation) and configure `/etc/td-agent/td-agent.conf` as:
```apache
@type copy
deep_copy true
@type record_reformer
renew_record true
renew_time_key time
tag datadog.${tag}
metric ${metric_outlier}
value ${score_outlier}
time ${record["time"]}
@type record_reformer
renew_record true
renew_time_key time
tag datadog.${tag}
metric ${metric_change}
value ${score_change}
time ${record["time"]}
@type dd
dd_api_key YOUR_API_KEY
```
Since the configuration depends on [fluent-plugin-dd](https://github.com/winebarrel/fluent-plugin-dd) and [fluent-plugin-record-reformer](https://github.com/sonots/fluent-plugin-record-reformer), you need to install the plugins via `td-agent-gem`.
Finally, restart td-agent: `$ sudo service restart td-agent`.
### 2. Configure your detector
Clone this repository:
$ git clone [email protected]:takuti/datadog-anomaly-detector.git
$ cd datadog-anomaly-detector
Create `config/datadog.ini` as demonstrated in `config/example.ini`.
$ cat config/datadog.ini
[general]
pidfile_path: /var/run/changefinder.pid
; Datadog API access interval (in sec. range)
interval: 600
[datadog.cpu]
query: system.load.norm.5{chef_environment:production,chef_role:worker6-staticip} by {host}
; ChangeFinder hyperparameters
r: 0.02
k: 7
T1: 10
T2: 5
[datadog.queue]
query: avg:queue.system.running{*}
r: 0.02
k: 7
T1: 10
T2: 5
You can insert a new config for a different query (metric) by creating a new **[datadog.xxx.yyy]** section as:
```
[datadog.add1]
query: additional.metric.1{foo}
r: 0.02
k: 7
T1: 10
T2: 5
...
```
Here, the above Fluentd configuration enables to create a new Datadog metrics **changefinder.outlier.xxx.yyy** and **changefinder.change.xxx.yyy*** for a configured section **[datadog.xxx.yyy]**. Since the names are very important to monitor the anomaly scores, you have to decide it carefully.
Note that `r`, `k`, `T1` and `T2` are the parameters of our machine learning algorithm. You can set different parameters for each query if you want. In case that you do not write the parameters on the INI file, default parameters will be set. In particular, optimal `k` is chosen by a model selection logic as described in **[doc/changefinder.md#model-selection](https://github.com/takuti/datadog-anomaly-detector/blob/master/doc/changefinder.md#model-selection)**.
### 3. Start a detector daemon
In order to get Datadog metrics, we need to first set API and APP keys as environmental variables `DD_APP_KEY` and `DD_API_KEY`.
Now, we are ready to start a detector daemon as:
```
$ python daemonizer.py start
```
For the `.pid` file specified in `config/datadog.ini`, please make sure if the directories exist correctly and you have write permission for the path.
You can stop the daemon as follows.
```
$ python daemonizer.py stop
```
## License
MIT