Sherlock is an anomaly detection service built on top of Druid. It leverages EGADS (Extensible Generic Anomaly Detection System) to detect anomalies in time-series data. Users can schedule jobs on an hourly, daily, weekly, or monthly basis, view anomaly reports from Sherlock's interface, or receive them via email.
Timeseries generation is the first phase of Sherlock's anomaly detection. The user inputs a full Druid JSON query with a metric name and group-by dimensions. Sherlock validates the query, adjusts the time interaval and granularity based on the EGADS config, and makes a call to Druid. Druid responds with an array of time-series, which are parsed into EGADS time-series.
{
"metric": "metric(metric1/metric2)",
"aggregations": [
{
"filter": {
"fields": [
{
"type": "selector",
"dimension": "dim1",
"value": "value1"
}
],
"type": "or"
},
"aggregator": {
"fieldName": "metric2",
"type": "longSum",
"name": "metric2"
},
"type": "filtered"
}
],
"dimension": "groupByDimension",
"intervals": "2017-09-10T00:00:01+00:00/2017-10-12T00:00:01+00:00",
"dataSource": "source1",
"granularity": {
"timeZone": "UTC",
"type": "period",
"period": "P1D"
},
"threshold": 50,
"postAggregations": [
{
"fields": [
{
"fieldName": "metric1",
"type": "fieldAccess",
"name": "metric1"
}
],
"type": "arithmetic",
"name": "metric(metric1/metric2)",
"fn": "/"
}
],
"queryType": "topN"
}[ {
"timestamp" : "2017-10-11T00:00:00.000Z",
"result" : [ {
"groupByDimension" : "dim1",
"metric(metric1/metric2)" : 8,
"metric1" : 128,
"metric2" : 16
}, {
"groupByDimension" : "dim2",
"metric(metric1/metric2)" : 4.5,
"metric1" : 42,
"metric2" : 9.33
} ]
}, {
"timestamp" : "2017-10-12T00:00:00.000Z",
"result" : [ {
"groupByDimension" : "dim1",
"metric(metric1/metric2)" : 9,
"metric1" : 180,
"metric2" : 20
}, {
"groupByDimension" : "dim2",
"metric(metric1/metric2)" : 5.5,
"metric1" : 95,
"metric2" : 17.27
} ]
} ]Sherlock calls the user-configured EGADS API for each generated time-series, generates anomaly reports from the response, and stores these reports in a database. Users may also elect to receive anomaly reports by email.
Sherlock uses a Redis backend Redis to store job metadata, generated anomaly reports, among other information, and as a persistent job queue. Keys related to Reports have retention policy. Hourly job reports have retention of 14 days and daily/weekly/monthly job reports have 1 year of retention.
Sherlock's user interface is built with Spark. The UI enables users to submit instant anomaly analyses, create and launch detection jobs, view anomalies on a heatmap, and on a graph.
A Makefile is provided with all build targets.
make jarThis creates sherlock.jar in the target/ directory.
Sherlock is run through the commandline with config arguments.
java -Dlog4j.configuration=file:${path_to_log4j}/log4j.properties \
-jar ${path_to_jar}/sherlock.jar \
--version $(VERSION) \
--project-name $(PROJECT_NAME) \
--port $(PORT) \
--enable-email \
--failure-email $(FAILURE_EMAIL) \
--from-mail $(FROM_MAIL) \
--reply-to $(REPLY_TO) \
--smtp-host $(SMTP_HOST) \
--interval-hours $(INTERVAL_HOURS) \
--interval-days $(INTERVAL_DAYS) \
--interval-weeks $(INTERVAL_WEEKS) \
--interval-months $(INTERVAL_MONTHS) \
--egads-config-filename $(EGADS_CONFIG_FILENAME) \
--redis-host $(REDIS_HOSTNAME) \
--redis-port $(REDIS_PORT) \
--execution-delay $(EXECUTION_DELAY)| args | required | default | description |
|---|---|---|---|
| --help | - | false |
help |
| --config | - | null |
config |
| --version | - | v0.0.0 |
version |
| --egads-config-filename | - | provided |
egads-config-filename |
| --port | - | 4080 |
port |
| --interval-hours | - | 672 |
interval-hours |
| --interval-days | - | 28 |
interval-days |
| --interval-weeks | - | 12 |
interval-weeks |
| --interval-months | - | 6 |
interval-months |
| --enable-email | - | false |
enable-email |
| --from-mail | if email enabled |
from-mail | |
| --reply-to | if email enabled |
reply-to | |
| --smtp-host | if email enabled |
smtp-host | |
| --smtp-port | - | 25 |
smtp-port |
| --failure-email | if email enabled |
failure-email | |
| --execution-delay | - | 30 |
execution-delay |
| --valid-domains | - | null |
valid-domains |
| --redis-host | - | 127.0.0.1 |
redis-host |
| --redis-port | - | 6379 |
redis-port |
| --redis-ssl | - | false |
redis-ssl |
| --redis-timeout | - | 5000 |
redis-timeout |
| --redis-password | - | - | redis-password |
| --redis-clustered | - | false |
redis-clustered |
| --project-name | - | - | project-name |
| --external-file-path | - | - | external-file-path |
| --debug-mode | - | false |
debug-mode |
Prints commandline argument help message.
Path to a Sherlock configuration file, where the above configuration may be specified. Config arguments in the file override commandline arguments.
Version of sherlock.jar to display on the UI
Path to a custom EGADS configuration file. If none is specified, the default configuration is used.
Port on which to host the Spark application.
Number of historic data points to use for detection on hourly time-series.
Number of historic data points to use for detection on daily time-series.
Number of historic data points to use for detection on weekly time-series.
Number of historic data points to use for detection on monthly time-series.
Enable the email service. This enables users to receive email anomaly report notifications.
The handle's FROM email displayed to email recipients.
The handle's REPLY TO email where replies will be sent.
The email service's SMTP HOST.
The email service's SMTP PORT. The default value is 25.
A dedicated email which may be set to receive job failure notifications.
Sherlock periodically pings Redis to check scheduled jobs. This sets the ping delay in seconds. Jobs are scheduled with a precision of one minute.
A comma-separated list of valid domains to receive emails, e.g. 'yahoo,gmail,hotmail'. If specified, Sherlock will restrict who may receive emails.
The Redis backend hostname.
The Redis backend port.
Whether Sherlock should connect to Redis via SSL.
The Redis connection timeout.
The password to use when authenticating to Redis.
Whether the Redis backend is a cluster.
Name of the project to display on UI.
Specify the path to external files for Spark framework via this argument.
Debug mode enables debug routes. Ex. '/DatabaseJson' (shows redis data as json dump). Look at com.yahoo.sherlock.App for more details.
Jigar Patel, jigsdevbox@gmail.com
Jeff Niu, jeffniu22@gmail.com
Josh Walters, josh@joshwalters.com
Code licensed under the GPL v3 License. See LICENSE file for terms.