If you just want to test out Singularity, consider using our docker setup instead.
1. Set up a Zookeeper cluster
Singularity uses Zookeeper as its primary datastore -- it cannot run without it.
Chef recipe: https://supermarket.chef.io/cookbooks/zookeeper
Puppet module: https://forge.puppetlabs.com/deric/zookeeper
More info on how to manually set up a Zookeeper cluster lives here
For testing or local development purposes, a single-node cluster running on your local machine is fine. If using the docker testing/development setup, this will already be present.
2. Set up MySQL (optional)
Singularity can be configured to move stale data from Zookeeper to MySQL after a configurable amount of time, which helps reduce strain on the cluster. If you're running Singularity in Production environment, MySQL is encouraged. See the database reference for help configuring the database.
3. Set up a Mesos cluster
Mesosphere provides a good tutorial for setting up a Mesos cluster: https://open.mesosphere.com/getting-started/install/. You can skip the section on setting up Marathon since Singularity will be our framework instead.
4. Build or download the Singularity JAR
In order to run Singularity, you can either build it from scratch or download a precompiled JAR for the
Building from Source
mvn clean package in the root of the Singularity repository. The SingularityService JAR will be created in
Downloading a precompiled JAR
Singularity JARs are published to Maven Central for each release. You can view the list of SingularityService (the executable piece of Singularity) JARs here
Be sure to only use the
shaded.jar links -- the other JARs won't work.
5. Create a Singularity Config File
Singularity requires a YAML file with some configuration values in order to start up. Here's an example:
# Run SingularityService on port 7099 and log to /var/log/singularity-access.log server: type: simple applicationContextPath: /singularity connector: type: http port: 7099 requestLog: appenders: - type: file currentLogFilename: /var/log/singularity-access.log archivedLogFilenamePattern: /var/log/singularity-access-%d.log.gz # omit this entirely if not using MySQL database: driverClass: com.mysql.jdbc.Driver user: [database username] password: [database password] url: jdbc:mysql://[database host]:[database port]/[database name] mesos: master: [comma separated list of mesos master host:port] defaultCpus: 1 defaultMemory: 128 defaultDisk: 1024 frameworkName: Singularity frameworkId: Singularity frameworkFailoverTimeout: 1000000 zookeeper: quorum: [comma separated host:port list of ZK hosts] zkNamespace: singularity sessionTimeoutMillis: 60000 connectTimeoutMillis: 5000 retryBaseSleepTimeMilliseconds: 1000 retryMaxTries: 3 logging: loggers: "com.hubspot.singularity" : TRACE enableCorsFilter: true sandboxDefaultsToTaskId: false # enable if using SingularityExecutor ui: title: Singularity (local) baseUrl: http://localhost:7099/singularity # If hosting singularity on another domain, use that domain instead of localhost
Full configuration documentation lives here: configuration.md
6. Run MySQL migrations (if necessary)
If you're operating Singularity with MySQL, you first need to run a liquibase migration to create all appropriate tables: (this snippet assumes your Singularity configuration YAML exists as
java -jar SingularityService/target/SingularityService-*-shaded.jar db migrate singularity_config.yaml --migrations mysql/migrations.sql
It's a good idea to run a migration each time you upgrade to a new version of Singularity.
7. Start Singularity
java -jar SingularityService/target/SingularityService-*-shaded.jar server singularity_config.yaml
Warning: Singularity will abort (i.e. exit) when it hits an unrecoverable error, so it's a good idea to use monit, supervisor, or systemd to monitor and automatically restart the Singularity process. It does this in order to simplify the handling of state which must be kept in sync with the Mesos master.
Once started the Singularity UI will be available at http://(host):7099/singularity
8. Install extra Singularity tools on Mesos slaves (optional)
Singularity ships with a custom Mesos executor and extra background jobs to make running tasks easier. You can find more info about the slave extras here.