Skip to content
This repository has been archived by the owner on Apr 24, 2023. It is now read-only.

Latest commit

 

History

History
77 lines (55 loc) · 5.1 KB

README.adoc

File metadata and controls

77 lines (55 loc) · 5.1 KB
Raw

Cook Scheduler

The Cook scheduler is a Mesos framework that is able to schedule jobs submitted to it via its REST API from multiple users. It’s raison d’etre is to prioritize the user experience on a shared cluster. It does so by allowing users to use as much resources as possible, preempting users with greater shares when new users show up, and allowing users to have new jobs jump to the head of their queue. You can learn more about the algorithms used by Cook in this talk by Li Jin at MesosCon: https://www.youtube.com/watch?v=BkBMYUe76oI.

Building

You can build the final jar by running lein uberjar.

Unit Tests

You can run the Cook scheduler unit tests with lein test. If you want to see log output on the console, you can use the test-console profile, e.g. lein with-profile +test-console test. Console logging is accomplished by overriding the :begin-test-var variant of the report multimethod. The override is defined in the cook.test.testutil namespace, which is required by many of the actual test namespaces. If you are running tests only for a specific namespace that doesn’t require it, you can add the require in order to get console output.

Running

You can run the scheduler using the build system by running lein run $config, where config is an .edn file. See dev-config.edn for example usage. You should be able to start a sample instance (without any persistence) using the dev-config.edn. You’ll need to have the environment variable MESOS_NATIVE_JAVA_LIBRARY set.

Tip
What should I set MESOS_NATIVE_JAVA_LIBRARY to?

MESOS_NATIVE_JAVA_LIBRARY is supposed to point to the libmesos shared object file, which is libmesos.so on Linux, and libmesos.dylib on Mac. If you’re using Linux and running with the Mesosphere packages, you should use /usr/lib/libmesos.so. If you’re on a Mac and you compiled Mesos from source, you should use $MESOS_REPO/src/.libs/libmesos.dylib.

When Cook starts running, it prints out as it initializes itself. Once logging is initialized, all further logging data will be written to the log file, not stdout or stderr. libmesos, the Mesos client library, prints all of its logging to stdout and stderr. Thus, all messages printed to stdout and stderr are about the connection to Mesos.

To deploy the scheduler, simply build a jar as described above, and then run it with java -cp target/cook-0.1.0-SNAPSHOT-standalone.jar cook.components dev-config.edn. See the Asciidoc document docs/configuration.adoc for details on the config file format and options.

API

Cook comes with two APIs: the REST API, and the Datomic API. To use the REST API, see the docs/scheduler-rest-api.adoc Asciidoc document for the API description. Most users should use the REST API, since it’s compatible and will not change.

For advanced integration with JVM-based projects, you can write new jobs directly to Datomic, and Cook will automatically detect these jobs and scheduler them. To use this API, you should already be familiar with (Datomic)[docs.datomic.com]. You can see how to create a new job and query it by looking at the implementation of the REST API in src/cook/mesos/api.clj. Job creation is done in the submit-jobs function, and job querying is done in fetch-job-map. The direct Datomic access is especially useful for computing analytics on the Cook scheduler’s behavior. Because of Datomic’s architecture, you can spin up a Datomic peer and freely do heavy full-db scans and analytics on the production database without impacting the performance of the Cook server.

Dependencies

Cook is written in Clojure. To develop Cook, all you need is a JVM and Mesos installed and configured. Cook will automatically start embedded copies of the rest of its dependencies.

When deploying Cook in production, you’ll also want to provide a few other dependencies:

  • Zookeeper is used for leader election of the Cook schedulers, so that it can support high availability. Cook puts almost zero load on Zookeeper, so feel free to use a shared Zookeeper.

  • Datomic is used to store all data.

You’ll need to deploy Datomic as well. By default, you can use the Free edition of Datomic, which runs on a single machine and writes all its data locally. Be sure to make backups!

If you’re running Cook in an enviroment with higher availability requirements, you can use Datomic Pro or Pro Starter. To use Datomic Pro instead of Datomic Free, - Install Datomic Pro jars locally on the build host using mvn install. Datomic pro distribution provides a helper script called bin/maven-install that installs jar in local maven repository. (See http://aan.io/datomic-pro-and-leiningen/ for some helpful tips.) - Change the dependency in project.clj from com.datomic/datomic-free to com.datomic/datomic-pro. - Run lein uberjar

We recommend not changing the version of Datomic, although moving to a version shouldn’t cause any issues.

If you’d like email notifications on critical exceptions, make sure that JavaMail will work on the server you’ve set up.

© Two Sigma Open Source, LLC