web5design / chronos

Fault tolerant job scheduler for Mesos which handles dependencies and ISO8601 based schedules

Chronos

Chronos is Airbnb's replacement for cron. It is a distributed and fault-tolerant scheduler that runs on top of Apache Mesos. You can use it to orchestrate jobs. It supports custom Mesos executors as well as the default command executor. Thus by default, Chronos executes sh (on most systems bash) scripts. Chronos can be used to interact with systems such as Hadoop (incl. EMR), even if the Mesos slaves on which execution happens do not have Hadoop installed. Included wrapper scripts allow transfering files and executing them on a remote machine in the background and using asynchronous callbacks to notify Chronos of job completion or failures.

Chronos has a number of advantages over regular cron. It allows you to schedule your jobs using ISO8601 repeating interval notation, which enables more flexibility in job scheduling. Chronos also supports the definition of jobs triggered by the completion of other jobs. It supports arbitrarily long dependency chains.

Try out the interactive and personalized tutorial for Chronos.

For questions and discussions around Chronos, please use the Google Group "chronos-scheduler": Chronos Scheduler Group. Also join us on IRC in #mesos on freenode.

• Features
• Running Chronos
• Configuring Chronos
• License
• Contributors
• Video Introduction
• Chronos UI
• API
  • Leader
  • Listing Jobs
  • Deleting a Job
  • Deleting All Tasks for a Job
  • Manually Starting a Job
  • Adding a Scheduled Job
  • Adding a Dependent Job
  • Describing the Dependency Graph
  • Asynchronous Jobs
  • Obtaining Remote Executables
  • Job Configuration
  • Sample Job
• Job Management
• Debugging Chronos Jobs
• Notes
  • Environment Variables Mesos Looks For
• Reporting Bugs
• Appendix
  • Finding a Node to Talk To
  • Zookeeper
  • Install Chronos on Amazon Linux

If you get an error while compiling Mesos, please consult the FAQ.

Features

• Web UI
• 8601 Repeating Interval Notation
• Handles dependencies
• Job Stats (e.g. 50th, 75th, 95th and 99th percentile timing, failure/success)
• Fault Tolerance (Hot Master)
• Configurable Retries
• Multiple Workers (i.e. Mesos Slaves)

Running Chronos

We've included some example run scripts, but the basic syntax for launching chronos is:

java -cp chronos.jar --master zk://127.0.0.1:2181/mesos --zk_hosts 127.0.0.1:2181

Please note that you need to have both Mesos and Zookeeper running for this to work!

For more information on configuration options, please see configuring Chronos.

Example Run Scripts

• Example runit run script

  bin/run

• Example local run script

  bin/start-chronos.bash

Configuring Chronos

For information on configuring chronos, please see docs/CONFIG.md.

License

The use and distribution terms for this software are covered by the Apache 2.0 License (http://www.apache.org/licenses/LICENSE-2.0.html) which can be found in the file LICENSE at the root of this distribution. By using this software in any fashion, you are agreeing to be bound by the terms of this license. You must not remove this notice, or any other, from this software.

Contributors

• Florian Leibert (@flo)
• Andy Kramolisch (@andykram)
• Harry Shoff (@hshoff)

Video Introduction

• Replacing Cron & Building Scalable Data Pipelines: YouTube

Sample Architecture

Chronos UI

Chronos comes with a UI which can be used to add, delete, list, modify and run jobs. It can also show a graph of job dependencies. The screenshot should give you a good idea of what Chronos can do.

API

You can communicate with Chronos using a RESTful JSON API over HTTP. Chronos nodes usually listen on port 4400 for API requests. All examples in this section assume that you've found a running leader at chronos-node.airbnb.com:4400.

Leaders

When you have multiple Chronos nodes running, only one of them will be elected as the leader. The leader is the only node that responds to API requests, but if you attempt to talk to a non-leader your request will automatically be redirected to a leader.

Listing Jobs

• Endpoint: /scheduler/jobs
• Method: GET
• Example: curl -L -X GET chronos-node:4400/scheduler/jobs
• Response: JSON data

A job listing returns a JSON list containing all of the jobs. Each job is a JSON hash. Interesting fields in the hashes are:

• invocationCount: the number of times the job completed
• executor: auto-determined by Chronos, but will usually be "" for non-async jobs
• parents: for dependent jobs, a list of all other jobs that must run before this job will do so

If there is a parents field there will be no schedule field and vice-versa.

Deleting a Job

Get a job name from the job listing above. Then:

• Endpoint: /scheduler/job/jobName
• Method: DELETE
• Example: curl -L -X DELETE chronos-node:4400/scheduler/job/request_event_counter_hourly
• Response: HTTP 204

Deleting All Tasks for a Job

Deleting tasks for a job is useful if a job gets stuck. Get a job name from the job listing above. Then:

• Endpoint: /scheduler/task/kill/jobName
• Method: DELETE
• Example: curl -L -X DELETE chronos-node:4400/scheduler/task/kill/request_event_counter_hourly
• Response: HTTP 204

Manually Starting a Job

You can manually start a job by issuing an HTTP request.

• Endpoint: /scheduler/job
• Method: PUT
• Example: curl -L -X PUT chronos-node:4400/scheduler/job/request_event_counter_hourly
• Response: HTTP 204

Adding a Scheduled Job

The heart of job scheduling is a JSON POST request. The JSON hash you send to Chronos should contain the following fields:

• Name: the job name
• Command: the actual command that will be executed by Chronos
• Schedule: The scheduling for the job, in ISO8601 format. Consists of 3 parts separated by '/':
  • Number of times to repeat the job; put just 'R' to repeat forever
  • The start time of the job
  • The run interval
• Epsilon: If Chronos misses the scheduled run time for any reason, it will still run the job if the time is within this interval. Epsilon must be formatted like an ISO 8601 Duration.
• Owner: the email address of the person responsible for the job
• Async: whether the job runs in the background

Here is an example job hash:

{
  "schedule": "R10/2012-10-01T05:52:00Z/PT2S",
  "name": "SAMPLE_JOB1",
  "epsilon": "PT15M",
  "command": "echo 'FOO' >> /tmp/JOB1_OUT",
  "owner": "[email protected]",
  "async": false
}

Once you've generated the hash, send it to Chronos like so:

• Endpoint: /scheduler/iso8601

• Method: POST

• Example:

    curl -L -H 'Content-Type: application/json' -X POST -d '{json hash}' chronos-node:4400/scheduler/iso8601
  

• Response: HTTP 204

Adding a Dependent Job

A dependent job takes the same JSON format as a scheduled job. However, instead of the schedule field, it will accept a parents field. This should be a JSON list of all jobs which must run at least once before this job will run.

• Endpoint: /scheduler/dependency

• Method: POST

• Example:

    curl -L -X POST -H 'Content-Type: application/json' -d '{dependent hash}' chronos-node:4400/scheduler/iso8601
  

Here is a more elaborate example for a dependency job hash:

{
    "async": true,
    "command": "bash -x /srv/data-infra/jobs/hive_query.bash run_hive hostings-earnings-summary",
    "epsilon": "PT30M",
    "errorCount": 0,
    "lastError": "",
    "lastSuccess": "2013-03-15T13:02:14.243Z",
    "name": "hostings_earnings_summary",
    "owner": "[email protected]",
    "parents": [
        "db_export-airbed_hostings",
        "db_export-airbed_reservation2s"
    ],
    "retries": 2,
    "successCount": 100
}

Describing the Dependency Graph

Chronos allows to describe the dependency graph and has an endpoint to return this graph in form of a dotfile.

• Endpoint: /scheduler/graph/dot
• Method: GET
• Example:

    curl -L -X GET chronos-node:4400/scheduler/graph/dot

Asynchronous Jobs

If your job is long-running, you may want to run it asynchronously. In this case, you need to do two things:

1. When adding your job, ensure it is set as asynchronous.
2. Your, job, when complete, should reports its completion status to Chronos.

If you forget to do (2), your job will never run again because Chronos will think that it is still running. Reporting job completion to Chronos is done via another API call:

• Endpoint: /scheduler/task/task id
• Method: PUT
• Example:

    curl -L -X PUT -H "Content-Type: application/json" -d '{"statusCode":0}' chronos-node:4400/scheduler/task/my_job_run_555_882083xkj302

The task id is auto-generated by Chronos. It will be available in your job's environment as $mesos_task_id.

Note: You will probably need to url-encode the mesos task id in order to submit it as part of the URL.

Obtaining Remote Executables

When specifying the command field in your job hash, use the url-runner.bash (make sure it's deployed on all slaves). Alternatively, you can also use a url in the command field, if your mesos was compiled with cURL libraries.

Job Configuration

Field	Description	Default
name	Name of job.	-
command	Command to execute.	-
epsilon	If, for any reason, a job can't be started at the scheduled time, this is the window in which Chronos will attempt to run the job again	PT60S or --task_epsilon.
executor	Mesos executor. By default Chronos uses the Mesos command executor.	-
executorFlogs	Flags to pass to Mesos executor.	-
retries	Number of retries to attempt if a command returns a non-zero status	2
owner	Email addresses to send job failure notifications. Use comma-separated list for multiple addresses.	-
async	Execute using Async executor.	false
successCount	Number of successes since the job was last modified.	-
errorCount	Number of errors since the job was last modified.	-
lastSuccess	Date of last successful attempt.	-
lastError	Date of last failed attempt.	-
cpus	Amount of Mesos CPUs for this job.	0.1 or --mesos_task_cpus
mem	Amount of Mesos Memory for this job.	100 or --mesos_task_mem
disk	Amount of Mesos disk for this job.	100 or --mesos_task_cpus
disabled	If set to true, this job will not be run.	false
uris	An array of URIs which Mesos will download when the task is started.	-
schedule	ISO8601 repeating schedule for this job. If specified, parents must not be specified.	-
parents	An array of parent jobs for a dependent job. If specified, schedule must not be specified.	-

Sample Job

{
   "name":"camus_kafka2hdfs",
   "command":"/srv/data-infra/kafka/camus/kafka_hdfs_job.bash",
   "epsilon":"PT30M",
   "executor":"",
   "executorFlags":"",
   "retries":2,
   "owner":"[email protected]",
   "async":false,
   "successCount":190,
   "errorCount":3,
   "lastSuccess":"2014-03-08T16:57:17.507Z",
   "lastError":"2014-03-01T00:10:15.957Z",
   "cpus":1.0,
   "disk":10240,
   "mem":1024,
   "disabled":false,
   "uris":[
   ],
   "schedule":"R/2014-03-08T20:00:00.000Z/PT2H"
}

Job Management

For larger installations, the web UI may be insufficient for managing jobs. At Airbnb, there are well over 700 production Chronos jobs. Rather than using the web UI for making edits, we created a script called chronos-sync.rb which can be used to synchronize configuration from disk to Chronos. For example, you may have a Git repository that contains all of the Chronos job configurations, and then you could run an hourly Chronos job that checks out the repository and runs chronos-sync.rb.

You can initialize the configuration data by running:

$ bin/chronos-sync.rb -u http://chronos/ -p /path/to/jobs/config -c

After that, you can run the normal sync like this:

$ bin/chronos-sync.rb -u http://chronos/ -p /path/to/jobs/config -c

You can also forcefully update the configuration in Chronos from disk by passing the -f or --force parameter. In the example above, /path/to/jobs/config is the path where you would like the configuration data to live.

Note: chronos-sync.rb does not delete jobs. If you want to remove a job, you must manually remove it using the API or web UI.

Debugging Chronos Jobs

Chronos itself can be configured just like dropwizard-logging via the configuration file. If there's something going wrong with the framework itself look here for information. Individual jobs log with their task id on the mesos slaves. Look in the standard out log for your job name and the string "ready for launch", or else "job ct:" and your job name. The job is done when the line in the log says:

Task with id 'value: TASK_ID **FINISHED**

To find debug logs on the mesos slave, look in /tmp/mesos/slaves on the slave instance (unless you've specifically supplied a different log folder for mesos). For example:

/tmp/mesos/slaves/

In that dir, the current slave run is timestamped so look for the most recent. Under that is a list of frameworks; you're interested in the Chronos framework. For example:

/tmp/mesos/slaves/STAMP/frameworks/

Notes

The curl executor is even more powerful if the specified URLs are packaged and self-contained executables. This can be done for example via arx, which bundles code into an executable archive. Arx applications in turn contain shell commands and an archive (e.g. a jar file and a startup-script). It's easy to use and there are no libraries required to unpack and execute the archive.

Signed URLs can be used to publish arx files (e.g. on s3).

To start a new scheduler you have to give the JVM access to the native mesos library. You can do so by either setting the java.library.path to the build mesos library or create an environment variable MESOS_NATIVE_LIBRARY and set it to the mesoslib.dylib / mesoslib.so file

Environment Variables Mesos Looks For

• MESOS_NATIVE_LIBRARY: Absolute path to the native mesos library. This is usually /usr/local/lib/libmesos.so on Linux and /usr/local/lib/libmesos.dylib on OSX.
• MESOS_LAUNCHER_DIR: Absolute path to the src subdirectory of your mesos build, such that the shell executor can be found (e.g. If mesos was built in /Users/florian/airbnb_code/mesos/build then the value for this variable would be /Users/florian/airbnb_code/mesos/build/src).
• MESOS_KILLTREE: Absolute path to the location of the killtree.sh script. (e.g. /Users/florian/airbnb_code/mesos/src/scripts/killtree.sh)

If you're using the installer script this should be setup for you.

Reporting Bugs

To make all of our lives easier we ask that all bug reports include at least the following information:

The output of:

mvn -X clean package

and

java -version

If the error is in running tests, then please include the output of running all the tests.

# Mac/FreeBSD
tail +1 target/surefire-reports/*.txt
# GNU Coreutils
tail -n +1 target/surefire-reports/*.txt

If the error is in the installer, please include all the output from running it with debug enabled:

bash -x bin/installer.bash

If the bug is in building Mesos from scratch, please submit those bugs directly to mesos.

If the bug occurs while running Chronos, please include the following information:

• The command used to launch Chronos, for example:

    java -cp target/chronos.jar com.airbnb.scheduler.Main <args>
  

• The version of Mesos you are running.

• The output of

    java -version
  

Appendix

Finding a Node to Talk to

As we mentioned, Chronos is designed (not required) to run with multiple nodes of which one is elected master. If you use the cURL command line tool, you can use the -L flag and hit any Chronos node and you will get a 307 REDIRECT to the leader.

Zookeeper

Chronos registers itself with Zookeeper at the location /airbnb/service/chronos. This value can be changed via the configuration file.

Install Chronos on Amazon Linux

Follow these steps to install Chronos on Amazon Linux:

Install Dependencies

Debian Linux:

sudo apt-get install autoconf make gcc cpp patch python-dev git libtool default-jdk default-jdk-builddep default-jre gzip libghc-zlib-dev libcurl4-openssl-dev

Fedora Linux:

sudo yum install autoconf make gcc gcc-c++ patch python-devel git libtool java-1.7.0-openjdk-devel zlib-devel libcurl-devel openssl-devel cyrus-sasl-devel

Make sure you're using Java 7: sudo alternatives --config java

Build and Install Mesos

git clone https://github.com/apache/mesos.git
cd mesos/
git checkout 
export JAVA_HOME=/usr/lib/jvm/java-1.7.0-openjdk-1.7.0.45.x86_64/
./bootstrap
./configure --with-webui --with-included-zookeeper --disable-perftools --enable-frame-pointers
make
sudo make install

Build Chronos

Install Node first. On OSX, try brew install node.

Start up Zookeeper, Mesos master, and Mesos slave(s). Then try

export MESOS_NATIVE_LIBRARY=/usr/local/lib/libmesos.so
git clone https://github.com/airbnb/chronos.git
cd chronos
mvn package
java -cp target/chronos*.jar com.airbnb.scheduler.Main --master zk://localhost:5050/mesos --zk_hosts localhost:2181