Spark-jobserver is available as a Docker container! This might be the easiest way to get started and deploy.
NOTE: In all the docker commands below, you need to append the tag, such as
To get started:
docker run -d -p 8090:8090 velvia/spark-jobserver
This will start job server on port 8090 in a container, with H2 database and Mesos support, and expose that port to the host on which you run the container.
If you would like to debug job server using JMX / VisualVM etc., then also expose port 9999.
By default, the container has an embedded Spark distro and runs using Spark local mode (
To change the spark master the container runs against, set SPARK_MASTER when you start the container:
docker run -d -p 8090:8090 -e SPARK_MASTER=mesos://zk://mesos.master:5050 velvia/spark-jobserver
You can easily change the amount of memory job server uses with
JOBSERVER_MEMORY, or replace the entire config job server uses at startup with
The standard way to replace the config is to derive a custom Docker image from the job server one by overwriting the default config at
app/docker.conf. The Dockerfile would look like this:
from velvia/spark-jobserver add /path/to/my/jobserver.conf /app/docker.conf
Similarly, to change the logging configuration, inherit from this container and overwrite
Jars / Passing Arguments to the Start Script
spark-submit arguments can be passed to the tail of the
docker run command. A very common use of this is to add custom jars to your Spark job environment. For example, to add the Datastax Spark-Cassandra Connector to your job:
docker run -d -p 8090:8090 velvia/spark-jobserver --packages com.datastax.spark:spark-cassandra-connector_2.10:1.3.0-M1
Database, Persistence, Logs
Docker containers are usually stateless, but it wouldn't be very useful to have the jars and job config reset every time you had to kill and restart a container.
The job server docker image is configured to use H2 database by default and to write the database to a Docker volume at
/database, which will be persisted between container restarts, and can even be shared amongst multiple job server containers on the same host. Note that in order to persist them to new containers, you need to create a local directory, something like this:
docker run -d -p 8090:8090 -v /opt/job-server-db:/database velvia/spark-jobserver
See the Docker Volumes Guide for more info.
Another option is to configure job server to persist metadata in PostGres, MySQL, or similar database. To do that, create a new config, pass it into the docker container as above using
JOBSERVER_CONFIG and the
/config volume, and point to your shared database, perhaps using
--link to a PostGres or MySQL container.
Logging goes to stdout, as per standard Docker conventions. Therefore:
docker logs -f <containerHash>to follow logs
docker logs --tail=100 <containerHash>to list the last 100 lines
- Use Docker logging drivers to redirect logs to syslog, SumoLogic, etc.
Could you please push an image with spark-2.0-preview and latest mesos (1.2.0)?
This dockerfile link is invalid, who can support a new one?Thanks
Thanks for uploading 0.6.2, we needed it just yesterday. It saved our day, because we need to be able to change the Java version of the jars.
When I add a argument --jars at the end of docker run, trigger a warning :
Warning: Local jar /spark/lib/JobServerDepentTest.jar does not exist, skipping.
and It result the job:
Could you help me resolve the problem?
+1 for Dockerfile / GitHub link
+1 for the Dockerfile source.
Could you please provide the Dockerfile for this project, or a link to it on Github?
This container is missing the Mesos library files and i cannot find the docker file to read how i can pass this to the start script. Can you provide either one?