sheepkiller/presto-marathon repository overview

⁠Deploying presto with docker/marathon/mesos

⁠Requirements

(working)

• mesos cluster
• marathon >= 0.7.5 (tested with marathon 0.7.5, should work with >= 0.70)
• some data accessible by presto
• a web/FTP server (anything accessible via curl)
• a private docker registry

⁠Software used

• confd (https://github.com/kelseyhightower/confd⁠)

⁠Howto

⁠Setup

docker pull sheepkiller/presto-marathon

Then, create a tarball containing your catalog properties files, located in a directory called "catalog", and put it in a place where curl can grab it. Here's an example for jmx

mkdir catalog
echo "connector.name=jmx" > catalog/jmx.properties
tar zcvf catalog.tar.gz catalog
scp catalog.tar.gz my.server:

⁠Let's play !

Please see examples/deploy_presto.json... As you can see, there's 4 mandatory environment variable to set:

1. CATALOGS_URL : where to find catalog.tar.gz
2. MARATHON_APPNAME : application name which must match args in your json
3. MARATHON_APPGROUP : applications group which must match group in id
4. MARATHON_URL : which must match your marathon master URL

You can also set JVM options via environment variables

• JAVA_PERM_SIZE
• JAVA_MAXPERM_SIZE
• JAVA_RCCS_SIZE
• JAVA_MAXHEAP_SIZE

Or some presto configuration

• PRESTO_TASK_MAXMEMORY

Or UUID (uuid is generated at runtime)

• PRESTO_UUID

If you need more tuning, modify confd files (presto-marathon-docker/docker/presto/confd)

⁠Start cluster

curl -i -H 'Content-Type: application/json' -d @deploy_presto.json http://my.marathon.master:8080/v2/groups

⁠Destroy cluster

 curl -X DELETE http://my.marathon.master/v2/groups/presto

⁠presto CLI

• via docker exec (docker >= 1.3.0) (you

$ docker exec -it <container id> /bin/bash
[container]# ./bin/presto  \
              --server \
              $(curl -sSfLk -m 10 -H \
                'Accept: text/plain' ${MARATHON_URL}/v2/tasks | \
                 egrep ^${MARATHON_APPGROUP}_meta_coordinator | \
                 awk 'NF == 3 {printf("http://%s"), $NF}')

• from anywhere: configure and use

sh connect.sh

Has been tested on bridged networks.

⁠Limitations

• multiple concurrent execution should work if you change group name (didn't test)

⁠ToDo's

• Better discovery "code"
• Better catalog handling
• Maybe split /${MARATHON_APPGROUP}/meta in 2: discovery & coordinator (or merge coordinator and discovery)
• Allow integration with etcd/consul (supported by confd)
• HA for marathon master
• handler more presto configuration options