Title: Running Spark On Mesos
Speaker: Mr. Chris Sidiropoulos (https://linkedin.com/in/chris-sidiropoulos-2a6b156a//)
Date: Tuesday, November 27, 2018
Event: https://meetup.com/Athens-Big-Data/events/256098657/
3. About
● Advanced Security Analytics and Response Orchestration
● Early compromise detection.
● Capture and analyze traffic logs.
4. Alternatives
● Standalone
○ Easy to deploy (scripts are bundled with spark distribution).
○ Can easily run on localhost for development.
○ Master-Worker setup.
○ HA supported utilizing Zookeeper.
○ Web UI for monitoring cluster and job statistics.
● Hadoop YARN
○ Harder to bring up.
○ Combination of the ResourceManager, NodeManager, Application Master & Container.
○ HA Supported utilizing Zookeeper.
○ ResourceManager/NodeManager UI.
● Kubernetes
○ Still experimental (v2.4.0).
○ Easy to get up and running if you are already familiar with k8s.
● Nomad
○ Good if you are into experimenting with hashicorp products.
5. A few things about Mesos
● Using the same principles as the Linux kernel, only at a different level of abstraction.
● Dynamic resource sharing and isolation (CPU, RAM, …).
● Turn your data center into one very large computer (global resource manager).
● Dominant Resource Fairness.
● Scales on 10,000s of nodes.
● Packages and commercial support through Mesosphere
● Even YARN can run on mesos(myriad).
● Three main components
○ Mesos Master
○ Mesos Agent
○ Mesos Framework
8. DC/OS
● Distributed operating system based on the Apache Mesos distributed systems kernel.
● A Cluster Manager.
● A Container Platform.
● An Operating System.
● Great documentation.
● Easy to spin up.
● A great catalog with packages (Universe).
13. Spark on Mesos
● Client Mode
○ A Spark Mesos framework is launched directly on the client machine and waits for the driver
output.
● Cluster mode
○ The driver is launched in the cluster and the client can find the results of the driver from the Mesos
Web UI.
● Mesos run modes:
○ Fine-grained mode (deprecated)
○ Coarse-grained mode, each Spark executor is represented by a single Mesos task. As a result,
executors have a constant size throughout their lifetime.
17. Viewing the logs
● dcos spark log
dcos spark log driver-20181126153522-0001 --file="stderr" --lines_count=4
18/11/26 16:05:36 INFO ShutdownHookManager: Deleting directory
/tmp/spark-19fad8b1-b162-44c4-a6ad-3cf3d9f3f004
18/11/26 16:05:36 INFO ShutdownHookManager: Deleting directory
/tmp/spark-19fad8b1-b162-44c4-a6ad-3cf3d9f3f004/pyspark-b93f92bc-bf9e-40b7-8ccd-65
8d18c7eade
I1126 16:05:37.522994 7645 executor.cpp:675] Container exited with status 137
W1126 16:05:37.522994 7644 logging.cpp:93] RAW: Received signal SIGTERM from process
2589 of user 0; exiting