Spark UI Proxy
If you are running a Spark Standalone cluster behind a firewall (let’s say it is running on Amazon AWS and your EC2 instances don’t have public IP addresses), you might have issues accessing the UI of your cluster, especially because each worker has its own UI, making it difficult if not impossible to reroute all the ports using only SSH tunnels.
Firewall
|
| ------------------------------
| | Spark Master |
| | e.g. http://10.0.0.1:8080 |
| ------------------------------
|
---------------------- | ------------------------------
| Your computer | ----->X | Spark Worker |
| e.g. 192.168.0.10 | | | e.g. http://10.0.0.2:8080 |
---------------------- | ------------------------------
|
| ------------------------------
| | Spark Worker |
| | e.g. http://10.0.0.3:8080 |
| ------------------------------
|
I created Spark UI Proxy to work around this issue. The Python script creates a lightweight HTTP server that proxies all the requests to your Spark Master and Spark Workers. All you have to do is create a single SSH tunnel to this proxy, and the proxy will forward all the requests for you. All the links between the nodes will be functional.
Firewall
|
| ------------------------------
| | Spark Master |
| -> | e.g. http://10.0.0.1:8080 |
| / ------------------------------
| /
---------------------- tunnel ------------------------ / ------------------------------
| Your computer | -----------> | spark-ui-proxy | ----> | Spark Worker |
| e.g. 192.168.0.10 | :9999 :9999| http://10.0.0.1:9999 | \ | e.g. http://10.0.0.2:8080 |
---------------------- | ------------------------ \ ------------------------------
| \
| \ ------------------------------
| ->| Spark Worker |
| | e.g. http://10.0.0.3:8080 |
| ------------------------------
|
How to use it
Let’s say the Spark Master has its UI running on localhost:8080 (localhost refers to the Spark Master node), and we want to access that UI on localhost:9999 (localhost here refers to your computer).
Start by creating an SSH tunnel from your computer to the Spark Master (but it could be to any of the nodes):
$ ssh -L 9999:localhost:9999 <public IP/name of the node>
On this node, run the Python proxy:
$ python spark-ui-proxy.py localhost:8080 9999
You can stop the proxy at any time by hitting Ctrl+C.
Alternatively, you may run the proxy in background:
$ nohup python spark-ui-proxy.py localhost:8080 9999 &
You can also run it with docker:
$ docker build -t spark-ui-proxy .
$ docker run -d --net host spark-ui-proxy localhost:8080 9999
Now, on your computer, open http://localhost:9999 and you should see the UI of your Spark cluster!