Hue is unable to upload large files into a kerberized cluster when it is configured to talk to a HDFS HTTPFS server. Here is what's going on:
- Hue's webhdfs creates a session that preserves cookies.
- On Hue's first access to HTTPFS, it makes a request in which HTTPFS returns a 401. Hue then gets a kerberos ticket and passes it along to talk to HTTPFS, which also returns a cookie in the response.
- When Hue uploads a file, it first makes an empty request in order to get redirected to the actual place to perform the upload. It also passes along the cookies it received previously.
- Finally it uploads the file. This time however it creates a new session without the cookies. Normally we would just go through the standard kerberos authentication process (401 and a retry), but if the file is over a certain size, it outright rejects the upload and abruptly just closes the connection in order to protect against malicious users.
- Hue then displays an Error: Undefined popup because of the connection error not being properly rendered in the UI.
The fix is simple, just make sure to reuse the session when uploading the file.