Exec timeout after 5 minutes

[oradwell]

Describe the bug
Exec times out after 5 minutes. Possibly caused by network proxies.

Is it worth implementing a ping-pong or a keepalive in kubernetes JavaScript client?

Client Version
1.3.0

Server Version
1.33.10

To Reproduce
Steps to reproduce the behavior:

Expected behavior
A clear and concise description of what you expected to happen.

Example Code
Code snippet for what you are doing

Environment (please complete the following information):

• OS: Linux
• Node.js version 20
• Cloud runtime GKE

Additional context

Like #478

[oradwell]

This is implemented in kubectl via websocket go package and in python websockets module.

In JavaScript ws package, it's recommended to detect broken connections.

Some of the projects affected by this issue:

• fix(k8s): work around a rare websocket connection issue & warn garden-io/garden#5908
• Adding websocket heartbeat actions/runner-container-hooks#333

[brendandburns]

I'm ok with implementing optional ping/pong.

That said, having an exec that hangs for 5 minutes without sending data (or indeed an exec lasting 5 minutes in general) isn't really what exec was designed for in k8s, there's much better options for long-running actions inside of containers.

[oradwell]

thanks for your response. what would be better than exec for this use case? I can try to suggest it to the maintainers.

I am able to reproduce this by running a sleep in a github actions workflow step. any command that hangs for more than 5 mins gets dropped by proxies with this exec implementation. Workarounds are, implementing ping pong on the service or a background command in workflow printing dummy output. I also have a ticket with the cloud provider trying to understand if the proxy timeout can be adjusted.

[brendandburns]

@oradwell apologies for the delay in responding. In general if you want to run a long-running action on a Pod, you're better off building some simple infrastructure in the Pod that knows how to manage the long running process.

The easiest thing to do is use the screen command line tool, but there's more complicated options like building your own simple REST API or other things.

The truth is that expecting to be able to hold a connection open on the internet for long periods of time, especially with no data traversing, is fairly brittle, you need the ability to reconnect, which is something exec doesn't provide for.