dashboard: allow 60 seconds for the dashboard to start

[morotti]

Hello,

I'd like to upstream this fix that was done internally at my company.

When running on older machines with less impressive CPU, and when under a bit of load. We've seen the dashboard failing to start in the expected 200 iterations (sleep(0.1) for 20 seconds).
We had to patch ray to allow 60 seconds for startup.

Regards.

[gemini-code-assist]

Code Review

This pull request increases the startup timeout for the Ray dashboard from 20 to 60 seconds. This is a sensible adjustment to accommodate slower machines or systems under heavy load, preventing premature startup failures. My feedback focuses on improving code readability by making the timeout value more explicit.

[edoakes]

@morotti thanks for the contribution, this looks reasonable to me. While you're at it, do you mind making this controllable via an environment variable? You can add it here:

ray/python/ray/_private/ray_constants.py

Line 4 in 88655df

import logging

I'd call it RAY_DASHBOARD_STARTUP_TIMEOUT_S with a default value of 60.

[morotti]

While you're at it, do you mind making this controllable via an environment variable? You can add it here:

@edoakes May I ask for you to do the patch if you want something more specific? I'm really not familiar with your codebase to look into how to read environment variable and give you tests and etc.. I'm sure you can do it in 10 minutes.

I really don't mind about not getting credit for the commit/PR. I just need the library fixed. :)

[edoakes]

While you're at it, do you mind making this controllable via an environment variable? You can add it here:

@edoakes May I ask for you to do the patch if you want something more specific? I'm really not familiar with your codebase to look into how to read environment variable and give you tests and etc.. I'm sure you can do it in 10 minutes.

I really don't mind about not getting credit for the commit/PR. I just need the library fixed. :)

yes, done