canvascloudai commented on PR #3855: URL: https://github.com/apache/celeborn/pull/3855#issuecomment-5989568851
> > Curious whether you've exercised a per-zone burst end to end: with the memory trigger at 0.70 and pause-receive at 0.85, do new workers become usable soon enough to relieve pressure on the existing workers, or can those workers still hit pause-receive while the added capacity is coming online? > > A small Cloud World Model sketch motivated the question, but it used assumed capacity and CPU-based scaling rather than Celeborn's disk/memory triggers or shuffle behavior, so it is not validation of these thresholds. > > it works flawlessly in my case , 3 weeks in production already Thanks for sharing. Good to hear it’s been running smoothly for three weeks in production. In our Cloud World Model sketch, retries amplified the warm-up gap when demand exceeded the starting fleet’s assumed capacity. That depended on our capacity and startup assumptions, so it doesn’t establish a Celeborn threshold. Appreciate the real-world context! -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
