Zongheng Yang
Zongheng Yang
I'd say this is a must-have, not "not that important". I was just debugging a HDFS issue and would like to see the free space on all of the nodes....
+1 to what OP said. Can confirm all those symptoms have hit me.
@ftynse Thanks. I'm using the conda-installed version of TC, commit `git_version: "8e112e9dccda62c30ef29208a827e783b9a7f156"` where `--logtosdterr` is not available. Is there a workaround? Fundamentally, is there a way to figure out the...
@ftynse @nicolasvasilache I will give building from source a try. Regarding whether or not correct launch bounds should be stored on disk after auto-tuning: it seems obvious it should be...
+1. This will be tremendously more user-friendly. When a single job becomes the straggler orders-of-magnitude slower (than any of best, median, or worst), it does not make sense to continue...
Let’s ship it!
Let’s ship it! On Wed, Aug 31, 2022 at 14:31 Woosuk Kwon ***@***.***> wrote: > @concretevitamin If you don't have > any more concern about this PR, I'll merge it....
Nice catch @WoosukKwon! Wdyt about leaving the unsupported VM types out of the catalog for now? This way users can get a nicer "VM type not found/supported" error. In the...
> @concretevitamin Sounds good. Then I think we can keep E2 and only remove t2a and f1-micro VMs. WDYT? Sounds good.
Seems like this can affect non-spot jobs as well? The 2nd option, placing the job back to the queue with some backoff, seems better because the user doesn't need to...