Jim Garlick

Results 461 comments of Jim Garlick

OK, thanks for that context. One solution might be to embed the Flux subprocess server in the flux shell. This was prototyped in #2298. I think there has since been...

They're coming straight toward our proximity! https://www.youtube.com/watch?v=Km_1NMUHjfA

A few quick ideas **flux instance**: a group of `flux-broker` processes which are launched via any parallel launch utility that supports PMI (process manager interface) **kvs**: a component of a...

IMHO throw everything in the glossary, acronym or backronym or no.

Hmm, getting an ASAN failure here: ``` 2024-03-15T15:21:32.2114399Z Making check in common 2024-03-15T15:21:32.2245352Z AddressSanitizer:DEADLYSIGNAL 2024-03-15T15:21:32.2246109Z ================================================================= 2024-03-15T15:21:32.2249768Z ==37186==ERROR: AddressSanitizer: SEGV on unknown address 0x63abbfc31750 (pc 0x7ab1a6bb7cb7 bp 0x000000000000 sp 0x7ffe4ec09ea0...

Looking at the first sched test failure, the test fakes resources consisting of 4 nodes, each with 44 cores and 4 gpus. In the first failure, it allocates two nodes...

FWIW I was able to fix the sched test failures by returning the allocated resource set from the job-manager directly in `resource.status` instead of marking those resources allocated the full...

Ah: the allocated set from `job-manager.resource-status` does not include properties so tests involving queues are failing in `t2350-resource-list.t` and `t2801-top-cmd.t`. I'm not sure if this is because sched-simple isn't returning...

> I wonder how much work it would be to properly support gpus in librlist? Perhaps that would then allow sched-simple to allocate > them (currently it cannot). Yeah that...

Hmm, actually `rlist_append()` seems to pass properties through OK, but it doesn't look like sched-simple is returning queue properties in allocated Rs. Is it possible that was by design?