Nicholas Smith

This PR is a bit stale, I guess. I think it would be a nice feature, users could for example say a replica is suspicious if a job fails to...

This looks to solve a perennial issue for me. Is there any reason not to take this less-than-perfect solution for now and work on the nanny-restarts-worker solution in the longer...

If memory usage is high enough that we hit the pause fraction watermark _and_ there are no keys to spill to disk, then I think what we could do is:...

What a coincidence, I've also been looking at this in the last few days. Regarding the reading more bytes than necessary, I think there the best course of action is...

So I've been running a skim benchmark where I access the first 50k events out of a few hundred files hosted throughout the US, comparing the new and old xrootd...

It was at first my understanding that `block_size` is supposed to set what granularity we fetch data at any position from the file, e.g. with it set to e.g. `1024`...

Ok I implemented a file handle cache in https://github.com/CoffeaTeam/fsspec-xrootd/pull/54 and now the sites that previously had a lot of trouble with opening the file often are doing better.  On...

Here's a summary of what I've attempted:  The legend entries correspond to progressively more patches to improve this situation: - `+fsspec`: introducing the FSSpec source, as of v5.3.0 -...

All the updates are merged. I'll close this for now.

The removal of the stateful handle is motivated by https://github.com/scikit-hep/uproot5/issues/1157#issuecomment-1981832282