Chris Burr

In that case I think the lack of package would be clear enough to point the user to use an older date. Trying to be clever would reduce the debuggability...

Would this be nicer as a generic `lb-conda` application?

I agree: https://gitlab.cern.ch/lhcb-dirac/LHCbDIRAC/-/issues/31

From a bit more debugging, it looks like it's reading way more data than it needs to, see the `size:` parts of below: ```bash $ XRD_LOGLEVEL=Debug python /tmp/uproot-perf.py $pfn default...

I've done some very crude testing with a high latency connection (Australia CERN) and it seems like the new fsspec implementation does well in comparision to the old XRootD source...

`fsspec-xrootd` is converting the requested ranges into bigger blocks: https://github.com/CoffeaTeam/fsspec-xrootd/blob/551692a472fcff2cdba3799ebc75670f0199ddf2/src/fsspec_xrootd/xrootd.py#L709-L742 If I force `self.blocksize = 0` 9.6 seconds becomes 1.6 seconds. This is still twice as long as XRootDSource but...

The rest of the difference is coming from the `fsspec` source opening the file twice: ``` [2024-03-05 23:00:45.735325 +0100][Debug ][ExDbgMsg ] [ccsrm.ihep.ac.cn:1094] MsgHandler created: 0x15e5ffa0 (message: kXR_open (file: /dpm/ihep.ac.cn/home/lhcb/LHCb/Collision18/LEPTONIC.MDST/00092248/0000/00092248_00002347_1.leptonic.mdst, mode:...

> That's connected to the reason the file is opened twice, right? Does a File need to access the beginning of a file before it can extract a fixed byte...

Thanks for the analysis @nsmith-! 🍰 All of your conclusions match with what I found, with the bonus of numbers to back them up. For the caching open files, I...

Can this be merged for the next release? I nearly made the same PR as a result of https://github.com/conda-forge/staged-recipes/pull/12028/