Attendees: Philippe, Jim, Pratyush Das, Guilherme, Oksana, Jakob.
Oksana: Catching up on older task and prepare for next week Sprint. I want to achieve the following:
* RTuple and (ZSTD) Compression discussion and plans.
* ZTD, compression dictionary and TTree.
* Discuss Jim with and etc.
Jakob: Exciting week with many merges in RTuple. Most visible added parallel support for RDataFrame. Added support for more types including std::variant. Enough functionality for CMS NanoAOD, so ready for benchmarking.
2nd implementation of page storage class ( “RawFile” ).
Philippe: RawFile will eventually need a vector interface.
[Discussion on granularity of reads and write … one idea: cache whole cluster and when storing make sure the branch’s data is stored consecutively … no matter what readv is needed for the case of sparse reads].
Philippe: Why was the name was remove from Column?
Jakob: I found that keeping a column name because it is already stored in the meta data.
Jim: Finally working C++ implementation of awkward-array and trying to understand how it relates to RTuple.
Philippe: Made progress on schema evolution of split sub-objects. (need to make ‘whole’ StreamerInfo as being cached and maybe repeated).
Philippe: ART is reporting yet another pch-file-is-out-of-date problem, what do we do?
Axel: The long term plan (needed by Windows by the SDK changes all files) is to ‘disable’ carefully the ‘stat-ing’ of the file (we have a copy inside the pch) but in the short term we could simply make sure the error message is convenient/clear enough and added the offending file to our list of file we ‘cache’ locally.
Philippe: There will be an I/O meeting next to provide for a debrief of next week’s I/O sprint.