2024-12-09T18:28:22.208Z

<Casey Bodley> i've still been seeing these jsonnet-bundler-build failures, so opened <https://tracker.ceph.com/issues/69164> to track them

2024-12-09T18:42:26.971Z

<Tyler Stachecki> Mostly curious... I noticed `pg_upmap` is defined as `mempool::osdmap::map` and not `::unordered_map`. Any reason for it to be ordered?

2024-12-09T18:49:59.899Z

<gregsfortytwo> unordered_map is usually implemented as a hash map and thus a lot larger than a regular map is (I want to say they tend to take up 4K each?). So we don’t generally use it unless we expect large membership and can afford the memory use at whatever rate it’s allocated (since we keep a lot of OSDmaps at once, a 4k object per map can add up to real memory)

Not sure if that’s why here, but that’s my first thought

2024-12-09T18:51:40.473Z

<Tyler Stachecki> ahhh... yeah I knew about the hashmap and was thinking the lookup time for what you'd be doing with upmaps are generally beneficial for faster `std::find` ops, but wasn't thinking about the size.

2024-12-09T23:22:01.309Z

<Mark Nelson (nhm)> If the number of elements is small, wouldn't a sorted vector make more sense?