Hi, Conferences * SCaLe got pushed back from early march to late july. Moving to LA * SREcon: working with vendor. They seem to be proceeding * Cephalocon: Omicon is peaking now looking much better shape by then. Let's make more careful decistions if we know who's registering. We need minimum number of registrations. Speakers got notified already. Q for speakers: in-presence or virtual? * New subvariant of omicron coming. Any impact? Matrix bridge: continued brokenness * how many users are using matrix? * problems with irc: super old protocol with no bouncer. People expect bouncers by now. * what do we want to list on the Ceph website? * AI: let's add the matrix bridge to the ceph.io website and see how many people will use it. Gibba * We encountered an encode issue. Fixed by switching to tip of quincy. * Upgrade went well. * AI: Need something to test on. Cache tiering * Has been deprecated for many releases. Problem: many people continue to use it. * David: RBD workloads still pretty bad without cache-tiering. Especially with EC. Even with DB on nvme OSDs. Using a huge RBD page cache also not a good solution. Alternative solutions (e.g. hardware raid) are all have their own problems. * Dan: If we hesitate to formally deprecate, can we at least document the status clearly at https://docs.ceph.com/en/pacific/rados/operations/cache-tiering/ ; Otherwise companies / labs could make long term strategy and not be aware it'll be dropped at some point * Ilya: Cache tiering V1 was completely broken. Then came V2, turned out that it also had a few major issues. Ilya: we need to have a strong statement. * People need to do very detailed investigation about every feature they want to implement. It's way too easy to shoot themselves in the food. We really have to write strong statements. * Mark: DMcache promising, but also not perfect. dm-cache has some issues with very large drives and minimum chunk sizes / write-amplification afaik, but it gets fairly complicated * Alternative to investigate is opencas, but in general block-level caching has *lots* of intricacies. * Dan: "cas" has the ability to tag IO. * AI: find out extend of the breakage. * AI: We need a good alternative. Quincy RC * RBD job gets stuck related to cache tiering * Neha: Let's make cache tiering failues not block the RC * Greg: Let's decide on cache tiering before. * February 9th? Bye, Sebastian _______________________________________________ Dev mailing list -- dev@xxxxxxx To unsubscribe send an email to dev-leave@xxxxxxx
Attachment:
OpenPGP_signature
Description: OpenPGP digital signature
_______________________________________________ Dev mailing list -- dev@xxxxxxx To unsubscribe send an email to dev-leave@xxxxxxx
