I'd be happy to take a look at reviewing this as well as I would be excited to see Auto Repair in 5.0. Thank you for the patch, Paulo!
Andy On Fri, Jan 30, 2026 at 5:13 PM Paulo Motta <[email protected]> wrote: > I have submitted a patch porting AutoRepair to 5.0 on CASSANDRA-21138[1] > and tagged Jaydeep Chovatia for review. I would greatly appreciate other > sets of eyes, especially those involved with the original CEP-37 effort. > > The feature is disabled by default and no schema changes are made unless a > JVM flag is enabled to reduce upgrade risk to users who do not intend to > enable this feature. > > Please let us know if you have any questions or concerns about having this > merged in 5.0. > > [1] - https://issues.apache.org/jira/browse/CASSANDRA-21138 > > On Mon, Jan 12, 2026 at 8:34 PM Jaydeep Chovatia < > [email protected]> wrote: > >> Sure, I am happy to review it whenever it's ready, Paulo. Please let me >> know. >> >> Jaydeep >> >> On Mon, Jan 12, 2026 at 8:32 AM Paulo Motta <[email protected]> wrote: >> >>> I agree with Scott. I don't think we should backport this to 4.1 due to >>> the compatibility issues raised plus this branch has already been >>> stabilized for a while. >>> >>> I think backporting auto-repair to 5.0 would be more appropriate as it >>> would encourage users to adopt this version and get closer to trunk, rather >>> than encouraging users to stick to an older version. >>> >>> I decided to take a stab at backporting auto-repair + additional fixes >>> to 5.0 on this preliminary PR: >>> https://github.com/apache/cassandra/pull/4558 >>> >>> It's not ready for review yet since I need to gate the schema changes >>> under a feature flag, but I think I can get it ready by the end of week. >>> >>> If there's no opposition against shipping this in 5.0 maybe I can create >>> a JIRA and have Jaydeep review it ? >>> >>> On Mon, Jan 12, 2026 at 11:15 AM C. Scott Andreas <[email protected]> >>> wrote: >>> >>>> For the purpose of a quick straw poll, I’m not opposed to backporting >>>> to 5.x, but I don’t support backporting to 4.x-series releases for the >>>> compatibility and upgrade complexity reasons previously discussed. >>>> >>>> - Scott >>>> >>>> > On Jan 12, 2026, at 1:27 AM, Štefan Miklošovič < >>>> [email protected]> wrote: >>>> > >>>> > Hi everybody, >>>> > >>>> > I want to refresh this thread after the holidays. Is there an >>>> > agreement we reached? Is everybody on board with backporting to 4.1+? >>>> > How are we going to do this concretely? I guess Jaydeep would be >>>> > involved in the backporting as he just said. I honestly do not think >>>> > there is anybody else better suited to make it happen and your >>>> > willingness to do that is really appreciated. >>>> > >>>> > Regards >>>> > >>>> >> On Tue, Dec 23, 2025 at 5:38 AM Jaydeep Chovatia >>>> >> <[email protected]> wrote: >>>> >> >>>> >> FYI—regardless of the outcome, you can count on me to port CEP-37 in >>>> whatever form the community agrees on. As mentioned earlier, I’m already >>>> maintaining a private 4.1.6 fork ( >>>> https://github.com/apache/cassandra/pull/3367). >>>> >> Thank you! >>>> >> >>>> >> Jaydeep >>>> >> >>>> >>> On Mon, Dec 22, 2025 at 7:43 AM Micah Green <[email protected]> >>>> wrote: >>>> >>> >>>> >>> I'm really interested in this thread, but don't see an update on >>>> where we landed in terms of backporting and also the amount of work >>>> involved. I'm all for backporting to 5.x minimally! I'm planning our 2026 >>>> work and where this discussion goes will really help me optimally plan, >>>> which is why I'm asking. >>>> >>> Thanks! >>>> >>> >>>> >>> On Sun, Dec 7, 2025 at 4:44 PM Ekaterina Dimitrova < >>>> [email protected]> wrote: >>>> >>>> >>>> >>>> Seems like the 4.1 branch would still require some work to cover >>>> everything raised on this thread? Have anyone evaluated how much work that >>>> can be? >>>> >>>> >>>> >>>> I agree porting to 4.1, but not 4.0 is kind of weird. Then >>>> probably we better have it only in 5.0? >>>> >>>> >>>> >>>> Do people think it makes sense to create some kind of user survey >>>> around this work, too? Posted in @user >>>> >>>> >>>> >>>> On Fri, 5 Dec 2025 at 9:00, Josh McKenzie <[email protected]> >>>> wrote: >>>> >>>>> >>>> >>>>> Otherwise it feels weird backporting to 4.1 but not 4.0, and >>>> backporting to both would increase the risk and maintenance burden. >>>> >>>>> >>>> >>>>> It would but by how much? >>>> >>>>> >>>> >>>>> 2 things jump out to me re: risk and maintenance: >>>> >>>>> >>>> >>>>> Risk: We kind of need to tackle the "version straddle w/schema >>>> table diffs is currently Bad and makes rollbacks manual and brittle" >>>> broadly; this feature is just one more example of that though it's a little >>>> exacerbated by discussing doing something like this in a patch release. The >>>> ergonomics of the "one-way-door without a human manually deleting columns" >>>> part holds true cross-MAJOR too. "Progress" here seems like it's either we >>>> handle this on a case-by-case basis w/flags to remove those schema entries >>>> on rollback (kinda ew), or more durably with an elegant solution in the >>>> long term, i.e. capabilities framework, though that doesn't answer the "we >>>> explode when schemas don't match" bit. >>>> >>>>> >>>> >>>>> Maintenance: maintaining this across 4 branches is clearly more >>>> toil than across 2. >>>> >>>>> >>>> >>>>> I'm personally kind of keen for us to tackle that Risk bit; I'd >>>> like all of us to be able to more freely consider making changes to schema >>>> tables w/out the complexity burden we have right now and the operator toil >>>> and risk that comes along with it. >>>> >>>>> >>>> >>>>> The maintenance toil bit - I have less opinions on. Kind of >>>> depends on how many people are on 4.0/4.1 right now that we'd expect to be >>>> on 4.1 for another year until 7.0 hits and whether we think they'd benefit >>>> from the feature (and contribute to bettering it) during that year I guess. >>>> >>>>> >>>> >>>>> On Thu, Dec 4, 2025, at 5:57 PM, Paulo Motta wrote: >>>> >>>>> >>>> >>>>> Otherwise it feels weird backporting to 4.1 but not 4.0, and >>>> backporting to both would increase the risk and maintenance burden. >>>> >>>>> >>>> >>>>> >>>> >>>
