I believe that in the early stages of a major version, some breaking changes are unavoidable. We can't ensure 100% prevention.
However, once an issue arises, we need to fix and resolve it promptly, rather than maintaining the breaking change. In other words, how many users are currently using version 4.0 in production? We should aim to deliver future 4.0.x versions with high stability and compatibility with version 3.0. We can mention in the documentation or community that versions 4.0.0 to 4.0.2 have compatibility issues, which will be fixed in 4.0.3 and future versions. Thanks, Baodi Shi Enrico Olivelli <eolive...@gmail.com> 于2025年1月13日周一 22:51写道： > > Given that Pulsar 4.0 has been out for a while and it is considered > production ready it is better to keep compatibility with Pulsar 4 (set the > flag to 'true') and document this flag in the release notes or any upgrade > guide from 3 to 4. > > So: > - keep the current behavior in Pulsar 4.0.2 > - document that upgrading from 3.x to 4.x requires to set that flag in > order to support rollbacks > > I know that this is not ideal, but current users of Pulsar 4 (that maybe > started on 4) will see bad surprises when upgrading to a "point release". > You can expect big surprises when you switch to a major release (2 -> 3, 3 > -> 4) but you really don't expect any surprises when you upgrade/rolback > from x.y.z to x.y.k > > I am sorry that we did not catch this behavior change, we should add some > "rollback testing" to our CI or at least to some checklist we run while > voting a new "major release". > > > > Enrico > > > Il giorno lun 13 gen 2025 alle ore 11:42 Lari Hotari <lhot...@apache.org> > ha scritto: > > > I hope to get broader opinions from the community before any such changes > > are made. > > Especially from Rajan and Enrico. Rajan is the author of > > https://github.com/apache/pulsar/pull/9292 and Enrico has approved the PR. > > There was a dev mailing list discussion regarding PIP-381 where PR 9292 > > was discussed. I found these messages by searching: > > https://lists.apache.org/thread/dvhwo1jp5opyqkfspl3lynm6w4y9wg5g. > > > > -Lari > > > > On 2025/01/13 09:54:14 Baodi Shi wrote: > > > hi, > > > > > > We should ensure forward and backward compatibility between 3.0 and > > > 4.0. We can enable > > > configuration(managedLedgerPersistIndividualAckAsLongArray) in 5.0. > > > > > > Therefore, I believe the default value should be false. > > > > > > We shouldn't be constrained by the already released 4.0.x versions. I > > > believe users prefer to choose the newer 4.0.x versions and have > > > seamless switching with 3.0.x. > > > > > > > > > Thanks, > > > Baodi Shi > > > > > > Yubiao Feng <yubiao.f...@streamnative.io.invalid> 于2025年1月13日周一 09:50写道： > > > > > > > > Hi Lari > > > > > > > > > We have already made releases 4.0.0 and 4.0.1 > > > > > with the changed defaults. Changing it now is too late. > > > > > > > > I think it is a fault and we should correct it, So let set the default > > > > value to `false`. > > > > > > > > > PR 9292 has been approved by the community and there > > > > > have been discussions in the past. > > > > > > > > I have never seen there is a discussion in the email list, and it was > > > > merged with a unique approvement. > > > > > > > > Since it already introduced a regression, and no others mentioned how > > to > > > > upgrade/downgrade gracefully so far, we should set the default value to > > > > `false`. > > > > > > > > On Thu, Jan 9, 2025 at 9:15 PM Lari Hotari <lhot...@apache.org> wrote: > > > > > > > > > On Thu, 9 Jan 2025 at 14:26, Yubiao Feng > > > > > <yubiao.f...@streamnative.io.invalid> wrote: > > > > > > > Isn't the typical way to handle regressions and bugs to fix them? > > > > > > > > > > > > It may have other potential issues. > > > > > > > > > > > > > I don't see a sensible argument for changing configuration > > defaults for > > > > > > 4.0.2 . > > > > > > > > > > > > Is there a PIP related to #9292, which defined the default > > behavior has > > > > > > been changed? > > > > > > > > > > We have already made releases 4.0.0 and 4.0.1 with the changed > > > > > defaults. Changing it now is too late. > > > > > PR 9292 has been approved by the community and there have been > > > > > discussions in the past. > > > > > I don't see why we would be postponing such an important improvement > > > > > to Pulsar 5.0 . Let's fix any possible regressions asap and adapt. > > > > > > > > > > -Lari > > > > > > > > > >
