There is a thread about cherry-picking the changes: https://lists.apache.org/thread/g1jq0vkq0wt28mdnbg63bpw8fpx9ml42
Similar changes have also been made in the past to maintenance branches, so this is not anything exceptional to bring up for discussion and then cherry-pick to maintenance branches. > I propose to revert that PR. At least, we should add a broker level > config or namespace level policy to control the behavior. But still, > we definitely need a PIP for such changes. We need a further > discussion for the compatibility issue rather than merging it and > cherry-picking it into old branches. So I suggest reverting it first. Since there's a reason to have PR 24118 & PR 24154 changes in maintenance branches, I'd suggest adding a way to configure this behavior. Adding a PIP would be a good way to document this change. -Lari On 2025/04/22 12:50:13 Yunze Xu wrote: > Hi all, > > https://github.com/apache/pulsar/pull/24118 introduces breaking > changes on the downstream applications without a PIP. > > For example, our internal Kafka protocol implementations heavily rely > on the behavior that partitions can be produced or consumed even if > the partition metadata is deleted. Many tests failed due to this > change. > > I didn't look into the technical details at the moment. But it should > be noted that sometimes it's valid to read a partition without > assuming the topic's partition metadata exists. > > I propose to revert that PR. At least, we should add a broker level > config or namespace level policy to control the behavior. But still, > we definitely need a PIP for such changes. We need a further > discussion for the compatibility issue rather than merging it and > cherry-picking it into old branches. So I suggest reverting it first. > > Please note that this PR also modifies many existing tests in Pulsar. > It's a clear signal that breaking changes were made. We should be > careful of such PRs. > > Thanks, > Yunze >
