The title of the KIP seems a little odd, because if I understand correctly, the main change you want to make is to bundle multiple logging backends with Kafka and make them selectable via a system property, and upgrading sfl4j is a means to achieve that, not the goal itself?
Den fre. 21. mar. 2025 kl. 12.18 skrev Chia-Ping Tsai <chia7...@gmail.com>: > hi Teng > > > The KIP will document that log4j2 is the only officially supported > server-side logging framework, and we will expose its configuration file > to > users. > > on the server-side, we should keep current scope - compileOnly and > releaseOnly - Otherwise, downstream projects could encounter dependency > conflicts [0] > > > I will revise the motivation section of the KIP to emphasize that the > > Please include the following description. > > "The rationale for this KIP is that upgrading SLF4J necessitates > corresponding provider upgrades, which constitutes a breaking change." > > Also, we must upgrade the Log4j2 dependency based on SLF4J 2 (i.e., > log4j-slf4j-impl to log4j-slf4j2-impl) in 5.0 if we upgrade to slf4j2 > > [0] https://github.com/apache/kafka/pull/17373#issuecomment-2577813317 > > Best, > > Chia-Ping > > > TengYao Chi <kiting...@gmail.com> 於 2025年3月21日 週五 下午6:52寫道： > > > Hello everyone, > > > > Sorry for the late reply. > > Based on the previous discussion, I would like to conclude with the > > following points: > > > > 1. The upgrade to slf4j2 should be postponed to Kafka 5.0 due to > > compatibility issues. > > 2. I will revise the motivation section of the KIP to emphasize that > the > > key benefit is allowing users to select logging backends through > > configuration rather than modifying JAR files. > > 3. After the slf4j upgrade, users will be able to use the > > `-Dslf4j.provider` system property to configure their preferred > logging > > backend. > > 4. The KIP will document that log4j2 is the only officially supported > > server-side logging framework, and we will expose its configuration > > file to > > users. > > To avoid breaking downstream compatibility, we will not bind the > client > > side to any specific logging framework. Users will need to manage > their > > own > > logging libraries, but they can utilize the `-Dslf4j.provider` > property > > once slf4j is upgraded. > > 5. We have rejected alternatives that involve warnings and classpath > > ordering as they do not provide a solid solution to compatibility > > issues. > > > > Does this summary make sense? > > > > Best, > > TengYao > > > > Ismael Juma <m...@ismaeljuma.com> 於 2025年3月21日 週五 上午9:32寫道： > > > > > A solution that involves a warning and classpath ordering doesn't meet > > the > > > bar for me. Good clarification though. > > > > > > Ismael > > > > > > On Thu, Mar 20, 2025 at 8:37 AM Farid Zakaria > > > <fzaka...@confluent.io.invalid> > > > wrote: > > > > > > > AFAIR SLF4J you don't have to remove the other backends; merely make > > > > sure yours is first on the CLASSPATH list :P > > > > (SLF4J pre 2.0 would always emit a warning that it found 2+ > > > StaticBinders) > > > > > > > > Interestingly, you could still whatever backend (i.e. Log4J) and pipe > > > > it through to another backend via another appender. > > > > This is what SLF4J refers to bridges -- although you have to be sure > > > > not to create a circular loop. > > > > > > > > Then there is something also general like syslog. > > > > > > > > On Thu, Mar 20, 2025 at 1:15 AM Chia-Ping Tsai <chia7...@gmail.com> > > > wrote: > > > > > > > > > > hi Ismael > > > > > > > > > > thanks for all your response. > > > > > > > > > > > All that said, I am not actually sure we can do what I described > > > above > > > > > while maintaining the compatibility required by a minor release. > > > > > > > > > > Excuse me, are your concerns related to version matching, as > > discussed > > > in > > > > > [0]? If so, I concur that this represents a compatibility issue, > and > > > the > > > > > target version for this change should be 5.0. Additionally, there > > was a > > > > > related discussion previously documented in [1]. While we have not > > > > strictly > > > > > enforced version matching during prior SLF4J updates, this KIP > > provides > > > > an > > > > > opportunity to establish guidelines for upgrading sl4fj that have > > > direct > > > > > compatibility implications. > > > > > > > > > > [0] https://www.slf4j.org/faq.html#compatibility > > > > > [1] > > https://github.com/apache/kafka/pull/16324#discussion_r1644671854 > > > > > > > > > > To Teng > > > > > > > > > > Could you please revise the KIP according to following benefit? > > > > > > > > > > > The key benefit of this KIP is that you can add new logging > > backends > > > > and > > > > > select it via a config. This is how most pluggable things work. But > > it > > > is > > > > > *not* how slf4j 1.x works. slf4j 1.x requires you to *remove* the > > > default > > > > > logging library picked by the project as well. That's much more > > > > intrusive. > > > > > > > > > > Best, > > > > > Chia-Ping > > > > > > > > > > Ismael Juma <m...@ismaeljuma.com> 於 2025年3月20日 週四 上午8:10寫道： > > > > > > > > > > > Hi Chia-Ping, > > > > > > > > > > > > I don't think we're in the business of shipping multiple logging > > > > libraries. > > > > > > Here's my take: > > > > > > > > > > > > 1. We should pick one logging library for services/servers, ship > it > > > and > > > > > > include a configuration file for it. > > > > > > 2. For clients, we leave it to the users to select the logging > > > library > > > > - > > > > > > clients run alongside applications and it's desirable to use the > > same > > > > > > logging library for both (if we were starting from scratch, we > may > > > have > > > > > > decided to also include our default logging library for clients > as > > > > well, > > > > > > but it's hard to make that change now). > > > > > > 3. For the cases where users want to use a different logging > > library > > > > for > > > > > > services/servers (perhaps because they have standardized on a > > > different > > > > > > logging library), they would have to add the additional jar to > the > > > > > > classpath and change the relevant logging config. This is no > > > different > > > > than > > > > > > adding a different authorizer or any other pluggable component. > > > > > > > > > > > > The key benefit of this KIP is that you can add new logging > > backends > > > > and > > > > > > select it via a config. This is how most pluggable things work. > But > > > it > > > > is > > > > > > *not* how slf4j 1.x works. slf4j 1.x requires you to *remove* the > > > > default > > > > > > logging library picked by the project as well. That's much more > > > > intrusive. > > > > > > > > > > > > All that said, I am not actually sure we can do what I described > > > above > > > > > > while maintaining the compatibility required by a minor release. > > > > > > > > > > > > Ismael > > > > > > > > > > > > On Wed, Mar 19, 2025, 2:03 PM Chia-Ping Tsai <chia7...@gmail.com > > > > > > wrote: > > > > > > > > > > > > > hi Ismael > > > > > > > > > > > > > > > but they must also add whichever logging library they want to > > > use. > > > > > > > > > > > > > > If users are required to modify JAR files to alter the SLF4J > > > > provider, > > > > > > the > > > > > > > value of this KIP is significantly diminished. I believe the > > > primary > > > > > > > benefit of this KIP lies in enabling users to configure a > system > > > > property > > > > > > > for SLF4J provider changes without JAR modifications. > > Furthermore, > > > by > > > > > > > managing all SLF4J and provider JARs within Kafka, we can > ensure > > > > SLF4J > > > > > > > version updates without compatibility concerns, as we can > > guarantee > > > > > > > provider JAR consistency with the SLF4J version in the > > > distribution. > > > > > > > > > > > > > > Best, > > > > > > > Chia-Ping > > > > > > > > > > > > > > Ismael Juma <m...@ismaeljuma.com> 於 2025年3月19日 週三 下午12:00寫道： > > > > > > > > > > > > > > > Hi TengYao, > > > > > > > > > > > > > > > > It's a bit difficult to review the KIP. I don't follow most > of > > > the > > > > > > > > motivation. The only one that I follow is: > > > > > > > > > > > > > > > > "Our current build configuration employs fragile dependency > > > > management > > > > > > > > tricks to handle SLF4J backends. We can eliminate these > brittle > > > > build > > > > > > > > mechanisms by transitioning to explicit provider dependencies > > > after > > > > > > > > upgrading to 2.0." > > > > > > > > > > > > > > > > Additionally, I don't think we should do the following: > > > > > > > > > > > > > > > > "Add other popular slf4j backend binding provider > > dependencies." > > > > > > > > > > > > > > > > A reasonable approach, in my opinion, would be: > > > > > > > > > > > > > > > > 1. Include the log4j2 dependency with the server modules and > > not > > > > > > include > > > > > > > > them with the client modules. > > > > > > > > 2. Automatically configure the log4j2 dependency for the > server > > > > > > modules. > > > > > > > > Users can override them via the system property, but they > must > > > > also add > > > > > > > > whichever logging library they want to use. > > > > > > > > 3. Somehow configure slf4j 2.x to work like 1.x out of the > box > > > for > > > > the > > > > > > > > client module (for compatibility reasons). > > > > > > > > > > > > > > > > But I don't know if `3` is possible. If `3` is not possible, > I > > > > don't > > > > > > see > > > > > > > > how we can make this a compatible change. > > > > > > > > > > > > > > > > Ismael > > > > > > > > > > > > > > > > On Tue, Mar 18, 2025 at 7:41 PM TengYao Chi < > > kiting...@gmail.com > > > > > > > > > > wrote: > > > > > > > > > > > > > > > > > Hello everyone, > > > > > > > > > > > > > > > > > > I want to bump this thread manually. > > > > > > > > > Any feedback or vote would be appreciated. > > > > > > > > > > > > > > > > > > Best, > > > > > > > > > TengYao > > > > > > > > > > > > > > > > > > TengYao Chi <kiting...@gmail.com> 於 2025年3月10日 週一 > 上午11:47寫道： > > > > > > > > > > > > > > > > > > > Hello guys, > > > > > > > > > > I would like to remind you in the vote thread that the > KIP > > > has > > > > been > > > > > > > > > > updated, and I apologize for repeating it. > > > > > > > > > > I have taken over this KIP from Muralidhar. > > > > > > > > > > Since the original content is outdated as the logging > > > > framework has > > > > > > > > been > > > > > > > > > > widely changed, I have updated the content of the KIP. > > > > > > > > > > Please take a look and share your thoughts. > > > > > > > > > > > > > > > > > > > > Best Regards, > > > > > > > > > > TengYao > > > > > > > > > > > > > > > > > > > > Muralidhar Basani <muralidhar.bas...@aiven.io.invalid> 於 > > > > > > 2024年9月24日 > > > > > > > 週二 > > > > > > > > > > 上午5:09寫道： > > > > > > > > > > > > > > > > > > > >> Hi, > > > > > > > > > >> > > > > > > > > > >> I wanted to gently follow up on this thread in case > anyone > > > > has any > > > > > > > > > >> thoughts > > > > > > > > > >> or would like to take a look. > > > > > > > > > >> > > > > > > > > > >> Thanks, > > > > > > > > > >> Murali > > > > > > > > > >> > > > > > > > > > >> On Mon, Sep 16, 2024 at 10:23 AM Muralidhar Basani < > > > > > > > > > >> muralidhar.bas...@aiven.io> wrote: > > > > > > > > > >> > > > > > > > > > >> > Thanks Chia. > > > > > > > > > >> > > > > > > > > > > >> > I have updated KIP with this quote, in the migration > > plan > > > > > > section. > > > > > > > > > >> > > > > > > > > > > >> > Thanks, > > > > > > > > > >> > Murali > > > > > > > > > >> > > > > > > > > > > >> > On Sun, Sep 15, 2024 at 3:30 PM Chia-Ping Tsai < > > > > > > > chia7...@gmail.com> > > > > > > > > > >> wrote: > > > > > > > > > >> > > > > > > > > > > >> >> > > > > > > > > > >> >> > Muralidhar Basani <muralidhar.bas...@aiven.io > > .invalid> > > > 於 > > > > > > > > > 2024年9月15日 > > > > > > > > > >> >> 晚上9:02 寫道： > > > > > > > > > >> >> > > > > > > > > > > >> >> > With this, I think, users don't have to make any > > > explicit > > > > > > > changes > > > > > > > > > in > > > > > > > > > >> >> their > > > > > > > > > >> >> > code, if their provider is reload4j. And if it's a > > > > different > > > > > > > > > provider > > > > > > > > > >> >> (like > > > > > > > > > >> >> > logback, log4j), they would have to upgrade that to > > > > match it > > > > > > > with > > > > > > > > > >> slf4j. > > > > > > > > > >> >> > > > > > > > > > >> >> If upgrading the matched provider is the only > explicit > > > > change > > > > > > and > > > > > > > > we > > > > > > > > > >> >> expect users have responsibility to keep consistent > > > version > > > > > > when > > > > > > > > > using > > > > > > > > > >> >> other providers , could we write it down to the KIP? > > > > > > > > > >> >> > > > > > > > > > >> >> That means we will update slf4j without KIP in the > > future > > > > > > except > > > > > > > > for > > > > > > > > > >> >> specific reason. > > > > > > > > > >> >> > > > > > > > > > >> >> Best, > > > > > > > > > >> >> Chia-Ping > > > > > > > > > >> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > >
