Hi, I appreciate everyone's work on the JDK upgrade. It would be ideal if we could migrate JDK like this. That will give us an incremental migration path.
1. (now) Hive 4.0 compiled by JDK 8, and run by JDK 8 2. Hive 4.x compiled by JDK 8, and run by JDK 17 3. Hive 4.x compiled by JDK 17, and run by JDK 17 I guess it is technically possible that Hive supports JDK 8 and 17 at runtime. Apache Spark supports JDK 8, 11, and 17. What is the most challenging point? Regards, Okumin On Thu, Aug 29, 2024 at 10:48 PM Butao Zhang <butaozha...@163.com> wrote: > Hi Akshat, > > Thanks for your reply! I agree that supporting both JDK 8 and JDK 17 is > not easy. I think we can safely to move to JDK 17 if we can ensure *hive > application* can access to HiveServer2/Metastore without forcing to use > the JDK 17. > For example, if a hive application is developed by JDK 8, it can import > the *hive-exec.jar & hive-jdbc.jar* which is compiled by JDK 17 and can > access to hs2/hms successfully. If this case can be supported, I think we > can move to JDK 17 as soon as possible. > > BTW, If Apache Hive , Apache Tez and Apache Hadoop are all compiled with > JDK17 or higer version, i think the hive job will run more faster. Users > will enjoy significant performance improvements due to JDK version > upgrades. > I'm looking forward to the performance improvement. :) > > Thanks, > Butao Zhang > ---- Replied Message ---- > From Akshat m<akshatats...@gmail.com> <undefined> > Date 8/29/2024 18:44 > To <dev@hive.apache.org> <dev@hive.apache.org> > Subject Re: Move to JDK-11 > Hi Butao, > > It won’t be feasible to support both JDK 8 and JDK 17 in the same Hive > version due to compatibility issues and behavior changes, as detailed in > the PR. The internal changes in JDK implementations have made this > challenging. Additionally, many other libraries maintain separate versions > for JDK 17+ and JDK 8 support. > > To align with this approach, I suggest we continue with Hive 4 supporting > JDK 8. Once the JDK 17 PR is merged, we could plan a separate release that > specifically targets JDK 17+ support. This would ensure that Hive stays > up-to-date while maintaining stability across different environments. > > Would love to hear your thoughts or any additional suggestions the > community might have! > > Regards > > Akshat > > > > On Wed, Aug 28, 2024 at 8:00 AM Butao Zhang <butaozha...@163.com> wrote: > >> Good job! Thanks everyone who put effort into the JDK upgrading. That's a >> big good news for Apache Hive community! >> Just a question: Given that many people are still using the lower jdk8 >> version. Do we consider to keeping compatibility with lower jdk8 version? >> >> >> Thanks, >> Butao Zhang >> ---- Replied Message ---- >> From Akshat m<akshatats...@gmail.com> <undefined> >> Date 8/27/2024 23:12 >> To dev<dev@hive.apache.org> <dev@hive.apache.org> >> Cc kokila narayanan<kokilanarayana...@gmail.com>, >> <kokilanarayana...@gmail.com>Raghav >> Aggarwal<raghavaggarwal03...@gmail.com>, >> <raghavaggarwal03...@gmail.com>Ayush Saxena<ayush...@gmail.com>, >> <ayush...@gmail.com>tanishq chugh<tanishqchugh4...@gmail.com>, >> <tanishqchugh4...@gmail.com><aturo...@cloudera.com> >> <aturo...@cloudera.com> >> Subject Re: Move to JDK-11 >> >> Hi everyone, >> >> I’m excited to share that after months of dedicated work, we’ve finally >> achieved a passing PR[1] for running *Apache Hive on JDK 17*! >> >> This was a challenging task, requiring to address various compatibility >> issues and ensure our comprehensive test suite passed without a hitch. >> >> Thanks Kokila and Tanishq for joining me in the initiative and Raghav for >> contributing fix for the errorProne issue, and thanks to Attila, Ayush and >> other Hive PMC members for their guidance in the initiative >> >> It’s been a real team effort, and I’m grateful for the collaboration and >> support from everyone involved. >> >> I’d appreciate it if you could take some time to review the PR. Community >> feedback is really important to make this happen. >> >> The PR #5404 addresses the integration of Apache Hive with JDK 17, >> introducing a range of changes aimed at ensuring compatibility and >> improving the codebase. Here’s a detailed breakdown: >> >> 1. Core Code Adjustments: >> >> - Compatibility Fixes: Various parts of the Hive codebase have been >> modified to ensure they work seamlessly with JDK 17. This includes >> addressing language or API changes that JDK 17 introduces. >> >> - Reflection and Modules: Adjustments related to how Hive interacts >> with Java's reflection API, which has stricter module system rules in JDK >> 17. >> >> 2. Dependency Management: >> >> - Version Upgrades: Several third-party dependencies have been updated >> to versions that are compatible with JDK 17. This is critical because older >> versions of some libraries may not work properly or might use deprecated >> features removed in JDK 17. >> >> - Data nucleus upgrade: To ensure compatibility with Java 17, we >> upgraded DataNucleus to version 6.x. This upgrade, however, introduced an >> issue with Hive schema creation. The issue stems from a commit in >> DataNucleus[2] that alters the handling of primary key column names by >> enclosing them in double quotes. This change caused conflicts during Hive >> schema generation. As an interim solution, we implemented a patched >> DataNucleus-rdbms module to address the issue in Hive. The necessary fix is >> now included in the DataNucleus repository[3]. Once DataNucleus 6.0.8 is >> officially released, we will upgrade to this version and remove the >> temporary patched module. >> >> - Dependency Cleanup: Unused or outdated dependencies were removed, >> reducing potential conflicts and simplifying the build process. >> >> 3. Testing Enhancements: >> >> - Test Suite Updates: The PR includes updates to Hive’s test suite to >> ensure all tests pass under JDK 17. This likely involved modifying tests >> that needed adjustments to align with JDK 17’s behaviour. >> >> - New Tests: New tests may have been added to cover areas where JDK 17 >> introduced new features or changes that could impact Hive. >> >> 4. Compatibility Checks: >> >> - JDK Version Verification: A new compatibility check has been >> introduced to verify the JDK version at runtime. This ensures that Hive is >> running on an appropriate JDK version, preventing unexpected errors or >> behaviour due to version mismatches. >> >> 5. Build Updates: >> >> - Build System Tweaks: The PR likely includes changes to the Maven build >> configuration to accommodate JDK 17, ensuring that the build process is >> smooth and compatible. >> >> Overall, this PR is a comprehensive effort to bring Apache Hive up to >> date with the latest Java platform, ensuring long-term maintainability and >> performance improvements. >> >> As the next steps, we’ll be focusing on further sanity checks and >> extensive testing to make sure everything functions smoothly. Community >> involvement in this process will be crucial, so please do share any issues >> or feedback as you test the changes. >> >> I’m looking forward to seeing the benefits this will bring to our >> community and users. >> >> >> [1] https://github.com/apache/hive/pull/5404 >> >> [2] >> https://github.com/datanucleus/datanucleus-rdbms/commit/b99df3c803862f9e9feacc7891461bcb21160b38 >> >> [3] >> https://github.com/datanucleus/datanucleus-rdbms/commit/b99df3c803862f9e9feacc7891461bcb21160b38 >> >> >> Best regards, >> >> Akshat Mathur >> >> >> On Tue, 11 Jul 2023 at 2:25 PM, Attila Turoczy < >> aturo...@cloudera.com.invalid> wrote: >> >>> Returning to this topic, I kindly request those who would like to >>> advocate >>> for the continued support of JDK8 to please share their reasoning and >>> insights with us. Your input and perspective are greatly appreciated! >>> Thank you. >>> >>> -Attila >>> >>> On Fri, Jun 2, 2023 at 12:43 PM Attila Turoczy <aturo...@cloudera.com> >>> wrote: >>> >>> > Hi All, >>> > >>> > I know my opinion might not be the most popular, but I advocate for >>> using* >>> > JDK 17*. Here's why: >>> > >>> > Let's consider a scenario where a customer wants to use the latest >>> version >>> > of Apache Hive. They would typically install it locally or on a small >>> > cluster. In 2023, is it realistic to assume that this customer won't be >>> > able to install JDK 17 on their cluster? Even in large enterprises, it >>> > should be feasible to install an LTS JDK, especially considering the >>> > widespread adoption of cloud computing. Sungwoo Park's measurements >>> also >>> > support this recommendation to go with JDK 17. It outperforms JDK 11 >>> by 8% >>> > in terms of runtime speed, and JDK 11 itself is 10+% faster than JDK 8. >>> > This is a significant value proposition. Who would be the customer that >>> > says, "I don't want faster query execution! I'd rather use JDK 8 and >>> pay >>> > more for cloud or data center resources instead of using JDK 17!" It >>> > doesn't make sense to me. >>> > >>> > The tech industry has been evolving at an incredible pace, with >>> > improvements in serialization, IPC mechanisms, and parallelized >>> frameworks >>> > since the release of JDK 8 ten years ago. We should leverage these >>> > advancements! Couple years ago, we invested a lot to improve 1-2% of >>> the >>> > execution. We prayed for 3 gods, sacrificed 2 ships and traveled >>> around the >>> > world to make it happen. :-) Now, the JDK itself provides a substantial >>> > amount of improvement. So, why would we resist progress just because >>> there >>> > are a few lazy or conservative admins who don't want to spend two >>> minutes >>> > installing a JDK? >>> > >>> > A platform needs to be modern and incorporate the latest technologies >>> to >>> > attract developers and users. I understand that some may prefer to stay >>> > with JDK 8 as it seems like the safest position, but I believe in >>> taking >>> > bold bets to achieve big wins. Even if we decide to stick with JDK 8, I >>> > would still be happy since we are moving forward and not dwelling on a >>> JDK >>> > that is a decade old. Personally, I think focusing on one thing that >>> brings >>> > more value to us and our users is the idealistic path forward. >>> > >>> > -Attila >>> > >>> > >>> > >>> > >>> > >>> > >>> > >>> > >>> > >>> > On Thu, Jun 1, 2023 at 11:23 AM Stamatis Zampetakis <zabe...@gmail.com >>> > >>> > wrote: >>> > >>> >> Hey everyone, >>> >> >>> >> If we claim that Hive supports a certain JDK then we should compile >>> and >>> >> run >>> >> tests with it. >>> >> >>> >> The more JDKs we can support the better for everyone but this comes >>> at a >>> >> cost (resources mostly). We should have a precommit run for every >>> >> supported >>> >> JDK (frequency to be determined once per day/week) that compiles and >>> run >>> >> all tests. >>> >> >>> >> From my perspective, I would be pretty happy if we could cover the two >>> >> edge >>> >> LTS releases at every point in time. >>> >> >>> >> Then we have to decide also which JDK shall we use for the pull >>> requests >>> >> and local dev environment. I think it makes sense to use the latest. >>> >> People >>> >> like working on modern stuff and also it makes sense that newer >>> releases >>> >> will also use newer versions. It would be pretty awkward if someone >>> wants >>> >> to use the latest Hive version and it turns out that it can only run >>> on >>> >> JDK8. >>> >> >>> >> Best, >>> >> Stamatis >>> >> >>> >> On Thu, Jun 1, 2023, 3:42 AM Sungwoo Park <glap...@gmail.com> wrote: >>> >> >>> >> > Hi, everyone. >>> >> > >>> >> > I have not tested the master branch with Java 11/17 yet, but I would >>> >> like >>> >> > to share my experience with testing a fork of branch-3.1 with Java >>> 11/17 >>> >> > (as part of developing Hive-MR3), in case that it can be useful for >>> the >>> >> > discussion. I merged the patches listed in [1] HIVE-22415 and >>> updated >>> >> the >>> >> > Maven configuration for Java 11. >>> >> > >>> >> > 1. Building Hive was fine and I was able to run it with Java 11 as >>> well >>> >> as >>> >> > Java 17. So, it seems that the work reported in [1] is indeed >>> complete >>> >> for >>> >> > upgrading to Java 11 (and Java 17) and getting Hive to work. >>> >> > >>> >> > 2. However, there was a problem with running tests, so this can be >>> >> > additional work for upgrading to Java 11. >>> >> > >>> >> > 3. For performance, Java 17 gives about 8 percent of (free) >>> performance >>> >> > improvement. When tested with 10TB TPC-DS, Java 8 takes 8074 >>> seconds, >>> >> > whereas Java 17 takes 7415 seconds. Considering the maturity of >>> Hive, I >>> >> > think this is not a small improvement because almost every query >>> gets >>> >> some >>> >> > speedup. >>> >> > >>> >> > Thanks, >>> >> > >>> >> > --- Sungwoo >>> >> > >>> >> > [1] https://issues.apache.org/jira/browse/HIVE-22415 >>> >> > >>> >> > >>> >> > On Thu, Jun 1, 2023 at 3:53 AM Sai Hemanth Gantasala >>> >> > <saihema...@cloudera.com.invalid> wrote: >>> >> > >>> >> > > Hi All, >>> >> > > >>> >> > > I would strongly advocate keeping support for JDK8. >>> >> > > Between JDK11 and JDK17, Depending on the amount of effort on the >>> >> upgrade >>> >> > > I'm inclined towards JDK17 (JDK21 LTS will be released in Sep >>> 2023). >>> >> > > >>> >> > > Thanks, >>> >> > > Sai. >>> >> > > >>> >> > > On Wed, May 31, 2023 at 5:39 AM László Bodor < >>> >> bodorlaszlo0...@gmail.com> >>> >> > > wrote: >>> >> > > >>> >> > > > *Hi!* >>> >> > > > >>> >> > > > >>> >> > > > *Should we support both JDK-11 & JDK-8?* >>> >> > > > IMO absolutely yes, let's not break up with JDK-8: according to >>> its >>> >> > > > lifecycle, it's going to stay with us for a long time. >>> >> > > > >>> >> > > > I believe >>> >> > > > a) we should be able to compile on JDK8, JDK11, and JDK17 >>> (github >>> >> > actions >>> >> > > > can cover this conveniently in precommit time, like tez >>> >> > > > < >>> >> https://github.com/apache/tez/blob/master/.github/workflows/build.yml >>> >> > >) >>> >> > > > b) the release artifacts should be compatible with JDK8 as long >>> as >>> >> it >>> >> > is >>> >> > > > with us. >>> >> > > > >>> >> > > > Regards, >>> >> > > > Laszlo Bodor >>> >> > > > >>> >> > > > >>> >> > > > Butao Zhang <butaozha...@163.com> ezt írta (időpont: 2023. máj. >>> >> 31., >>> >> > > Sze, >>> >> > > > 14:33): >>> >> > > > >>> >> > > > > Thanks Ayush for driving this! Good to know that Hive is >>> getting >>> >> > ready >>> >> > > > for >>> >> > > > > newer JDK. >>> >> > > > > From my opinon, if we have more community energy to put into >>> it, >>> >> we >>> >> > can >>> >> > > > > support both JDK-11 and JDK-17 like Spark[1]. If we have to >>> make >>> >> a >>> >> > > > choice >>> >> > > > > between a JDK-11 and JDK-17, i would like to choose the >>> relatively >>> >> > new >>> >> > > > > version JDK-17, meanwhile, we should maintain compatibility >>> with >>> >> > jdk8, >>> >> > > as >>> >> > > > > JDK-8 is still widely used in most big data platforms. >>> >> > > > > >>> >> > > > > >>> >> > > > > Thanks, >>> >> > > > > Butao Zhang >>> >> > > > > >>> >> > > > > >>> >> > > > > [1]https://issues.apache.org/jira/browse/SPARK-33772 >>> >> > > > > ---- Replied Message ---- >>> >> > > > > | From | Ayush Saxena<ayush...@gmail.com> | >>> >> > > > > | Date | 5/31/2023 18:39 | >>> >> > > > > | To | dev<dev@hive.apache.org> | >>> >> > > > > | Subject | Move to JDK-11 | >>> >> > > > > Hi Everyone, >>> >> > > > > Want to pull in the attention of folks towards moving to >>> JDK-11 >>> >> > compile >>> >> > > > > time support in Hive. There was a ticket in the past [1] which >>> >> talks >>> >> > > > about >>> >> > > > > it and If I could decode it right, it was blocked because the >>> >> Hadoop >>> >> > > > > version used by Hive didn't had JDK-11 runtime support, But >>> with >>> >> [2] >>> >> > in >>> >> > > > we >>> >> > > > > have upgraded the Hadoop version, so that problem is sorted >>> out. I >>> >> > > > couldn't >>> >> > > > > even see any unresolved tickets in the blocked state either. >>> >> > > > > >>> >> > > > > I quickly tried* a mvn clean install -DskipTests -Piceberg >>> >> -Pitests >>> >> > > > > -Dmaven.javadoc.skip=true >>> >> > > > > >>> >> > > > > And no surprises it failed with some weird exceptions towards >>> the >>> >> > end. >>> >> > > > But >>> >> > > > > I think that should be solvable. >>> >> > > > > >>> >> > > > > So, Questions? >>> >> > > > > >>> >> > > > > - What do folks think about this? Should we put in some effort >>> >> > towards >>> >> > > > > JDK-11 >>> >> > > > > - Should we support both JDK-11 & JDK-8? >>> >> > > > > - Ditch JDK-11 and directly shoot for JDK-17? >>> >> > > > > >>> >> > > > > Let me know your thoughts, In case anyone has some experience >>> in >>> >> this >>> >> > > > area >>> >> > > > > and have tried something in the context, feel free to share or >>> >> may be >>> >> > > if >>> >> > > > > someone has any potential action plan or so >>> >> > > > > >>> >> > > > > -Ayush >>> >> > > > > >>> >> > > > > [1] https://issues.apache.org/jira/browse/HIVE-22415 >>> >> > > > > [2] https://issues.apache.org/jira/browse/HIVE-24484 >>> >> > > > > >>> >> > > > > * changed the maven.compiler.source & maven.compiler.target >>> to 11 >>> >> > > > > >>> >> > > > >>> >> > > >>> >> > >>> >> >>> > >>> >>
