[Java][Dataset] JNI Error when reading parquet file

[fb64]

Describe the bug, including details regarding any error messages, version, and platform.

An JNI error happened when I try to read a parquet file with arrow-dataset (15.0.0) in a java program
Runtime : OpenJDK 64-Bit Server VM (build 17.0.2+8-86, mixed mode, sharing)
It seems that's due to the usage of NativeMemoryPool.createListenable( DirectReservationListener.instance()) instead of NativeMemoryPool.getDefault()
By the way the error not happened with all parquet files
Full code for reproduce the error
https://gist.github.com/fb64/71880cde297bc5234b02b68b785670fd

Stack trace

/Users/runner/work/crossbow/crossbow/arrow/java/dataset/src/main/cpp/jni_util.cc:79: Failed to update reservation while freeing bytes: JNIEnv was not attached to current thread
0   jnilib-10391875251891153172.tmp     0x00000001424a99ac _ZN5arrow4util7CerrLogD2Ev + 204
1   jnilib-10391875251891153172.tmp     0x00000001424a98ce _ZN5arrow4util7CerrLogD0Ev + 14
2   jnilib-10391875251891153172.tmp     0x00000001424a1112 _ZN5arrow4util8ArrowLogD1Ev + 34
3   jnilib-10391875251891153172.tmp     0x00000001412608ed _ZN5arrow7dataset3jni31ReservationListenableMemoryPool4Impl4FreeEPhxx + 237
4   jnilib-10391875251891153172.tmp     0x00000001427b2acb _ZN5arrow10PoolBufferD2Ev + 59
5   jnilib-10391875251891153172.tmp     0x00000001427b268e _ZN5arrow10PoolBufferD0Ev + 14
6   jnilib-10391875251891153172.tmp     0x000000014140f508 _ZN7parquet14SerializedFileD2Ev + 168
7   jnilib-10391875251891153172.tmp     0x000000014140ebfe _ZN7parquet14SerializedFileD0Ev + 14
8   jnilib-10391875251891153172.tmp     0x000000014153a8ef _ZN7parquet5arrow12_GLOBAL__N_114FileReaderImplD0Ev + 159
9   jnilib-10391875251891153172.tmp     0x000000014155048a _ZN5arrow8internal6FnOnceIFvRKNS_10FutureImplEEE6FnImplINS_6FutureINS0_5EmptyEE20WrapResultOnComplete8CallbackINSA_14ThenOnCompleteIZN7parquet5arrow17RowGroupGenerator9FetchNextEvEUlvE_NSA_17PassthruOnFailureISH_EEEEEEED0Ev + 122
10  jnilib-10391875251891153172.tmp     0x00000001424a503d _ZN5arrow18ConcreteFutureImpl22DoMarkFinishedOrFailedENS_11FutureStateE + 189
11  jnilib-10391875251891153172.tmp     0x00000001428471d1 _ZN5arrow6FutureINS_8internal5EmptyEE14DoMarkFinishedENS_6ResultIS2_EE + 129
12  jnilib-10391875251891153172.tmp     0x0000000142847055 _ZN5arrow6FutureINS_8internal5EmptyEE12MarkFinishedIS2_vEEvNS_6StatusE + 53
13  jnilib-10391875251891153172.tmp     0x000000014154f4a4 _ZZN5arrow8internal8Executor10DoTransferINS0_5EmptyENS_6FutureIS3_EENS_6StatusEEENS4_IT_EES8_bENUlRKS6_E_clESA_ + 100
14  jnilib-10391875251891153172.tmp     0x00000001424a5650 _ZN5arrow8internal6FnOnceIFvvEE6FnImplIZNS_18ConcreteFutureImpl21RunOrScheduleCallbackERKNSt3__110shared_ptrINS_10FutureImplEEEONS8_14CallbackRecordEbEUlvE_E6invokeEv + 32
15  jnilib-10391875251891153172.tmp     0x000000014248e055 _ZNSt3__1L14__thread_proxyINS_5tupleIJNS_10unique_ptrINS_15__thread_structENS_14default_deleteIS3_EEEEZN5arrow8internal10ThreadPool21LaunchWorkersUnlockedEiE4$_10EEEEEPvSC_ + 709
16  libsystem_pthread.dylib             0x00007ff810c77202 _pthread_start + 99
17  libsystem_pthread.dylib             0x00007ff810c72bab thread_start + 15

Component(s)

Java

[isimanovich]

seeing similar issue as in the linked issue from Kotlin/dataframe after upgrade to dataset 15.0.0

Exception in thread "main" java.util.concurrent.ExecutionException: java.lang.UnsatisfiedLinkError: /opt/amazon/var/tmp/jnilib-17078712129987048802.tmp: 
/opt/amazon/var/tmp/jnilib-17078712129987048802.tmp: undefined symbol: _ZTIN6google8protobuf7MessageE
	at org.apache.arrow.dataset.jni.JniLoader.load(JniLoader.java:92)
	at org.apache.arrow.dataset.jni.JniLoader.loadRemaining(JniLoader.java:75)
	at org.apache.arrow.dataset.jni.JniLoader.ensureLoaded(JniLoader.java:61)
	at org.apache.arrow.dataset.jni.NativeMemoryPool.getDefault(NativeMemoryPool.java:34)

[fb64]

seeing similar issue as in the linked issue from Kotlin/dataframe after upgrade to dataset 15.0.0

Exception in thread "main" java.util.concurrent.ExecutionException: java.lang.UnsatisfiedLinkError: /opt/amazon/var/tmp/jnilib-17078712129987048802.tmp: 
/opt/amazon/var/tmp/jnilib-17078712129987048802.tmp: undefined symbol: _ZTIN6google8protobuf7MessageE
	at org.apache.arrow.dataset.jni.JniLoader.load(JniLoader.java:92)
	at org.apache.arrow.dataset.jni.JniLoader.loadRemaining(JniLoader.java:75)
	at org.apache.arrow.dataset.jni.JniLoader.ensureLoaded(JniLoader.java:61)
	at org.apache.arrow.dataset.jni.NativeMemoryPool.getDefault(NativeMemoryPool.java:34)

It's not excatly the same issue, we also faced to it on Kotlin dataframe : Kotlin/dataframe#577 (comment)
It seems that it only happened on Linux, I think it's native library compilation issue as the library seems using Protobof but is not statically linked to it. It's working with arrow 14 by the way ....

[isimanovich]

yes, i did see in your thread on Kotlin/dataframe#577 (comment) that downgrade to 14.0.2 helped, so I will be doing that for now. I wanted to point out that in my case it is also issue on Linux and it works fine on Mac M1 Pro.

While I understand that this is not exactly the same issue as current ticket, I was pointing out that it was similar to Kotlin/dataframe#577 (comment)

It seems that's due to the usage of NativeMemoryPool.createListenable( DirectReservationListener.instance()) instead of NativeMemoryPool.getDefault()

in my case, im seeing this behavior with both NativeMemoryPool.createListenable( DirectReservationListener.instance()) and NativeMemoryPool.getDefault()

[davisusanibar]

Hi Team,

The following Java jars were downloaded from the v15 release and compared with the v14 release:

V15

• Jar: https://repo1.maven.org/maven2/org/apache/arrow/arrow-dataset/15.0.0/arrow-dataset-15.0.0.jar

$ unzip arrow-dataset-15.0.0.jar
$ shasum -a 256 aarch_64/libarrow_dataset_jni.so
fd4d8a470815522e69e174c34840295a9e8e2e021575d203697ae09e1c864cca  aarch_64/libarrow_dataset_jni.so
$ objdump -p aarch_64/libarrow_dataset_jni.so | grep NEED
  NEEDED       librt.so.1
  NEEDED       libpthread.so.0
  NEEDED       libdl.so.2
  NEEDED       libstdc++.so.6
  NEEDED       libm.so.6
  NEEDED       libgcc_s.so.1
  NEEDED       libc.so.6
  NEEDED       ld-linux-aarch64.so.1
$ nm aarch_64/libarrow_dataset_jni.so | grep ZTIN6google8protobuf7Messag
                 U _ZTIN6google8protobuf7MessageE

U: The symbol is undefined (not found in this object)

V14:

• Jar: https://repo1.maven.org/maven2/org/apache/arrow/arrow-dataset/14.0.2/arrow-dataset-14.0.2.jar

$ unzip arrow-dataset-14.0.2.jar
$ shasum -a 256 aarch_64/libarrow_dataset_jni.so
c21cabcd8aaaea8c19197de1955ae0456830c386ea50bd9a298b0e047c4ce932  aarch_64/libarrow_dataset_jni.so
$ objdump -p aarch_64/libarrow_dataset_jni.so | grep NEED
  NEEDED       librt.so.1
  NEEDED       libpthread.so.0
  NEEDED       libdl.so.2
  NEEDED       libstdc++.so.6
  NEEDED       libm.so.6
  NEEDED       libgcc_s.so.1
  NEEDED       libc.so.6
  NEEDED       ld-linux-aarch64.so.1
$ nm aarch_64/libarrow_dataset_jni.so | grep ZTIN6google8protobuf7Messag
0000000002ecddb8 V _ZTIN6google8protobuf7MessageE

Glossary:
T: The symbol is in the text (code) section
U: The symbol is undefined (not found in this object)
W: The symbol is weak, which means it has lesser priority than a strong global symbol
V: The symbol is weak in the symbol table

Using https://github.com/apache/arrow/tree/a61f4af724cd06c3a9b4abd20491345997e532c0, I have built the libraries and can read dataset files.

[davisusanibar]

v15 release branch also uploaded to Gihub all Java artifacts as assets (v15 release: https://github.com/ursacomputing/crossbow/releases/tag/release-15.0.0-rc1-0-github-java-jars)

@raulcd Are there some way to try to upload again release branch to Github as an assets to test it again (commit: a61f4af).

It seems like there was a problem linking protobuf static libraries in the build process or something like that.

[raulcd]

The only thing we could do is create a new patch release but I would really like this to be diagnosed and tested / validated if we decide to do a new release. @lidavidm any thoughts? Could 15.0.0 be somehow yanked? BTW we seem to have been able to reproduce the same issue on the java cookbooks:
apache/arrow-cookbook#343 (comment)

[lidavidm]

I'm not sure we can yank a maven package.

Why isn't this caught in the java-jars pipeline?

[fb64]

Any hint about the main error of this issue ?
/Users/runner/work/crossbow/crossbow/arrow/java/dataset/src/main/cpp/jni_util.cc:79: Failed to update reservation while freeing bytes: JNIEnv was not attached to current thread
It also happened with Arrow 14 BTW

[isimanovich]

seeing following error when running on 14.0.1 NativeMemoryPool pool = NativeMemoryPool.createListenable(listener); or NativeMemoryPool.createListenable(DirectReservationListener.instance())

    /arrow/java/dataset/src/main/cpp/jni_util.cc:79: Failed to update reservation while freeing bytes: JNIEnv was not attached to current thread
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(+0x1623f88)[0x7f34b1feef88]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(_ZN5arrow4util8ArrowLogD1Ev+0xed)[0x7f34b1fef33d]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(_ZN5arrow7dataset3jni31ReservationListenableMemoryPool4FreeEPhll+0x47d)[0x7f34b1591d1d]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(_ZNSt23_Sp_counted_ptr_inplaceIN5arrow9ArrayDataESaIS1_ELN9__gnu_cxx12_Lock_policyE2EE10_M_disposeEv+0x10e)[0x7f34b1796d3e]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(_ZNSt23_Sp_counted_ptr_inplaceIN5arrow3ipc25RecordBatchFileReaderImpl28CachedRecordBatchReadContextESaIS3_ELN9__gnu_cxx12_Lock_policyE2EE10_M_disposeEv+0x9e)[0x7f34b19c739e]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(+0xbbf2ca)[0x7f34b158a2ca]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(_ZN5arrow8internal6FnOnceIFvRKNS_10FutureImplEEE6FnImplINS_6FutureINS0_5EmptyEE21WrapResultyOnComplete8CallbackINSA_14ThenOnCompleteIZZNS_3ipc25RecordBatchFileReaderImpl21ReadCachedRecordBatchEiNS8_ISt10shared_ptrINSE_7MessageEEEEENKUlRKSI_E0_clESL_EUlvE_NSA_17PassthruOnFailureISN_EEEEEEED0Ev+0x2e)[0x7f34b19c66ae]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(_ZN5arrow18ConcreteFutureImpl22DoMarkFinishedOrFailedENS_11FutureStateE+0x1a9)[0x7f34b20171e9]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(_ZN5arrow6FutureINS_8internal5EmptyEE12MarkFinishedIS2_vEEvNS_6StatusE+0xaa)[0x7f34b15e212a]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(+0x1647846)[0x7f34b2012846]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(_ZN5arrow18ConcreteFutureImpl22DoMarkFinishedOrFailedENS_11FutureStateE+0x195)[0x7f34b20171d5]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(+0x1688865)[0x7f34b2053865]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(_ZNO5arrow8internal6FnOnceIFvvEEclEv+0x165)[0x7f34b1fed1c5]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(+0x161e848)[0x7f34b1fe9848]
    /opt/amazon/var/tmp/jnilib-639979346264002289.tmp(+0x2a971a0)[0x7f34b34621a0]
    /lib64/libc.so.6(+0x9f7f2)[0x7facc71477f2]
    /lib64/libc.so.6(clone+0x44)[0x7facc70e7314]
    #
    # A fatal error has been detected by the Java Runtime Environment:
    #
    #  SIGSEGV (0xb) at pc=0x00007facc70d0898, pid=1, tid=234
    #
    # JRE version: OpenJDK Runtime Environment 1.0.1618.0 (17.0.10+8) (build 17.0.10+8-LTS)
    # Java VM: OpenJDK 64-Bit Server VM 1.0.1618.0 (17.0.10+8-LTS, mixed mode, tiered, compressed class ptrs, g1 gc, linux-amd64)
    #
    # Problematic frame:
    # C  [libc.so.6+0x28898]  abort+0x178
    #
    # Core dump will be written. Default location: //core.1
    #
    # An error report file with more information is saved as:
    # /tmp/hs_err_pid1.log
    #
    # If you would like to submit a bug report, please visit:
    #   https://bugreport.java.com/bugreport/crash.jsp

[davisusanibar]

I'm not sure we can yank a maven package.

Why isn't this caught in the java-jars pipeline?

Java-jars is running unders macOS environment, then all the validation finished without problem on macOS: https://github.com/ursacomputing/crossbow/actions/runs/7543210134/job/20535105142

Operating System
  macOS
  12.7.2
  21G1974
Runner Image
  Image: macos-12
  Version: 202[3](https://github.com/ursacomputing/crossbow/actions/runs/7543210134/job/20535105142#step:1:3)121[6](https://github.com/ursacomputing/crossbow/actions/runs/7543210134/job/20535105142#step:1:7).1
  Included Software: https://github.com/actions/runner-images/blob/macOS-12/20231216.1/images/macos/macos-12-Readme.md
  Image Release: https://github.com/actions/runner-images/releases/tag/macOS-12%2F20231216.1

[davisusanibar]

I'm not sure we can yank a maven package.
Why isn't this caught in the java-jars pipeline?

Java-jars is running unders macOS environment, then all the validation finished without problem on macOS: https://github.com/ursacomputing/crossbow/actions/runs/7543210134/job/20535105142

Operating System
  macOS
  12.7.2
  21G1974
Runner Image
  Image: macos-12
  Version: 202[3](https://github.com/ursacomputing/crossbow/actions/runs/7543210134/job/20535105142#step:1:3)121[6](https://github.com/ursacomputing/crossbow/actions/runs/7543210134/job/20535105142#step:1:7).1
  Included Software: https://github.com/actions/runner-images/blob/macOS-12/20231216.1/images/macos/macos-12-Readme.md
  Image Release: https://github.com/actions/runner-images/releases/tag/macOS-12%2F20231216.1

Error related to undefined symbol: _ZTIN6google8protobuf7MessageE also appear with nightly versions:

Error with Dataset 15.0.0:
- git clone https://github.com/davisusanibar/dataset-error-linux-env.git
- mvn compile exec:java -Dexec.mainClass="dataset.TestMe" -Dmaven.compiler.target=8 -Dmaven.compiler.source=8


Error with Dataset 16.0.0-SNAPSHOT:
- git clone https://github.com/davisusanibar/dataset-error-linux-env.git
- git checkout test-with-nightly-version
- mvn compile exec:java -Dexec.mainClass="dataset.TestMe" -Dmaven.compiler.target=8 -Dmaven.compiler.source=8

[davisusanibar]

I'm not sure we can yank a maven package.
Why isn't this caught in the java-jars pipeline?

Java-jars is running unders macOS environment, then all the validation finished without problem on macOS: https://github.com/ursacomputing/crossbow/actions/runs/7543210134/job/20535105142

Operating System
  macOS
  12.7.2
  21G1974
Runner Image
  Image: macos-12
  Version: 202[3](https://github.com/ursacomputing/crossbow/actions/runs/7543210134/job/20535105142#step:1:3)121[6](https://github.com/ursacomputing/crossbow/actions/runs/7543210134/job/20535105142#step:1:7).1
  Included Software: https://github.com/actions/runner-images/blob/macOS-12/20231216.1/images/macos/macos-12-Readme.md
  Image Release: https://github.com/actions/runner-images/releases/tag/macOS-12%2F20231216.1

Error related to undefined symbol: _ZTIN6google8protobuf7MessageE also appear with nightly versions:

Error with Dataset 15.0.0:
- git clone https://github.com/davisusanibar/dataset-error-linux-env.git
- mvn compile exec:java -Dexec.mainClass="dataset.TestMe" -Dmaven.compiler.target=8 -Dmaven.compiler.source=8


Error with Dataset 16.0.0-SNAPSHOT:
- git clone https://github.com/davisusanibar/dataset-error-linux-env.git
- git checkout test-with-nightly-version
- mvn compile exec:java -Dexec.mainClass="dataset.TestMe" -Dmaven.compiler.target=8 -Dmaven.compiler.source=8

Reviewing crossbow http://crossbow.voltrondata.com/ JNI libraries uploaded to Github Assets, we found that 2023-12-09 was the first time we saw U=Undefined on the libarrow_dataset_jni.so library.

2023-12-08: OK - https://github.com/ursacomputing/crossbow/actions/runs/7139118592
nm libarrow_dataset_jni.so | grep ZTIN6google8protobuf7Messag
0000000002f5dd78 V _ZTIN6google8protobuf7MessageE

2023-12-09: NOK --> First undefined appear U - https://github.com/ursacomputing/crossbow/actions/runs/7150103719
nm libarrow_dataset_jni.so | grep ZTIN6google8protobuf7Messag
                 U _ZTIN6google8protobuf7MessageE