Rasmus Thygesen created FLINK-35358:
---------------------------------------
Summary: Breaking change when loading artifacts
Key: FLINK-35358
URL: https://issues.apache.org/jira/browse/FLINK-35358
Project: Flink
Issue Type: New Feature
Components: Client / Job Submission, flink-docker
Affects Versions: 1.19.0
Reporter: Rasmus Thygesen
We have been using the following code snippet in our Dockerfiles for running a
Flink job in application mode
{code:java}
FROM flink:1.18.1-scala_2.12-java17
COPY --from=build /app/target/my-job*.jar /opt/flink/usrlib/artifacts/my-job.jar
USER flink {code}
Which has been working since at least around Flink 1.14, but the 1.19 update
has broken our Dockerfiles. The fix is to put the jar file a step further out
so the code snippet becomes
{code:java}
FROM flink:1.18.1-scala_2.12-java17
COPY --from=build /app/target/my-job*.jar /opt/flink/usrlib/my-job.jar
USER flink {code}
We have not spent too much time looking into what the cause is, but we get the
stack trace
{code:java}
myjob-jobmanager-1 | org.apache.flink.util.FlinkException: Could not load the
provided entrypoint class.myjob-jobmanager-1 | at
org.apache.flink.client.program.DefaultPackagedProgramRetriever.getPackagedProgram(DefaultPackagedProgramRetriever.java:230)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.container.entrypoint.StandaloneApplicationClusterEntryPoint.getPackagedProgram(StandaloneApplicationClusterEntryPoint.java:149)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.container.entrypoint.StandaloneApplicationClusterEntryPoint.lambda$main$0(StandaloneApplicationClusterEntryPoint.java:90)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.runtime.security.contexts.NoOpSecurityContext.runSecured(NoOpSecurityContext.java:28)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.container.entrypoint.StandaloneApplicationClusterEntryPoint.main(StandaloneApplicationClusterEntryPoint.java:89)
[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | Caused by:
org.apache.flink.client.program.ProgramInvocationException: The program's entry
point class 'my.company.job.MyJob' was not found in the jar
file.myjob-jobmanager-1 | at
org.apache.flink.client.program.PackagedProgram.loadMainClass(PackagedProgram.java:481)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.client.program.PackagedProgram.<init>(PackagedProgram.java:153)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.client.program.PackagedProgram.<init>(PackagedProgram.java:65)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.client.program.PackagedProgram$Builder.build(PackagedProgram.java:691)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.client.program.DefaultPackagedProgramRetriever.getPackagedProgram(DefaultPackagedProgramRetriever.java:228)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | ... 4
moremyjob-jobmanager-1 | Caused by: java.lang.ClassNotFoundException:
my.company.job.MyJobmyjob-jobmanager-1 | at
java.net.URLClassLoader.findClass(Unknown Source) ~[?:?]myjob-jobmanager-1 |
at java.lang.ClassLoader.loadClass(Unknown Source) ~[?:?]myjob-jobmanager-1
| at
org.apache.flink.util.FlinkUserCodeClassLoader.loadClassWithoutExceptionHandling(FlinkUserCodeClassLoader.java:67)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.util.ChildFirstClassLoader.loadClassWithoutExceptionHandling(ChildFirstClassLoader.java:74)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.util.FlinkUserCodeClassLoader.loadClass(FlinkUserCodeClassLoader.java:51)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
java.lang.ClassLoader.loadClass(Unknown Source) ~[?:?]myjob-jobmanager-1 |
at
org.apache.flink.util.FlinkUserCodeClassLoaders$SafetyNetWrapperClassLoader.loadClass(FlinkUserCodeClassLoaders.java:197)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
java.lang.Class.forName0(Native Method) ~[?:?]myjob-jobmanager-1 | at
java.lang.Class.forName(Unknown Source) ~[?:?]myjob-jobmanager-1 | at
org.apache.flink.client.program.PackagedProgram.loadMainClass(PackagedProgram.java:479)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.client.program.PackagedProgram.<init>(PackagedProgram.java:153)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.client.program.PackagedProgram.<init>(PackagedProgram.java:65)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.client.program.PackagedProgram$Builder.build(PackagedProgram.java:691)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | at
org.apache.flink.client.program.DefaultPackagedProgramRetriever.getPackagedProgram(DefaultPackagedProgramRetriever.java:228)
~[flink-dist-1.19.0.jar:1.19.0]myjob-jobmanager-1 | ... 4 more{code}
I have changed some text in the stack trace to keep it anonymous so it is
possible there is a typo but that is not the issue. As you can see, the stack
trace leads to PackagedProgram and DefaultPackagedProgramRetriever to which the
only commits after Flink 1.18 are [PackagedProgram
commit|https://github.com/apache/flink/commit/d0ce5349fdf1a611518eba20a169c475ee0b46c5]
and [DefaultPackagedProgramRetriever
commit|https://github.com/apache/flink/commit/e63aa12252843d0098a56f3091b28d48aff5b5af]
and we suspect the culprit is the latter, specifically [this
line|https://github.com/apache/flink/commit/e63aa12252843d0098a56f3091b28d48aff5b5af#diff-11b5162d6745014c68e96303d26c71bdb88bac068c27834dbdbb7c9089ffbe9fL227]
which we think has made the artifact check non-recursive. We assume it is
intended to have your artifacts directly in /opt/flink/usrlib without the
artifacts directory so we are planning on changing that for our Dockerfiles
anyway, but it is still a breaking change so we wanted to make an issue on it
first.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)