Extracting information from a a java core dump with jmap(1.5)
Asked Answered
E

3

10

Long story short, some coworkers are running a pretty old setup(oc4j jdk1.5.6 in x86_64) with an application which happens to be mission critical. They recently have tried to deploy a new version of the application, but as soon as they do the java process(es) throw a core dump and die.

The problem is, the core dumps seem to be fine, gdb can open them, but jmap and other tools refuse to process them:

# /usr/java/jdk1.5.0_06/bin/jmap /usr/java/jdk1.5.0_06/bin/java core
Attaching to core core from executable /usr/java/jdk1.5.0_06/bin/java, please wait...
Error attaching to core file: Can't attach to the core file

And newer versions throw a exception:

# jdk1.6.0_45/bin/jmap /usr/java/jdk1.5.0_06/bin/java core
Attaching to core core from executable /usr/java/jdk1.5.0_06/bin/java, please wait...
Exception in thread "main" java.lang.reflect.InvocationTargetException
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:39)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
        at java.lang.reflect.Method.invoke(Method.java:597)
        at sun.tools.jmap.JMap.runTool(JMap.java:179)
        at sun.tools.jmap.JMap.main(JMap.java:110)
Caused by: sun.jvm.hotspot.runtime.VMVersionMismatchException: Supported versions are 20.45-b01. Target VM is 1.5.0_06-b05
        at sun.jvm.hotspot.runtime.VM.checkVMVersion(VM.java:224)
        at sun.jvm.hotspot.runtime.VM.<init>(VM.java:287)
        at sun.jvm.hotspot.runtime.VM.initialize(VM.java:357)
        at sun.jvm.hotspot.bugspot.BugSpotAgent.setupVM(BugSpotAgent.java:594)
        at sun.jvm.hotspot.bugspot.BugSpotAgent.go(BugSpotAgent.java:494)
        at sun.jvm.hotspot.bugspot.BugSpotAgent.attach(BugSpotAgent.java:348)
        at sun.jvm.hotspot.tools.Tool.start(Tool.java:169)
        at sun.jvm.hotspot.tools.PMap.main(PMap.java:67)
        ... 6 more

gdb offers little information without symbols:

Reading symbols from /usr/java/jdk1.5.0_06/bin/java...(no debugging symbols found)...done.

[New Thread 9841]
[New Thread 31442]
[New Thread 31441]
...
Core was generated by `/usr/java/jdk1.5.0_06/bin/java -server -XX:+UseConcMarkSweepGC -XX:MaxHeapFreeR'.
Program terminated with signal 6, Aborted.
#0  0x0000003bbf030285 in ?? ()
(gdb) bt
#0  0x0000003bbf030285 in ?? ()
#1  0x0000003bbf031d30 in ?? ()
#2  0x0000000000000000 in ?? ()

The only valuable information I've gathered from the core is that most threads are blocked(I'm far from being a gdb guru):

  35 Thread 10093  0x0000003bbfc0b1c0 in pthread_cond_timedwait@@GLIBC_2.3.2 ()
   from /lib64/libpthread.so.0
  34 Thread 10097  0x0000003bbfc0b1c0 in pthread_cond_timedwait@@GLIBC_2.3.2 ()
   from /lib64/libpthread.so.0
  33 Thread 10099  0x0000003bbfc0b1c0 in pthread_cond_timedwait@@GLIBC_2.3.2 ()
   from /lib64/libpthread.so.0

Besides, I don't know if it's really relevant. The app is almost always heavily loaded, and my bet is that there were some lock contention already but since it's another's team app my knowledge about it it's pretty shallow.

I guess this is a long shot, but is there something that we can do to get a java thread dump or something like that? Do Sun used to offer debuginfo of the jdk as I guess is avalaible now with openjdk?

Thanks in advance.

UPDATE: The other team has resolved the issue without getting info from the core dump, just by trial and error after successfully replicating the problem in a test system. I'm still intrigued about the thing: how to debug an ancient java core dump which jmap can't process, it might be valuable info for the future, althought it seems is that there is no solution to that problem. Probably the JVM memory got corrupted and that's why jmap can't process it.

Ermey answered 25/3, 2017 at 9:5 Comment(8)
Have you tried upgrading the jvm to the latest 1.5 jvm ?Doubt
I have tried to use the jmap of the last 1.5 version without better results("Can't attach to the core file"). I can't decide about upgrading the jdk version in the production server since it's not my team's responsability(still!).Ermey
is it possible to update the JDK that you use to deploy the application? the latest version seems to be 1.5.0_22 - also worth a look: oracle.com/technetwork/java/javase/releasenotes-142123.html ; there were quite a few updates to the core dump between Update 05 and 22Picardi
Just a long shot but can you process induced core dumps? If yes then JVM memory is likely corrupted before it crashes and thats why jmap can't read it. See this link for how to induce a core dump palmr.co.uk/blog/13/JVM+Core+Dumping andEstimation
Thanks to both for the suggestions, both make a lot sense.Ermey
Try github.com/yanqi27/core_analyzer Guide how to use core-analyzer.sourceforge.net/index_files/Page1510.htmlHatfield
As far as I know, there was no jmap in JDK 1.5 so how did you get one for jdk 1.5.0_6?Adversaria
There was in Linux, may be not in other platforms.Ermey
L
5

You can add the following JVM option when starting your application, that will allow you to run any command you specify if a fatal JVM error occurs:

-XX:OnError="<cmd args>"

For instance, you could run a command (or a script) that will perform certain actions like get a heap or thread dump.

Laughlin answered 29/3, 2017 at 21:43 Comment(0)
I
3

Jmap and other JVM utilities are extremely version sensitive. From your error, it is self explanatory that hopefully the same jvm is not used in your case.

Java VisualVM can load core dumps directly. But you must use the same jvm that created the core file.

Resource Link:

https://mcmap.net/q/644439/-core-dump-taken-with-gcore-jmap-conversion-to-hprof-file-format-fails-with-error-message

Suggestion#1:

kjkoster has given a solution here in this tutorial.

You need to use the jmap that comes with the JVM. From your error message I gather that you are using a different version of jmap than of the JVM.

Please check what JVMs are installed on your machine and ensure than when you run jmap, you use the right version.

To solve such issues I never rely on path. Instead I set JAVA_HOME to be the one that the JVM uses and then invoke both the JVM and jmap like so:

Code:

$ JAVA_HOME=/usr/local/jdk1.6.0
$ export JAVA_HOME
$ ${JAVA_HOME}/bin/java ...

...

$ ${JAVA_HOME}/bin/jmap ...

Hope this helps.


Suggestion#2:

It is a full solution step by step is given by Chamilad. Hope it will clarify your root cause and solution procedure.

Almost every Java developer knows about jmap and jstack tools that come with the JDK. These provide functionality to extract heap and thread information of a running JVM instance. Easy.

What if there’s a running JVM that has produced a deadlock and you want to take a thread dump while the process is running? You go in and run the following.

jstack pid >> thread_dump.txt

Turns out the system doesn’t know what jstack is. You don’t panic, but you get a tiny sensation at the back of your head saying you’re not leaving early this Friday.

What has happened is the running JVM is based on a JRE and not a JDK. The JRE is a minimal runtime that doesn’t pack the monitoring and analysis tools that the JDK packs.

So what are our options here?

  1. Stop the process. Download JDK, start the process again on top of JDK and hope the deadlock happens again. Nope.
  2. Start JVisualVM on your laptop and hope the process has JMX enabled. Nope.
  3. tools.jar TO THE RESCUE!

Functionalities such as jstack are implemented in the tools.jar file which is packed inside <JDK_HOME>/lib folder. We can use this to invoke the JStack class and get a thread dump of the running process.

So we march on to download and extract the JDK, and then to run the following.

java -classpath <JDK_HOME>/lib/tools.jar sun.tools.jstack.JStack <pid> >> thread_dump.txt

..and come across the following error.

Exception in thread "main" java.lang.UnsatisfiedLinkError: no attach in java.library.path
        at java.lang.ClassLoader.loadLibrary(ClassLoader.java:1867)
        at java.lang.Runtime.loadLibrary0(Runtime.java:870)
        at java.lang.System.loadLibrary(System.java:1122)
        at sun.tools.attach.LinuxVirtualMachine.<clinit>(LinuxVirtualMachine.java:342)
        at sun.tools.attach.LinuxAttachProvider.attachVirtualMachine(LinuxAttachProvider.java:63)
        at com.sun.tools.attach.VirtualMachine.attach(VirtualMachine.java:208)
        at sun.tools.jstack.JStack.runThreadDump(JStack.java:163)
        at sun.tools.jstack.JStack.main(JStack.java:116)

Darn it! Spoiled again!

How do we solve this? The above error is caused when the process can’t find libattach.so file which is related to the Dynamic Attach function related to JStack. Setting the following environment variable will help the JVM to find the libattach.so file.

export LD_LIBRARY_PATH=<JDK_HOME>/jre/lib/amd64/

Now let’s run JStack again, this time with results!

java -classpath <JDK_HOME>/lib/tools.jar sun.tools.jstack.JStack <pid> >> thread_dump.txt

Now that we have the thread dump, we move on to the heap dump. The tool we normally use is jmap but that too is not available on the JRE. So what? We can use the binary in the JDK’s bin directory right? right?

root@snowflake1 latest]# <JDK_HOME>/bin/jmap -heap <pid>
Attaching to process ID <pid>, please wait…
Error attaching to process: sun.jvm.hotspot.runtime.VMVersionMismatchException: Supported versions are 25.102-b14. Target VM is 25.91-b14
sun.jvm.hotspot.debugger.DebuggerException: sun.jvm.hotspot.runtime.VMVersionMismatchException: Supported versions are 25.102-b14. Target VM is 25.91-b14
 at sun.jvm.hotspot.HotSpotAgent.setupVM(HotSpotAgent.java:435)
 at sun.jvm.hotspot.HotSpotAgent.go(HotSpotAgent.java:305)
 at sun.jvm.hotspot.HotSpotAgent.attach(HotSpotAgent.java:140)
 at sun.jvm.hotspot.tools.Tool.start(Tool.java:185)
 at sun.jvm.hotspot.tools.Tool.execute(Tool.java:118)
 at sun.jvm.hotspot.tools.HeapSummary.main(HeapSummary.java:49)
 at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
 at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
 at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
 at java.lang.reflect.Method.invoke(Method.java:498)
 at sun.tools.jmap.JMap.runTool(JMap.java:201)
 at sun.tools.jmap.JMap.main(JMap.java:130)
Caused by: sun.jvm.hotspot.runtime.VMVersionMismatchException: Supported versions are 25.102-b14. Target VM is 25.91-b14
 at sun.jvm.hotspot.runtime.VM.checkVMVersion(VM.java:227)
 at sun.jvm.hotspot.runtime.VM.<init>(VM.java:294)
 at sun.jvm.hotspot.runtime.VM.initialize(VM.java:370)
 at sun.jvm.hotspot.HotSpotAgent.setupVM(HotSpotAgent.java:431)
 … 11 more

Nope! Unless you match the JDK version with the exact version the JRE is at you get the above issue (which is pretty self-explanatory). So we download the JDK of the JRE of on top our process is running and run jmap again.

<JDK_HOME>/bin/jmap -dump:file=heap_dump.hprof <pid>

Resource Link:

  1. Extracting memory and thread dumps from a running JRE based JVM
Irreversible answered 29/3, 2017 at 14:27 Comment(0)
P
2

jmap is not going to help you debug a core dump. The JVM dumps core when either it has a bug or you have JNI code with a problem. Organizations with mission-critical applications should, sadly, see upgrading from unsupported versions of the JVM as mission-critical, or be prepared to pay Oracle a fortune for help.

Pneumonectomy answered 29/3, 2017 at 21:46 Comment(1)
"Organizations with mission-critical applications should, sadly, see upgrading from unsupported versions of the JVM as mission-critical, or be prepared to pay Oracle a fortune for help." You are completely right in this. As I said, I'm more intrigued for the technical challenge than worried.Ermey

© 2022 - 2024 — McMap. All rights reserved.