Re: PING: RFR: 8165736: Error message should be shown when JVMTI agent cannot be attached

[serguei.spit...@oracle.com]

Hi Yasumasa, Thank you for the update. Some comments. http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.04/src/jdk.attach/share/classes/sun/tools/attach/HotSpotVirtualMachine.java.frames.html 94 BufferedReader reader = new BufferedReader(new InputStreamReader(in)); 95 try (reader) { A classic way for the above is: try (BufferedReader reader = new BufferedReader(new InputStreamReader(in))) { Also, I have a doubt about the change in HotSpotVirtualMachine.java: private void loadAgentLibrary(String agentLibrary, boolean isAbsolute, String options) throws AgentLoadException, AgentInitializationException, IOException { + String msgPrefix = "return code: "; InputStream in = execute("load", agentLibrary, isAbsolute ? "true" : "false", options); - try { - int result = readInt(in); - if (result != 0) { - throw new AgentInitializationException("Agent_OnAttach failed", result); + BufferedReader reader = new BufferedReader(new InputStreamReader(in)); + try (reader) { + String result = reader.readLine(); + if (result == null) { + throw new AgentLoadException("Target VM did not respond"); + } else if (result.startsWith(msgPrefix)) { + int retCode = Integer.parseInt(result.substring(msgPrefix.length())); + if (retCode != 0) { + throw new AgentInitializationException("Agent_OnAttach failed", retCode); + } + } else { + throw new AgentLoadException(result); } - } finally { - in.close(); - } } Now the AgentLoadException is thrown where it was not before. So that there is a change in behavior. On the other hand the AgentLoadException was already specified to be thrown by the loadAgentLibrary. I wonder, if this is worth to file a CSR for this. Thanks, Serguei On 11/19/17 23:54, Yasumasa Suenaga wrote: Hi David, My own feeling is that it is up to the OnAttach function to properly deal with pending exceptions: report and/or clear them. The VM side just has to clear any pending exception to avoid it causing problems for later code. I removed the change to print pending exceptions in new webrev:   http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.04/   test/jdk/com/sun/tools/attach/StartManagementAgent.java The reporting of NumberFormatException may be accurate in terms of the low-level exception, but "Invalid com.sun.management.jmxremote.port number" was much clearer. This makes me wonder about whether the code that previously produced "Invalid com.sun.management.jmxremote.port number" needs updating if this change proceeds. (And alao makes me wonder about the impact of the change in general.) I tested StartManagementAgent.java without this change, and I got failure as below: -------------- JavaTest Message: Test threw exception: com.sun.tools.attach.AttachOperationFailedE xception: java.lang.RuntimeException: jdk.internal.agent.AgentConfigurationError: j ava.lang.NumberFormatException: For input string: "apa" JavaTest Message: shutting down test STATUS:Failed.`main' threw exception: com.sun.tools.attach.AttachOperationFailedExc eption: java.lang.RuntimeException: jdk.internal.agent.AgentConfigurationError: jav a.lang.NumberFormatException: For input string: "apa" -------------- Should we change this testcase whatever this change is not accepted? Thanks, Yasumasa On 2017/11/20 6:41, David Holmes wrote: Hi Yasumasa, I've been trying to leave these reviews to serviceability folk ... I've gone back through the original RFR from September last year to see what we did and what was left. The current proposal raises some concern for me - and IIRC Dmitry was also concerned about it last time: printing of the pending exception. If we print the pending exception we will report an error and throw AgentLoadException, even if execution of the OnAttach function returned JNI_OK. If that exception was not critical to the success of the loading the agent, and the agent was just sloppy about clearing it, then it will now fail to load - which would be a compatibility concern. Further, if the exception indicates an error and the OnAttach function returns JNI_ERR then we won't report that cleanly because the printing of the exception will prevent matching with "return code: -1". My own feeling is that it is up to the OnAttach function to properly deal with pending exceptions: report and/or clear them. The VM side just has to clear any pending exception to avoid it causing problems for later code. Some specific comments: HotSpotVirtualMachine.java The regex code seems overkill for the basic parsing you are doing. You just need to see if the strings starts with "return code: " and then parse the next bit as an integer to get the return code. ---   test/jdk/com/sun/tools/attach/StartManagementAgent.java The reporting of NumberFormatException may be accurate in terms of the low-level exception, but "Invalid com.sun.management.jmxremote.port number" was much clearer. This makes me wonder about whether the code that previously produced "Invalid com.sun.management.jmxremote.port number" needs updating if this change proceeds. (And alao makes me wonder about the impact of the change in general.) --- Sorry - not the quick second review you were looking for. David ----- On 19/11/2017 11:38 PM, Yasumasa Suenaga wrote: PING: Could you review it?    http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.03/ I want to merge this change to jdk 10. So I need a second reviewer. Yasumasa On 2017/11/16 21:09, Yasumasa Suenaga wrote: Hi David, Serguei, The test logic is adding it in AttachFailedTestBase.java:   45         return Paths.get(System.getProperty("test.nativepath"), "lib", libname)   46                     .toAbsolutePath()   47                     .toString(); Thanks! I've fixed it in new webrev:    http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.03/ I've tested it as below. It works fine: $ $JT_HOME/bin/jtreg -ignore:quiet -nativepath:$NATIVE_PATH hotspot/jtreg/serviceability/dcmd/jvmti/AttachFailed $ echo $NATIVE_PATH /<Path to configuration>/images/test/hotspot/jtreg/native Thanks, Yasumasa On 2017/11/16 16:49, serguei.spit...@oracle.com wrote: On 11/15/17 23:29, David Holmes wrote: On 16/11/2017 4:43 PM, serguei.spit...@oracle.com wrote: On 11/15/17 18:11, David Holmes wrote: Hi Serguei, > /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg/native/lib/libException.so' There should not be any "/lib/" in that path Right, it should not be. The test logic is adding it in AttachFailedTestBase.java:   45         return Paths.get(System.getProperty("test.nativepath"), "lib", libname)   46                     .toAbsolutePath()   47                     .toString(); but it shouldn't. Nice catch! I looked right to these lines and overlooked it. :) Thanks, Serguei David ----- This is the script I'm using to run the tests: #!/bin/sh REPO=/var/tmp/sspitsyn/jdk.attach IMAGES=${REPO}/build/linux-x86_64-normal-server-release/images export JAVA_HOME=${IMAGES}/jdk export NATIVE_PATH=${IMAGES}/../support/test/hotspot/jtreg/native export NATIVE_PATH=${IMAGES}/test/hotspot/jtreg/native echo "JAVA_HOME = $JAVA_HOME" /java/re/jtreg/4.2/nightly/binaries/jtreg/bin/jtreg -nativepath:${NATIVE_PATH} \ $REPO/open/test/hotspot/jtreg/serviceability/dcmd/jvmti/AttachFailed This is a part of log with the reported error from the AttachException.jtr: [TestNG] Running:    serviceability/dcmd/jvmti/AttachFailed/AttachException.java Running DCMD 'JVMTI.agent_load /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg/native/lib/libException.so' through 'PidJcmdExecutor' Executing command '[/var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/jdk/bin/jcmd, 8689, JVMTI.agent_load /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg/native/lib/libException.so]' Command returned with exit code 0 ---------------- stdout ---------------- 8689: /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg/native*/lib*/libException.so was not loaded. /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg/native*/lib*/libException.so: cannot open shared object file: No such file or directory These are the locations of the libException.so: build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg/native/libException.so build/linux-x86_64-normal-server-release/support/test/hotspot/jtreg/native/lib/libException.so The tests fail with the "NATIVE_PATH=${IMAGES}/test/hotspot/jtreg/native" but pass with the "export NATIVE_PATH=${IMAGES}/../support/test/hotspot/jtreg/native". When the "export NATIVE_PATH=${IMAGES}/../support/test/hotspot/jtreg/native" is used the log has this line: Running DCMD 'JVMTI.agent_load /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/../support/test/hotspot/jtreg/native*/lib*/libException.so' through 'JMXExecutor' Apparently, the sub-directory name "/lib" is added to the path. Thanks, Serguei David On 16/11/2017 4:34 AM, serguei.spit...@oracle.com wrote: Hi Yasumasa and David, On 11/15/17 04:56, David Holmes wrote: On 15/11/2017 10:15 PM, Yasumasa Suenaga wrote: Hi Serguei, Do the new tests pass in your runs? Of course. It seems not to exist jtreg native libraries. I've tested as below:    $ make build-test-hotspot-jtreg-native    $ cd test    $ $JT_HOME/bin/jtreg -ignore:quiet -nativepath:<builddir>/<confdir>/support/test/hotspot/jtreg/native/lib hotspot/jtreg/serviceability/attach hotspot/jtreg/serviceability/dcmd/jvmti jdk/com/sun/tools/attach Thanks. I missed to add the -nativepath flag, sorry. Please check that: make test-image followed by jtreg -nativepath:<build-dir>/images/test/hotspot/jtreg/native also works. It fails with the error:    63 Running DCMD 'JVMTI.agent_load /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg/native/lib/libException.so' through 'PidJcmdExecutor'    64 Executing command '[/var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/jdk/bin/jcmd, 28407, JVMTI.agent_load /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg /native/lib/libException.so]'    65 Command returned with exit code 0    66 ---------------- stdout ----------------    67 28407:    68 /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg/native/lib/libException.so was not loaded.    69 /var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/test/hotspot/jtreg/native/lib/libException.so: cannot open shared object file: No such file or directory    70 It seems, the '/lib' folder is added to the nativepath. Yasumasa, could you, double check it please? I'm using the jtreg: /java/re/jtreg/4.2/promoted/latest/binaries/jtreg/bin/jtreg which is: % ls -l /java/re/jtreg/4.2/promoted/latest lrwxrwxrwx 1 uucp 143 7 Nov  6 21:49 /java/re/jtreg/4.2/promoted/latest -> fcs/b10/ Thanks, Serguei Thanks, David Good news is that the attach-related tests from closed repository are passed. Thanks! Yasumasa On 2017/11/15 16:38, serguei.spit...@oracle.com wrote: Hi Yasumasa, Do the new tests pass in your runs? In my runs 3 of 4 tests are failed with the errors like this:   109 Running DCMD 'JVMTI.agent_load /var/tmp/sspitsyn/tst/jdk.attach/JTwork/scratch/null/lib/libException.so' through 'PidJcmdExecutor'   110 Executing command '[/var/tmp/sspitsyn/jdk.attach/build/linux-x86_64-normal-server-release/images/jdk/bin/jcmd, 21951, JVMTI.agent_load /var/tmp/sspitsyn/tst/jdk.attach/JTwork/scratch/null/lib/libException.so]'   111 Command returned with exit code 0   112 ---------------- stdout ----------------   113 21951:   114 /var/tmp/sspitsyn/tst/jdk.attach/JTwork/scratch/null/lib/libException.so was not loaded.   115 /var/tmp/sspitsyn/tst/jdk.attach/JTwork/scratch/null/lib/libException.so: cannot open shared object file: No such file or directory   116 Good news is that the attach-related tests from closed repository are passed. Thanks, Serguei On 11/14/17 16:40, serguei.spit...@oracle.com wrote: Hi Yasumasa, It looks good to me. Thanks, Serguei On 11/7/17 22:38, Yasumasa Suenaga wrote: Hi Serguei, I uploaded new webrev: http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.02/ I'd expect a check for some exception name, not for details like: For input string: "apa". Should we remove this comparison? I don't understand. Why do remove? Would it better to check for the exception name instead? I've changed to check exception name (NumberFormatException) in StartManagementAgent.java. I will sponsor this fix and run these tests before the push. Thanks! I'm waiting for second reviewer. Yasumasa 2017-11-08 11:55 GMT+09:00 serguei.spit...@oracle.com <serguei.spit...@oracle.com>: On 11/6/17 04:31, Yasumasa Suenaga wrote: Hi Serguei, On 2017/11/06 20:06, serguei.spit...@oracle.com wrote: Hi Yasumasa, The changes looks good. Thank you for making them! Thanks! On 11/3/17 05:10, Yasumasa Suenaga wrote: Hi Serguei, Thank you for your comment! I uploaded new webrev: http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.01/ http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.00/test/jdk/com/sun/tools/attach/StartManagementAgent.java.udiff.html - if (!ex.getMessage().contains("Invalid com.sun.management.jmxremote.port number")) { + if (!ex.getMessage().contains("For input string: \"apa\"")) {     What is the motivation for this change?     It seems, the original comparison is better. "ex" is AttachOperationFailedException. We can get the result as below when we run StartManagementAgent: ------------- [runApplication] Error: Invalid com.sun.management.jmxremote.port number: apa [runApplication] jdk.internal.agent.AgentConfigurationError: java.lang.NumberFormatException: For input string: "apa" [runApplication]        at jdk.management.agent/sun.management.jmxremote.ConnectorBootstrap.startRemoteConnectorServer(ConnectorBootstrap.java:336) ------------- I think we should check exception message in AttachOperationFailedException. In fact, jtreg fails at this point in my environment. I'd expect a check for some exception name, not for details like: For input string: "apa". Should we remove this comparison? I don't understand. Why do remove? Would it better to check for the exception name instead? What tests did you run to make sure there are no regressions? I've tested the following testcases:    - hotspot/jtreg/serviceability/attach    - hotspot/jtreg/serviceability/dcmd/jvmti    - jdk/com/sun/tools/attach There are more tests related to dynamic attach in closed, nsk.aod.testlist and 30+ attach tests in nsk.jvmti.testlist. I'm not sure, if they are included into any of the Mach5 testing levels. Will need to check. We have to make sure these tests are still passed. I cannot access JPRT and closed testcases because I'm not an Oracle employee. Can you run them with this change? Ok. I will sponsor this fix and run these tests before the push. It seems, another update and one more review is needed. Thanks, Serguei Thanks, Yasumasa Thanks, Serguei Thanks, Yasumasa On 2017/11/03 16:31, serguei.spit...@oracle.com wrote: Hi Yasumasa, Some comments. http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.00/test/jdk/com/sun/tools/attach/StartManagementAgent.java.udiff.html - if (!ex.getMessage().contains("Invalid com.sun.management.jmxremote.port number")) { + if (!ex.getMessage().contains("For input string: \"apa\"")) {     What is the motivation for this change?     It seems, the original comparison is better. http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.00/test/hotspot/jtreg/serviceability/dcmd/jvmti/AttachFailed/AttachException.java.html http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.00/test/hotspot/jtreg/serviceability/dcmd/jvmti/AttachFailed/AttachIncorrectLibrary.java.html http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.00/test/hotspot/jtreg/serviceability/dcmd/jvmti/AttachFailed/AttachNoEntry.java.html http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.00/test/hotspot/jtreg/serviceability/dcmd/jvmti/AttachFailed/AttachReturnError.java.html     37     public void run(CommandExecutor executor) {     38         try{     A space is missed after 'try'.     It is odd that all test java classes define exactly the same methods: sharedObjectName(), jmx() and cli().     Would it better to defin a common base class with these methods? Otherwise, it looks good. Thank you for taking care about it! What tests did you run to make sure there are no regressions? Thanks, Serguei On 11/1/17 05:59, Yasumasa Suenaga wrote: PING: Could you review and sponsor it? http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.00/ Thanks, Yasumasa On 2017/09/29 13:24, Yasumasa Suenaga wrote: Hi all, If we try to attach invalid JVMTI agent via JVMTI.agent_load dcmd, we will get "Command executed successfully". However, it implies error in JVMTIAgentLoadDCmd. This message is from JCmd.java when jcmd does not receive any output from target VM. I think HotSopt/jcmd should return useful error message to users to understand the cause of failure. I uploaded webrev for this issue. Could you review it? http://cr.openjdk.java.net/~ysuenaga/JDK-8165736/webrev.00/ This change is work fine on Fedora 26 x86_64 as following jtreg testcases:     - hotspot/jtreg/serviceability/attach     - hotspot/jtreg/serviceability/dcmd/jvmti     - jdk/com/sun/tools/attach I cannot access JPRT. So I need a sponsor. (I cannot test it on other platforms.) Thanks, Yasumasa