Bug 871606

+++ This bug was initially created as a clone of Bug #849394 +++

Description of problem:

I configured the 'script' plugin to watch for output when checking for availability of my resource.

I check the results using a regex check. In some cases, the output is empty. Then RHQ assumes the resource is down, when in fact it is up.

For example, /sbin/service is called with the name of my service. In some cases no output appears.

2012-08-18 21:09:38,052 DEBUG [ResourceContainer.invoker.daemon-115] (org.rhq.plugins.script.ScriptServerComponent)- [/sbin/service]: CLI results: exitcode=[0]; error=[null]; output=
2012-08-18 21:09:38,052 DEBUG [ResourceContainer.invoker.daemon-115] (org.rhq.plugins.script.ScriptServerComponent)- [/sbin/service]: CLI output [] did not match regex [(?s).*is running.*], resource is considered DOWN

Version-Release number of selected component (if applicable):

RHQ 4.2 plugin

How reproducible:

Eventually is.

Additional info:

modules/core/util/src/main/java/org/rhq/core/util/exec/ProcessExecutor.java contains this unfortunate note:

        // WARNING:
        // It seems there is no way to get around a possible race condition - what if the process
        // was so fast that it exited already?  We didn't get a chance to capture its output.
        // I see a unit test that periodically fails because it doesn't get any captured output when
        // it should - I think it is because of this race condition.  But there is no Java API that
        // let's me redirect a process' streams before the process is told to start.

It's unclear if this is really still the case or not...

http://stackoverflow.com/questions/1670798/using-a-thread-to-capture-process-output

There is a note that I/O needs to be consumed in a thread. It could be the process completes but the I/O was never consumed.

--- Additional comment from genman on 2012-08-19 02:15:08 EDT ---


The problem is that the I/O threads are never run to completion: The main process waits for the exit status, but the I/O threads may not have starts.

I will provide a patch, but the solution is simply to return a handle containing both stdout/stderr Thread instances and wait for these threads...

modules/core/util/src/main/java/org/rhq/core/util/exec/ProcessExecutor.java

    /**
     * Wrapper for threads used for capturing output.
     * Call {@link #join} to wait for output to be fully captured.
     */
    protected static class RedirectThreads {

        private final StreamRedirector stdout;
        private final StreamRedirector stderr;

...

    protected RedirectThreads redirectStreams(ProcessToStart process, Process childProcess) throws IOException {
...
        return new RedirectThreads(stdoutThread, stderrThread);

then ... 

+        final RedirectThreads redirect = redirectStreams(process, childProcess);
...
-                        int exitCode = childProcess.waitFor();
-                        retExitCode[0] = new Integer(exitCode);
+                    } finally {
+                        redirect.join();
                     }

Simple as this.

--- Additional comment from genman on 2012-09-13 18:02:49 EDT ---

Created attachment 612648 [details]
fixes process leak

--- Additional comment from mazz on 2012-09-18 15:19:42 EDT ---

(In reply to comment #2)
> Created attachment 612648 [details]
> fixes process leak

I think the wrong patch was attached - the attached patch deals with PIQL queries. Can you remove that patch and attach the one for this issue?

--- Additional comment from genman on 2012-09-18 17:07:02 EDT ---

Created attachment 614123 [details]
Ensure output thread completes before exiting startProgram

Sorry I had the wrong patch.

One thing you might want to adjust (for sake of the build) is adjust the number of times we check for correct output:

modules/core/util/src/test/java/org/rhq/core/util/exec/ProcessExecTest.java

diff --git a/modules/core/util/src/test/java/org/rhq/core/util/exec/ProcessExecTest.java b/modules/core/util/src/test/java/org/rhq/core/util/exec/ProcessExecTest.java
index 49ab3f5..60b7aed 100644
--- a/modules/core/util/src/test/java/org/rhq/core/util/exec/ProcessExecTest.java
+++ b/modules/core/util/src/test/java/org/rhq/core/util/exec/ProcessExecTest.java
@@ -35,7 +35,7 @@
 public class ProcessExecTest {
     public void testProcessExecOutputStream() {
         // run multiple times to ensure race condition fixed
-        for (int i = 0; i < 100; i++) {
+        for (int i = 0; i < 10; i++) {
             ProcessToStart start = new ProcessToStart();

             setupProgram(start);

I also have a version with thread pools being attached. I don't think a thread pool is worth it, but might be good idea.

--- Additional comment from genman on 2012-09-18 17:08:50 EDT ---

Created attachment 614124 [details]
optional use of a thread pool

Additional possible change to use a thread pool to handle I/O redirect. Might be good if there are a lot of script executions, though I don't know if it is worth it for the average case.

--- Additional comment from mazz on 2012-09-18 17:20:41 EDT ---

git commit to master: bf4b25f421769909e886eb42f7b99c66b207b200