Wrong Generation for CMS initial-mark about gcviewer

[ecki]

With this code, it works for now:

    for (Iterator i = model.getGCEvents(); i.hasNext();) {
        GCEvent event = (GCEvent) i.next();

        if (event.getGeneration() != Generation.YOUNG)
            continue;

        GCEvent young = event.getYoung();

        if (young.getType() == Type.UNDEFINED)
            continue;

[chewiebug]

Hi Bernd,

You are right, "getGeneration()" doesn't consider detail information of
an event. The parser usually adds a new detail event for every "[ ... ]"
information in a gc log.

So
"6.765: [GC [1 CMS-initial-mark: 0K(25165824K)] 410644K(47815104K),
0.0100670 secs] [Times: user=0.01 sys=0.00, real=0.01 secs] "
results in a main event "GC" (considered YOUNG) and a "CMS-initial-mark"
(considered TENURED) event as detail event inside the GC event.

getGeneration() just returns the main events generation, which is not
accurate in many cases.

I have pushed a fix to return the correct generation. In the case of
initial-mark now TENURED is returned. If more than one generation is
contained, ALL is returned.

If you want to be sure to find all events containing explicit YOUNG
generation information, you should consider all with getGeneration()==
YOUNG and those with getGeneration()==ALL and event.getYoung().getType()
!= UNDEFINED. By the way: Type == UNDEFINED means, that the event in the
logfile didn't contain any information about that generation but I
derived it from the present information (for initial-mark there is only
TENURED generation available in the logfile but it is easy to calculate,
what the size of the YOUNG generation was -> the type of the YOUNG
generation detail event is UNDEFINED).

Regards, Jörg

Am 29.09.2012 03:09, schrieb Bernd:

With this code, it works for now:

| for (Iterator i = model.getGCEvents(); i.hasNext();) {
GCEvent event = (GCEvent) i.next();

     if (event.getGeneration() != Generation.YOUNG)
         continue;

     GCEvent young = event.getYoung();

     if (young.getType() == Type.UNDEFINED)
         continue;

|
|

—
Reply to this email directly or view it on GitHub
#37 (comment).

[ecki]

I wanted to restrict the export to real events, no synthetic timestamps. This worked in my code case, but I will adjust it, need to check which UNDEFINED to include later on, when I export oldgen statistics as well.