Java7u4 G1 log opens as empty about gcviewer

[chewiebug]

Hi,

Thank you for providing me with a sample log file to analyse the
problem. It has nothing to do with Java7u4 but with some options you
used to generate the gc log output (-XX:+PrintTenuringDistribution,
-XX:+PrintGCApplicationStoppedTime,
-XX:+PrintGCApplicationConcurrentTime).

I have updated the source tree to ignore the output produced by these
options. Are you able to build GCViewer from source? It would be great
if you could check the fix against a few of your log files and tell me
if the problem really is fixed.

Regards, Joerg

On 05.04.2012 14:22, ktyurin wrote:

Hello. I'm trying to open log done by G1 garbage collector on Java7u4 (it's in beta at the moment). It recocgnizes it as 1.6 format, throws few Parse exceptions as usual and finally no graph and no summary. All values are equal 0. The only value it shows is "Total heap (usage / alloc. max)".

I use gcviewer-1.31-RC5.jar. Here is the link to GC log in case you are interested http://db.tt/m3MxmEMi (1MB)

Thanks for supporting this great tool!

---

Reply to this email directly or view it on GitHub:
#22

I built GCViewer from source with Maven but it still opens the sample as empty file.

Here is a new set of CMS GC files. I had a test running last weekend with the following options

-Xms2G -Xmx2G -XX:MaxPermSize=128m -XX:PermSize=128m -XX:CMSInitiatingOccupancyFraction=60 -XX:+UseCMSInitiatingOccupancyOnly -XX:+CMSParallelRemarkEnabled -XX:+DisableExplicitGC -XX:+CMSClassUnloadingEnabled -verbose:gc -Xloggc:gcpipe -XX:+PrintGCDateStamps -XX:+PrintGCDetails -XX:+PrintTenuringDistribution -XX:+PrintGCApplicationConcurrentTime -XX:+PrintGCApplicationStoppedTime

Logs are rotated every day. Finally I glued them up together. When I open it with GCViewer (both SNAPSHOT and RC5 versions) it shows only the last part of the log. When trying to open the first part it shows nothing (even heap size not recognized). I uploaded logs to http://db.tt/9SjiXAcP (10.3MB)

[chewiebug]

Hi,

Thank you for the description and the sample log files. I found that
GCViewer couldn't process log files written with -XX:+UseParallelGC
combined with -XX:+PrintTenuringDistribution. I have fixed it in the
latest source, now parsing of your log files is possible.

On my machine, performance of drawing the graph was very slow. Do you
experience something similar?

By the way: You forgot to use -XX:+UseConcMarkSweepGC to active CMS -
instead you used a parallel scavenge algorithm (which worked enormously
well for your application).

Regards, Jörg

PS: Did you get my answer concerning G1 and the keyword needed in the
first few lines (I can't find the answer on github)? Does it now work
for you?

On 09.04.2012 09:21, ktyurin wrote:

Here is a new set of CMS GC files. I had a test running last weekend with the following options

-Xms2G -Xmx2G -XX:MaxPermSize=128m -XX:PermSize=128m -XX:CMSInitiatingOccupancyFraction=60 -XX:+UseCMSInitiatingOccupancyOnly -XX:+CMSParallelRemarkEnabled -XX:+DisableExplicitGC -XX:+CMSClassUnloadingEnabled -verbose:gc -Xloggc:gcpipe -XX:+PrintGCDateStamps -XX:+PrintGCDetails -XX:+PrintTenuringDistribution -XX:+PrintGCApplicationConcurrentTime -XX:+PrintGCApplicationStoppedTime

Logs are rotated every day. Finally I glued them up together. When I open it with GCViewer (both SNAPSHOT and RC5 versions) it shows only the last part of the log. When trying to open the first part it shows nothing (even heap size not recognized). I uploaded logs to http://db.tt/9SjiXAcP (10.3MB)

---

Reply to this email directly or view it on GitHub:
#22 (comment)

The last set of logs (where I forgot to activate CMS, my bad) opens just fine. Now I clearly observe that GC Time growths... hehe.

The G1 log that I attached before still opens as empty. I removed first 10000 lines and finally it was opened. Does it work for you?

I had problems with performance of drawing the graph too on my work PC. But the same log opens fine on my home laptop. It seems to me that Antialiasing option doesn't work well on PC without video card. I think it would be nice to have this feature disabled by default (like it was in 1.29 if I'm not mistaken) or pass an argument on start to disable it.

After all you saved a lot of my time. I can't thank you enough for these fixes. That's great!

It seems Antialiasing is not an issue. Graph keeps scaling slow even whern it is disabled.

[chewiebug]

I have found that the options showing heap, young and tenured seem to be the culprits slowing the graph considerably. I don't know yet how to solve this, but I will have to look into it (I'll open an issue for this problem). Certain large gc log files like yours are very slow to display.

Concerning G1 my answer seems to have gone missing. I forgot to mention that GCViewer depends on keywords to determine the GC algorithm. It looks for those keywords in the first 4 KB of the log files (it is "pause (young)" in the case of G1 algorithm). Your sample starts somewhere in the middle of a log file. If you delete the first 380 lines of your sample the result should be better.

Using this information, are you able to analyse your G1 log?

Regards, Jörg

[chewiebug]

I have opened issue 23 adressing the performance issue.

It opens fine without first 380 lines. I think this issue can be closed. Thanks!

[chewiebug]

I am happy to have been able to help. The information on the issue you provided helped much to find the reasons, thank you.