Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [epp-dev] MAT in Java-based packages

Hi Andrey,
Hi Martin,

I briefly follow-up on your “points-to-keep-in-mind”.

"dumping of 4GB JVM data on network based home users”

Good point. This is an argument against enabling it by default and/or setting a default path where to dump the heap files.


“Instead of [sending] full memory dumps on OOM”

FWIW, I never intended to send the whole memory dump - just a condensed report (the leak suspect analysis to be concrete). See the discussion in the forum I linked in previous emails for more details.


"heap dumps require a lot of disk space”

Yes. We could delete the files after we run the analysis or define a periodic cleanup job to do so or just dump it into the tmp folder.


"one could collect some other information (jinfo, jstack & co)”

It may be me, but I don’t see any added value of using jstack and jinfo over the existing error reports. They basically offer the same information (a stack trace and a few system properties - both can already be included in Neon.1). Please correct me in case I miss an important detail here.


Given these concerns, it sounds like we should ask the user on AERI’s first startup and allow the user to specify the path where these files should be dumped. If the user enables this feature, AERI would rewrite the eclipse.ini and add/remove the corresponding lines. 



Andrey, 
Martin, 
let me know in case you still have concerns when we implement this feature (activation) as outlined above.


Marcel






On 03 Oct 2016, at 18:53, Andrey Loskutov <loskutov@xxxxxx> wrote:

Hi,

please note that in case of network based home users dumping of 4GB JVM data (some even more) to the network drive can be not really helpful and in worst case "lock" the I/O for some time and probably cause the quota limits to be reached. Next, transferring this dump data to the eclipse.org will probably cause our servers to explode and eat lot of traffic. So the users must have a chance to actively confirm that dumps & transfer and we should think if and how will we save that amount of bytes.

Instead of full memory dumps on OOM one could collect some other information (jinfo, jstack & co), which can be also useful but will not cause that big IO traffic costs etc.

Am 29.09.2016 um 17:17 schrieb Marcel Bruch:
Dear package-maintainers – and Gunnar in particular,



I got some feedback from MAT how we could run a “memory leak suspects" analysis in a headless mode - which is what I think is needed in most cases to spot the problems.

There are still two questions open on which I’d like to get your opinion on:


1.
If we’d like to analyze OOM exceptions, we need to instruct the VM to dump the heap whenever they occur. We could do so by (i) adding the  -XX:+HeapDumpOnOutOfMemoryError to eclipse.ini (I know it only works for Oracle VM but 99% of our users use it) or (ii) ask the user on AERI activation whether he want’s to enable heap dumps (if the option is not already set).


2.
Availability of MAT. There are two ways how we could run MAT. If it would be part of the EPP package, we could just start a new process which generates the report (see [1] for more details). If it’s not part of the IDE, we’d need to download it once and save it in the workspace folder.



The most convenient way for me would be to add the flag to eclipse.ini for the committers package and to add the MAT features to it as well.

Gunnar, whats your opinion on that?


Would other package maintainers be interested in sending OOME analyses results for errors that occur in their products? If so, who?

If there are others interested as well, we certainly need to allow users to enable/disable that feature easily and we can’t take the shortcut outlined above (i.e. automatically create heap dumps etc.).



Cheers,
Marcel






On 29 Sep 2016, at 09:56, Marcel Bruch <marcel.bruch@xxxxxxxxxxxxxx> wrote:

Thanks Kaloyan.

I asked in the MAT forum for details [1].


[1] https://www.eclipse.org/forums/index.php/m/1744561/?





On 12 Sep 2016, at 09:22, Kaloyan Raev <kaloyan.r@xxxxxxxx> wrote:

Hi,

On 09/05/2016 11:23 AM, Marcel Bruch wrote:
Just to say it clearly: I don’t know if MAT can be used reasonably well for this use case and I’ve no concrete plan when I’ll work on this item. I’ll follow up when there is something to show.


I remember that at SAP we used (and maybe it is still used) MAT to automatically analyze OOM crashes for a Java Server. MAT run headless and generated a report (text and charts) that was included in a mail alert. It should be able to run in a similar way for the Eclipse IDE too.

--
Kaloyan Raev | Zend Studio Team Lead
Rogue Wave Software, Inc.
Accelerating Great Code
M +359 887 648 663
www.roguewave.com / kaloyan.raev@xxxxxxxxxxxxx
_______________________________________________
epp-dev mailing list
epp-dev@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit
https://dev.eclipse.org/mailman/listinfo/epp-dev

--
Codetrails GmbH
The knowledge transfer company

Robert-Bosch-Str. 7, 64293 Darmstadt
Phone: +49-6151-276-7092
Mobile: +49-179-131-7721
http://www.codetrails.com/

Managing Director: Dr. Marcel Bruch
Handelsregister: Darmstadt HRB 91940



-- 
Kind regards,
Andrey Loskutov

http://google.com/+AndreyLoskutov
_______________________________________________
epp-dev mailing list
epp-dev@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit
https://dev.eclipse.org/mailman/listinfo/epp-dev

-- 
Codetrails GmbH
The knowledge transfer company

Robert-Bosch-Str. 7, 64293 Darmstadt
Phone: +49-6151-276-7092
Mobile: +49-179-131-7721
http://www.codetrails.com/

Managing Director: Dr. Marcel Bruch
Handelsregister: Darmstadt HRB 91940


Back to the top