Proof crashes with some data and does not with older ones

ROOT
1

Dear Proof experts,

I am using root 5.28/00 (trunk@37585).
I am facing a very anomalous problem, and do not understand whether it depends on Proof or not.
I have written a very stupid TSelector, which actually does nothing, and let it run over some root files. The same code that runs over a newer set of data crashes miserably, and I do not understand why.
Here attached an example of the output of the crash and the corresponding log; this is about proof-lite, but the problem is the same using “full” proof.

The output:

The log:

Is it possible that data prepared with different version of root behave differently under proof?
I am quite sure that the dataset that makes proof crash is ok, since other people are using it with normal root, and also access via browser raises no exceptions or warning.
I have also tryed to use root versions 28, 28c, 28d, 30rc, but nothing has changed.

Thank you very much in advance.

Cheers,
federico

2

Dear all,

I would like to link this post:

here: [url]Proof precess problem with ATLAS D3PD data
in the Proof section of roottalk; it seems that my problem is almost the same (my toy TSelector now is running after applying the suggested trickery), just checking further. anyhow, it would be useful to understand how to interpret the logs, since I could not even imagine how to solve the problem if I did not read the quoted post.
Thank you again,
federico

3

Dear Federico,

Debugging these issues is always complicated, especially when there are many layers involved.
In this case the mastre log did not gave hints, I agree, except that something bad happened on (all) the workers.
On the other end, the workers logs gave the exact location of the segmentation violation which helped to understand and fix the problem in the trunk, and to provide the workaround.

Gerri

4

Hi Gerri,

thank you very much for your answer. Sorry, I did not meant to be unpolite, just wanted to point out that I found very difficult to get my way out of this.

By the way, I do not understand where the trace-log should have helped me; could you please post the lines? in the next few days data will be changed again, so it is possible I will get trapped again in something similar.
Thanks a lot.

cheers,
federico

5

Hi Federico,

Sure, I did not interpreted in such a way.

The point that these problems are difficult to follow even for the ‘experts’, especially if the happen in the PROOF code, not in yours.

In the case in exam, the stack trace in the worker log indicated that a ‘segmentation violation’ occurred at line 519 of proof/proofplayer/src/TEventIter.cxx :

The only thing we can ask you in these cases is to post the logs,as you have done. But, of course, if you want to dig the code by yourself you are free to do it … :wink:

Gerri