Behaviour in forked processes about davix

[gbitzes]

Indeed, I don't think davix has been designed with explicit support for forked processes, I'm not sure how easy it'll be to add it. In theory, if there isn't an active davix context at the point of forking, both parent and child should be able to use the library by creating a new context.

Yes, example code on how to reproduce the problem would be much appreciated.

[gganis]

Thanks for the prompt reply.
Can you work with the ROOT master?

[gbitzes]

Yes, just let me know the commit of ROOT you're working with and I'll compile it.

[gganis]

Any commit since Friday afternoon (specifically any commit including c4712f03106c50c8fc1e789e7d21cae1434a22f3).
Under tutorials/multicore there is a new tutorial mp103_processSelector.C. To reproduce the problem uncomment the lines defining __reproduce_davix .
You can run the tutorial with
$ root -l -q tutorials/multicore/mp103_processSelector.C
If built with -Dtesting=ON, you can use
$ ctest -V -R mp103_processSelector

Let me know if you need any help.

[gbitzes]

Hi, I'm running the example. Is this the message that indicates the error? I don't get any crashes.
Error in <TDavixFile::Init>: http://root.cern.ch/files/h1/dstarmb.root not a ROOT file

If I disable connection reuse in davix, the message goes away. It seems that the forked process is trying to re-use the old TCP connection of the parent. I'll see about installing some kind of fork handler, which clears the session cache.

[gganis]

Hi,
Yes, that is the indication of the problem.
It does not crash any longer because I have added a protection and error handling.
The forked process get something unexpected when it establish a connection for a new file. I assume this is
because it re-uses improperly something from the parent.

Thanks for looking at it.

On 21 Sep 2016, at 11:09, Georgios Bitzes [email protected] wrote:

Hi, I'm running the example. Is this the message that indicates the error? I don't get any crashes.
Error in TDavixFile::Init: http://root.cern.ch/files/h1/dstarmb.root not a ROOT file

If I disable connection reuse in davix, the message goes away. It seems that the forked process is trying to re-use the old TCP connection of the parent. I'll see about installing some kind of fork handler, which clears the session cache.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub #15 (comment), or mute the thread https://github.com/notifications/unsubscribe-auth/ABp52uEa4Rt4aW7botEetc-yWVzitSbvks5qsPQugaJpZM4J-9LT.

[gbitzes]

The solution I implemented is to keep track of all allocated davix contexts, and after a fork go through all of them in the child process and clear their session caches.

Seems to work, the error message is gone. I'm not sure how fork-safe davix is now, certainly not if a transfer is currently taking place, but this specific case seems fine. Feel free to give it a try, you can get the latest development version from our unstable repos.

Thanks for the report! The fix will land in davix 0.6.5.