[Cocci] Working with pipes for parallel SmPL data processing?

[Cocci] Working with pipes for parallel SmPL data processing?

[Markus Elfring]

Hello,

The Coccinelle software supports also a variant of parallel data processing
after the parameter “--jobs” was passed.
https://github.com/coccinelle/coccinelle/blob/7ec31ed1fadf738bc487ccefdc63bfe0598f44cc/docs/manual/spatch_options.tex#L745

It is nice when it works to distribute analysis on source files to some processors.
Unfortunately, undesirable software behaviour can be observed if a database
like “PostgreSQL 11.3-7.1” would be used in such a system configuration.
Thus I imagine that it can be occasionally appropriate to perform desired
parallel data processing by the usage of a detailed pipeline instead.
https://ocaml.github.io/ocamlunix/pipes.html

How do you think about another application for this kind of
inter-process communication?

Regards,
Markus
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] Working with pipes for parallel SmPL data processing?

[Julia Lawall]

[-- Attachment #1: Type: text/plain, Size: 1374 bytes --]



On Sun, 9 Jun 2019, Markus Elfring wrote:

> Hello,
>
> The Coccinelle software supports also a variant of parallel data processing
> after the parameter “--jobs” was passed.
> https://github.com/coccinelle/coccinelle/blob/7ec31ed1fadf738bc487ccefdc63bfe0598f44cc/docs/manual/spatch_options.tex#L745
>
> It is nice when it works to distribute analysis on source files to some processors.
> Unfortunately, undesirable software behaviour can be observed if a database
> like “PostgreSQL 11.3-7.1” would be used in such a system configuration.
> Thus I imagine that it can be occasionally appropriate to perform desired
> parallel data processing by the usage of a detailed pipeline instead.
> https://ocaml.github.io/ocamlunix/pipes.html
>
> How do you think about another application for this kind of
> inter-process communication?

When you use -j, you fork processes that thus the child processes write to
a different region of memory than the parent process.

I don't have any interest in connecting Coccinelle to a database, so I'm
not going to spend any time on debugging your issue.  However, if you want
to look into it, this may be the source of the problem.

Maybe you can use a hashtable, then use merge in a finalize, and the write
the information to the database in the parent thread.

The use of merge is illustrated in eg tests/merge_vars.cocci

julia

[-- Attachment #2: Type: text/plain, Size: 136 bytes --]

_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] Working with pipes for parallel SmPL data processing?

[Markus Elfring]

Am 10.06.19 um 23:00 schrieb Julia Lawall:
>
>
> On Sun, 9 Jun 2019, Markus Elfring wrote:
>
>> Hello,
>>
>> The Coccinelle software supports also a variant of parallel data processing
>> after the parameter “--jobs” was passed.
>> https://github.com/coccinelle/coccinelle/blob/7ec31ed1fadf738bc487ccefdc63bfe0598f44cc/docs/manual/spatch_options.tex#L745
>>
>> It is nice when it works to distribute analysis on source files to some processors.
>> Unfortunately, undesirable software behaviour can be observed if a database
>> like “PostgreSQL 11.3-7.1” would be used in such a system configuration.
>> Thus I imagine that it can be occasionally appropriate to perform desired
>> parallel data processing by the usage of a detailed pipeline instead.
>> https://ocaml.github.io/ocamlunix/pipes.html
>>
>> How do you think about another application for this kind of
>> inter-process communication?
>
> When you use -j,

with a value bigger than one


> you fork processes that thus the child processes write to
> a different region of memory than the parent process.

This can be.

But such forked processes have got access also to copied data
from the the parent process.


> I don't have any interest in connecting Coccinelle to a database,

I am curious if any other users of your software can adjust
the corresponding development interests.


> so I'm not going to spend any time on debugging your issue.

I hope that collateral software evolution can happen if more advices
will be taken better into account from existing software documentation.
https://docs.sqlalchemy.org/en/13/core/pooling.html#using-connection-pools-with-multiprocessing


The data storage is working already if known database connections
would be performed only by a single process (instead of background processes).


> Maybe you can use a hashtable,

I would find an other data structure more helpful for the storage
of analysis results from parallel source code processing.


> then use merge in a finalize,

Thanks for such feedback.


> and the write the information to the database in the parent thread.

This is also my software development idea for working with a detailed
process pipeline.


> The use of merge is illustrated in eg tests/merge_vars.cocci

I would about the detail that a test source file is empty here.
https://github.com/coccinelle/coccinelle/commit/0207001b5241092e7b22cf4f99b22db13c01a24d#diff-27862a1fe13b63e871b358e9fc54b3b7

I find the usage of Coccinelle's merge functionality unclear at the moment.
Can such an area become better documented?

Regards,
Markus
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci