Ahh, sorry, i misunderstood your usecase and thought you were talking
about thread-safety rather than process-safety. Yes, you are correct
and the server (and C++ driver) use platform-specific tricks to ensure
that we don't generate duplicate OIDs. Unfortunately that isn't
possible for the C driver since it shouldn't have any
platform-specific code.

I think a good solution would be to add a set_oid_fuzz(int) method
that will allow users to set that as needed. I should probably also
allow you to use a custom inc function to use atomic incs if using
threading.

Could you file a JIRA ticket for those?

On Mon, Jan 3, 2011 at 2:48 PM, Greg Sabatino <gsabatino@dnsstuff.com> wrote:
>
> Mathias:
>
> Thanks for quick reply.  We did see the comment in the code, but don't
> believe the atomic increment would not completely solve our use-case issue.
>  Generally, we sort of hoped to get a call to 'rand' (or other API) each
> time through.  However, around the topic you mention, the only ANSI solution
> we have for you is a file system write-lock.  Other solutions of course get
> into system level resource management which will vary by platform.
> Some quick things we were thinking of to improve the seeding across
> processes would be to use microseconds (rather than seconds), or (on linux)
> use a read from /dev/urandom to seed.  Either should improve initial random
> values by process but are not 100% guaranteed to be unique.  We did plan to
> call the randomizer each time through the routine.
> We additionally planned to look into some other solutions for uid generation
> out there, but haven't put the time in yet...
> -Greg
>
> On 1/3/11 2:00 PM, "Mathias Stearn" <mathias@10gen.com> wrote:
>
> You are correct. If you look at line 114 of bson.c you'll see a TODO
> to make the inc atomic. Unfortunately there is no way to do that in
> ANSI C. Do you know any good way to do that?
>
> On Mon, Jan 3, 2011 at 1:43 PM, Greg Sabatino <gsabatino@dnsstuff.com>
> wrote:
>>
>> Hello Mathias:
>>
>> We are in a development project, and looking to implement the C version of
>> the mongoDB driver (pulled down from github a few weeks back).  The
>> 'oid_gen' routine in the bson.c file is not quite as robust as we would
>> like.  In our scenario, there will likely be multiple processes executing
>> concurrently on the same host and making connections.   It's reasonable to
>> assume that they may hit this code block in the same second, likely
>> generating the same oid.   Any planned rev on the driver in the near-term
>> that may touch this area?  If not, we may take a pass at it sometime in
>> the
>> next month or two and can send back to you.
>>
>> Thanks,
>> -Greg
>>
>
>