On Jul 31, 2007, at 1:09 PM, Ralf S. Engelschall wrote:

> On Tue, Jul 31, 2007, Jeff Johnson wrote:
>
>> On Jul 31, 2007, at 10:39 AM, Ralf S. Engelschall wrote:
>>
>>> We are using fcntl(2) locking for DB in RPM since over 5 years  
>>> now and
>>> never encountered any problems. And we drive RPM on really lots of
>>> different platforms. So, I expect that fcntl(2) locking will not  
>>> make
>>> any problems at all...
>>
>> With or without DB_INIT_CDB, i.e. with concurrent write acces or not?
>
> The "cdb" option is set in "macros" for the DB files if this is  
> what you
> mean...

Mostly. But if openpkg has an additional exclusive lock, then  
concurrency
has not been achieved.

Try rpm -qa in a loop, run rpm -i; rpm -e in another loop, repeat  
multiple times
is what I used to convince myself that rpmdb (note rpmdb, not rpm)  
concurrency
was feasible.

The problem with fcntl locking in db/mutex/README is what I worry,  
although
a highly unlikely scenario in the real rpm world so far.

73 de Jeff