Hello,

Thanks! This is very interesting.

Could you point me to the lmdb source files which uses this assumption?

Is it mdb_reader_pid() function in mdb.c ?

Do I interpret it correctly that only the processes which access the database simultaniously have to have different pids?

And it means that if I wrap lmdb operations with file locks (using flock) then the data will be consistent (but performance would degrade) ?

Den sön 20 aug. 2023 kl 21:56 skrev Howard Chu <hyc@symas.com>:
>
> > Is it the proper use case for lmdb? Is there any special "flush" API which I have to use to make sure that the record inserted in one container becomes visible
> > in another?
>
> LMDB requires every process using it to have a unique PID. Usually containers all run their job as PID number 1, you'll have to make sure
> they all use unique PIDs.

--
WBR & WBW, Vitaly