Segment fault in concurrent access to BDB

Question

Hi, All
I am extending a single thread BDB application to allow concurrent
access to the DB from 2 threads, transaction is not needed.
The environment is opened with the flag (DB_CREATE | DB_INIT_MPOOL | DB_INIT_LOCK).
Then I open a new DB under the same directory in each thread, and
modify the DB simultaneously.  my expectation was that each thread
will not conflict with each other and in turn improve the
throughput. While the program constantly crashed because of segment
fault in file (src/lock/lock.c:832, both threads crashed on this
line).
I inspected the back trace, the only suspecious thing I found is that
the 2 threads are using the same locker object[1], so I'm wondering if
anyone has similar experience implementing concurrent BDB
applications? And what's possible cause of my problem and possible
fixes? Thanks.
[1]
thread 1: #0  0x00002aaab3651a7f in __lock_get_internal (lt=0x13cbda0, sh_locker=0x2aab0119f258, flags=4, obj=0x1c55650, lock_mode=DB_LOCK_WRITE, timeout=0, lock=0x2aab449fe8a0)
    at ../src/lock/lock.c:832
thread 2: #0  0x00002aaab3651a7f in __lock_get_internal (lt=0x13cbda0, sh_locker=0x2aab0119f258, flags=4, obj=0x13da640, lock_mode=DB_LOCK_WRITE, timeout=0, lock=0x2aab3ffc78a0)
    at ../src/lock/lock.c:832

Answer 1

Hi, Mike
Thanks for the timely reply!
I referred to the Berkeley DB Transaction Guide, while the doc does not mention the consequence if you violate the scenario requirement. 
1. I choose the default Data Store model, and my program will allow multiple writers, what's the expected behavior? 2 threads using the same locker compete different locks is supposed to segment fault?  I'm also curious about how a put(key, data) correspond to a locker, keys in the memory page map to the same locker?
2. I try to enable concurrent reader & writer without transaction, while the guide assumes transaction is enable, is Data Store able to achieve this? Or at least Concurrent Data Store is required?
Thank you very much!