bsddb3 thread problem

[Rhamphoryncus]

On Apr 1, 1:29 am, "anuraguni... at yahoo.com" <anuraguni... at yahoo.com>
wrote:
> In my application I am trying to access(read) a DB thru a thread while
> my main thread is adding data to it and it gives following error(s)
>
> bsddb._db.DBRunRecoveryError: (-30974, 'DB_RUNRECOVERY: Fatal error,
> run database recovery -- PANIC: Permission denied')
>
> and sometimes
> bsddb._db.DBRunRecoveryError: (-30974, 'DB_RUNRECOVERY: Fatal error,
> run database recovery -- PANIC: fatal region error detected; run
> recovery')
>
> sometimes
> bsddb._db.DBInvalidArgError: (22, 'Invalid argument -- DB_LOCK-
>
> >lock_put: Lock is no longer valid')
>
> sometimes pure seg fault.
> Program received signal SIGSEGV, Segmentation fault.
> 0xb7c1b845 in __bam_adjust () from /usr/lib/libdb-4.4.so
>
> and some time memory usage keeps on increasing and cpu is 100%
> it crashes with memory error.
>
> This doesn't happen always, almost 1 in 10 cases.
> If i use simple python threaded function instead of threading class,
> it works.
>
> I have attached a simple script which tries to replicate the scenario.
>
> Do anybody has a clue what I am doing wrong here?
> I suppose bsddb3 DB can be accessed from mutiple threads?
> or do I need to specifically set DB_THREAD flag? though with
> db.DB_THREAD it hangs on some mutex?
>
> Thanks a lot
> Anurag
>
> -------
> import time
> import os
> import threading
> import thread
> import shutil
> from bsddb3 import db
>
> class DocQueueConsumer(threading.Thread):
>
>     def __init__(self, queueDB):
>         threading.Thread.__init__(self)
>         self.queueDB = queueDB
>         self.setDaemon(True)
>
>     def run(self):
>         while True: self.queueDB.cursor()
>
> def crash():
>     path = "/tmp/test_crash"
>     if os.path.exists(path):
>         shutil.rmtree(path)
>     os.mkdir(path)
>
>     aBigEnv = db.DBEnv()
>     aBigEnv.set_cachesize(0, 512*1024*1024)
>     aBigEnv.open(path, db.DB_INIT_CDB|db.DB_INIT_MPOOL|db.DB_CREATE)
>
>     queueDB = db.DB(aBigEnv)
>     queueDB.open('mydb', dbtype=db.DB_RECNO, flags=db.DB_CREATE)
>
>     DocQueueConsumer(queueDB).start()
>     for i in xrange(10**5):
>         if i%1000==0: print i/1000
>         queueDB.append("something")
>
> crash()
> -------

It's my understanding that the connection is NOT thread-safe.  Your
thread should be using an entirely separate connection.