Berkeley DB SELF DEADLOCK AND RECOVERY

Question

My berkeley db version is 6.0.20. OS is windows. c++ api.
key words:
DbEnv* m_pDbEnv = new DbEnv(0);
m_pDbEnv->set_cachesize(0,  134217728, 1);
m_pDbEnv->set_mp_mmapsize(1024*1024*1024);
m_pDbEnv->set_timeout(5000000,  DB_SET_LOCK_TIMEOUT);
m_pDbEnv->open("c:\\PICSS_HOME",  DB_CREATE | DB_INIT_CDB | DB_INIT_MPOOL | DB_THREAD, 0);
m_pDb = new Db(m_pDbEnv, 0);
m_pDb->set_flags(DB_DUP);
m_pDb->open(NULL, "c:\\PICSS_HOME\\picss.db", NULL, DB_BTREE, DB_CREATE|DB_THREAD, 0);
Dbt key(name, (u_int32_t)strlen(name));
Dbt data(buff, length);                      // length may large than 10*1024*1024
int ret = m_pDb->put(NULL, &key, &data, 0);
if (ret != 0)
      // do some thing and log
     return;
m_pDb->sync(0);
I have two probrem.
1. sometimes, when execute m_pDb->put(), it dose not return. I got a dump file and found It is executeing:
retry :  MUTEX_LOCK(env, hp->mtx_hash)                      // mp_fget.c, line 256
hp->mtx_hash is 0x2016.  I really do not know why It can not be locked.
I used db_deadlock.exe, but do not work.
I got this problem one time a week nearly.
2. sometimes, m_pDb->put() throw an exception,  BDB0087 DB_RUNRECOVERY, fatal error, run database recovery
it seems something with db environment, is that right? why i got this happened?
Addtional, when i stop program, delete __db.001, __db.002, __db.003, and start program ,all will be ok.
How will I resolve these two problems? Any idea is worthy.
Thanks.
James.
消息编辑者为：8b488b74-a71b-4eb1-8af5-8d9c312bbac1

Answer 1

Hi James,
on point #1 - spurious errors are difficult to debug.   About the only way that we can effectively over the forum is with a reproducible test case.    If you can create this and post it then we can take a look at it.   Since your record size is very large, you should be using a database page size that will fit at least one of your records. 
on point #2 - if you get a DB_RUNRECOVERY error message then it means that the code detected something wrong with the database.    There are too many possible reasons to list them here.    There is generally more information than just the return value, this would be useful to know.   Are there any other error messages?  Have you tried running a db_verify after getting the error message -- this will check for inconsistencies in the db files themselves.
thanks
mike