Multiple processes accessing a replicated database

Question

Hi
I am after some help with multiple processes and replicated databases.
I have a primary and secondary database replicated across a pair of servers and this seems to be working well. I'm trying to run another process on one of the machines that opens the environment and databases to view and/or modify the data.
The problem is that when I run this process it causes some sort of corruption such that the server process on the same box gets a DB_EVENT_PANIC the next time it accesses the database. I would like to understand what I am doing wrong.
The servers and standalone process all use the same code to open and close the environment and databases (see below). Just calling
     open_env();
     open_databases();
     close_databases();
     close_env();
in the utility process causes DB_EVENT_PANIC in the server process.
Can anybody spot what I am doing wrong? I am using DB Version 4.7
Thanks
Ashley
open_env() {
db_env_create(&dbenv, 0);
dbenv->app_private = &my_app_data;
dbenv->set_event_notify(dbenv, event_callback);
dbenv->rep_set_limit(dbenv, 0, REPLIMIT);
dbenv->set_flags(dbenv, DB_AUTO_COMMIT | DB_TXN_NOSYNC, 1);
dbenv->set_lk_detect(dbenv, DB_LOCK_DEFAULT)
int flags = DB_CREATE | DB_INIT_LOCK |
          DB_INIT_LOG | DB_INIT_MPOOL |
          DB_INIT_TXN | DB_RECOVER | DB_THREAD;
flags |= DB_INIT_REP;
dbenv->repmgr_set_local_site(dbenv, listen_host, port, 0);
dbenv->rep_set_priority(dbenv, 100);
dbenv->repmgr_set_ack_policy(dbenv, DB_REPMGR_ACKS_ONE);
for (x = 0; x < num_peers; x++) {
dbenv->repmgr_add_remote_site(dbenv, peers[x].name, peers[x].port, &peers[x].eid, 0);
dbenv->rep_set_nsites(dbenv, num_peers + 1);
dbenv->open(dbenv, ".", flags, S_IRUSR | S_IWUSR);
dbenv->repmgr_start(dbenv, 3, DB_REP_ELECTION);
sleep(SLEEPTIME);
close_env() {
dbenv_p->txn_checkpoint(dbenv_p, 0, 0, 0);
dbenv_p->close(dbenv_p, 0);
open_databases() {
db_create(&dbp, dbenv_p, 0)
flags = 0;
if (app_data->is_master)
flags |= DB_CREATE;
dbp->open(dbp, NULL, "primary", NULL, DB_HASH, flags, 0);
... Wait for db if slave and ENOENT ...
primary = dbp;
dbp->open(dbp, NULL, "secondary", NULL, DB_BTREE, flags, 0);
... Wait for db if slave and ENOENT
secondary = dbp;
while (app_data->client_sync) {
sleep(SLEEPTIME);
close_databases() {
     secondary->close(secondary, 0);
     primary->close(primary, 0);
     dbenv_p->txn_checkpoint(dbenv_p, 0, 0, 0);
}

Answer 1

Running recovery (DB_RECOVER flag to env->open()) must be done only in the first process to open the environment.
This is a general rule of Berkeley DB, not specific to replication. You can read more about it in the Reference Guide, on the page entitled "Architecting Transactional Data Store applications".