DB_RUNRECOVERY error after just opening and closing a DB_ENV and DB

Question

Hi,
I'm currently writing a bdb database maintanance utility for the OS project Netatalk which uses bdb for some data storage.
We're using transactions and therefor we open DB_ENV with
(DB_CREATE | DB_INIT_LOCK | DB_INIT_LOG | DB_INIT_MPOOL | DB_INIT_TXN | [DB_RECOVER]).
DB_RECOVER is only used from the main application, not from the utility I'm currently developing, because afaict running recovery on an environment which another process has opened is not allowed.
Now I can reproduce the following problem by just running my utility exactly 138 times(!) [1] while the utility just:
- opens the environment
- configures logging
- opens a database file which contains one main database with 2 additional indexes
- associates the indexes
- closes the database (indexes first)
- closes the environment
The main application (netatalk) is not running!
At the 138 time I'm getting these errors when closing the database:
---8<---
Run: 137
Run: 138
CNID database dump:
Mai 08 16:06:04.449917 [12991] {dbif.c:328} (E:CNID): error closing database cnid2.db: DB_RUNRECOVERY: Fatal error, run database recovery
Mai 08 16:06:04.450054 [12991] {dbif.c:348} (E:CNID): error closing DB environment: DB_RUNRECOVERY: Fatal error, run database recovery
Error closing database
---8<---
These log messages come from my utility. As I've setup bdb logging I've got some more interesting stuff from that:
---8<---
./log.0000000001: log file open failed: No such file or directory
PANIC: No such file or directory
DB_ENV->log_newfh: 1: DB_RUNRECOVERY: Fatal error, run database recovery
PANIC: fatal region error detected; run recovery
PANIC: fatal region error detected; run recovery
File handles still open at environment close
Open file handle: ./cnid2.db
PANIC: fatal region error detected; run recovery
---8<---
I've tested this on different platforms (Solaris, Linux) with the same result.
I was hoping this info might be enough for someone more experienced with bdb then I am. If not I can put up a minimal source file which reproduces the error. But as that requieres collecting code spread over quite some files I didn't come up with that this time.
Anybody?
Thanks!
-Ralph
[1]
like this:
$ i=1; while true; do echo Run: $i ; pfexec ./dbd -d /Volumes/ACL/ ; if [ $? -ne 0 ] ; then break ; fi ; i=$((i+1)) done
The utility is named dbd.

Answer 1

Ok, got it, sorry for the noise:
I had mixed up the order of int cwd=open(".", O_RDONLY), chdir to bdb env, open bdb stuff, do bdb stuff, close bdb stuff, chdir back to cwd.
I was changing dir back before closing the bdb stuff...
-Ralph