[Opendnssec-user] kaspConnect() causes ods-enforcer brittleness

[Petr Spacek]

On 8.12.2015 09:22, Havard Eidnes wrote:
> Hi,
> 
> the ods-enforcer daemon loops, periodically looking for things
> which need to be done / scheduled.  Inside its service loop it
> has this piece of code:
> 
>         log_msg(config, LOG_INFO, "Connecting to Database...");
>         kaspConnect(config, &dbhandle);
> 
> followed by misleadingly-indented code (that's a minor issue).
> 
> However...  What happens if an administrator just happens to run
> "ods-ksmutil key list ..." at exactly that time when the enforcer
> wants to connect to the database?  This is what the log can tell:
> 
> Dec  8 08:32:43 hugin ods-enforcerd: Connecting to Database...
> Dec  8 08:32:43 hugin ods-enforcerd: ERROR: error executing SQL - database is locked
> 
> and this time it looks like ods-enforcerd really means that this
> is an error, because kaspConnect() doesn't return an error but
> does an exit() if connecting to the database fails:
> 
> /*
>  * Connect to the DB
>  */
> void
> kaspConnect(DAEMONCONFIG* config, DB_HANDLE     *handle)
> {
>     /* Note that all these XML derived strings are unsigned chars */
>         if (DbConnect(handle, (char *)config->schema, (char *)config->host, (char *)config->password, (char *)config->user, (char *)config->port) != 0) {
>         unlink(config->pidfile);
>         exit(-1);
>     }
> 
> }
> 
> That's not a good way to design a library function interface!
> 
> This makes ods-enforcer brittle, provides no automatic error
> recovery, and requires operator intervention to restart the
> now-dead enforcer.
> 
> Again, this is with OpenDNSSEC 1.4.7, using sqlite3.

I confirm that I have seen the very same issue.

BTW what is the purpose of home-grown lock around SQL database? I did not dig
into it but I would somehow expect that SQL database can deal with access from
multiple processes...

-- 
Petr Spacek  @  Red Hat