[Pdns-dev] [PATCH][RFC] gmysqlbackend: reliability enhancement, reconnect if required

[Peter Fern]

Hi Bert/list,
I'm curious as to why the attached patch (or something similar) has not 
been implemented, since a mysql_ping() represents minimal overhead, and 
results in results being returned in a reliable manner, rather than 
relying on potentially infinite idle timeouts on the mysql connection.  
This issue has been raised a number of times, but I can't see a detailed 
response as to why this shouldn't be fixed.

With a shared MySQL server it may be undesirable to have unlimited 
wait_timeouts server-wide, however this represents a problem with 
infrequently queried pdns instances.  In our case, we have dedicated 
gmysqlbackend master pdns instances that are never queried directly for 
records, and my be idle for days if there are no zone updates.  When a 
notify is triggered, if the connection has gone away, the query simply 
fails, the notify is unsuccessful and the slave never updates, however 
the notified_serial may be updated.  This is obviously a sub-optimal 
scenario.

The attached simple patch provides for reliable retrieval of results in 
a heterogeneous environment.  Please provide any feedback or comments 
you may have.

Regards,
Peter Fern

-------------- next part --------------
A non-text attachment was scrubbed...
Name: mysql-reconnect.patch
Type: text/x-patch
Size: 832 bytes
Desc: not available
Url : http://mailman.powerdns.com/pipermail/pdns-dev/attachments/20080820/4b9241fc/mysql-reconnect.bin