Am 02.05.2011 23:22, schrieb Jesse Norell: > Hello, > > I just pushed a commit to check the database connection and reconnect. > Any errors that happen during a check should simply return "DUNNO" now, > so maybe a message slips by the quota check that would otherwise have > been rejected (dbmail itself will still catch and bounce that as > previously).
sounds good i will have an eye on this because i love the idea >> maybe someday in dbmail spread read/write makes sense too >> but here also i would prefer to not include another sofwtare layer >> >> * writes always to the master >> * if configured using a read-connection pool >> * choose one of the existing connections from the pool at the begin of >> user-connect >> * if it is dead, choose another one of the pool >> * perfectly let the pool-child connect again randomly to one of the >> available hosts > > Something like this might work right now based on DNS round-robin. > Setup your database host to point to a hostname and add multiple DNS > entries for it - if the connection goes away, it should reconnect. It > may work, or might not (eg. if the hostname isn't looked up again), but > is something to play with if you like, prior to actual support in the > daemon yes, but the app must sepearte read/write on the postfix-side no problem because read-only the postfix-implementation is perfect from the view of HA * if there are unix-sockets and inet configured sockets are preferred * best for performance if no load-balancing is needed * if you configure only inet-database-connections it goes random as long there is no really high load you should prefer localhost via unix-socket for mysql i found hughe performance differences of network for our dns-backend and since web-ui/mysql are on the same server the whole thing is so much faster (dns-servers fecthign generated configs)
signature.asc
Description: OpenPGP digital signature
_______________________________________________ DBmail mailing list [email protected] http://mailman.fastxs.nl/cgi-bin/mailman/listinfo/dbmail
