Hello, I'd like to understand how is failover mechanism, added back in 3daef94c7aa463be21b2d428eae2b5b21f10a6aa, supposed to work. My goal is to generalize its implementation for PostgreSQL to all (non-Oracle) databases, as the code in src/backends/postgresql/error.cpp is really not PostgreSQL-specific at all, but there are a couple of problems with this.
First one is https://github.com/SOCI/soci/issues/793, i.e. that the current implementation in PostgreSQL backend doesn't use the correct connection string when trying to reconnect. This is not a huge problem on its own, and I've already fixed it locally, but it made it impossible to test this and so makes me suspicious about whether it was actually tested at all with this database. Second problem is that for me failover means transparently continuing running even if the database becomes (temporarily) inaccessible. In particular, I thought the failover mechanism would re-attempt executing the SQL statement which resulted in an error. However this isn't the case currently in PostgreSQL backend at all. So I wonder if it's "just" another bug in this code (which seems relatively plausible, considering the absence of testing mentioned above) or if my assumption is wrong and this is not how it's supposed to work at all? This is partially related to the third problem/question: does anybody know how to test this mechanism with Oracle, which is the only backend with native support for this functionality? I've naively thought that just breaking connection to the server would be enough to trigger it, but what actually happens is that attempting to execute any SQL statement when connection is unavailable (I've just added iptables rule on the local router rejecting packets to the remote server to emulate this) simply hangs apparently forever -- or at least beyond the time I was prepared to wait (several minutes). Does anybody know how is this supposed to work and what kind of configuration do I need (either client or server-side) to test failover with Oracle? Finally, I wonder if it's really even worth spending time on this or if it would, perhaps, be better to just provide a ping() method checking if the connection is still alive and let the application use it when handling SOCI exceptions to try to reconnect on its own, if necessary? This would surely be much simpler and the only advantage of the current failover mechanism I see is that it's natively supported by Oracle. But without knowing how does it work there (and, in particular, whether the statements are still run if a failover happens during their execution), I don't know if it's really important. I.e. if Oracle doesn't run the statement to the completion neither and it's up to either SOCI or the application to do it anyhow, I don't see much point in using this mechanism instead of simply reconnecting when the connection is lost. Thanks in advance for any help/hints! VZ
pgpV7sat3rI_r.pgp
Description: PGP signature
_______________________________________________ soci-devel mailing list soci-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/soci-devel
