• ойти Ask yourself this question. If your application sends a COMMIT command to the database, and it never gets an answer back, what does it do? Does it assume the COMMIT failed? Does it assume the COMMIT worked? Either approach is clearly wrong, even if replication is not in use. After all, if the database crashed, it could have done so either before or after processing the COMMIT statement; your application has no way of knowing which it is. To be correct, your application must re-establish a database connection as soon as possible, examine the database, somehow determine whether or not the transaction in question committed prior to the crash, and if not, redo the whole transaction. Does your application do that?

Replies (0)