Hi, If Sinática Monitor reports a transaction as "Read Committed no Record Versions, Read Write", am I right in these assumptions:
1. Read/write mode, i.e. both selects, updates, inserts and deletes are allowed. 2. On each statement, for each record, it will access only the latest record version, in effect ignoring MVCC. 3. Assuming the lack of wait/no wait info means "no wait": If the latest record version is uncommitted, it will report a lock conflict, even if the attempted operation is a read/select. This is in an OR framework (ECO, a bit like NHibernate). I think it uses this mode for all operations. It would seem to me that this mode limits the number of concurrent operations more than is really motivated. I am considering to use "rec versions" instead., in which case I would expect the above to turn into: 2. On each statement, for each record, it will *read* the latest committed version. Does this count per statement start time or is it "asynchronous" so that a record version committed by another transaction in the middle of a statement execution will be read by that statement? 3. Assuming the lack of wait/no wait info means "no wait": On each statement, for each record, a *write* will fail with a lock conflict error if there is an uncommitted record version. In view of the OR framework, this transaction mode would mean that when reading it will always get the latest possible data. If the reads are "async" as I ask in point 2, the data received runs the risk of being inconsistent, which is not good. When writing data, if the OR framework doesn't add any explicit checks for record versions, it will overwrite changes written by other transactions, without noticing, assuming the other changes are committed. Uncommitted changes will result in an exception, due to lock conflicts. Kjell -- -------------------------------------- Kjell Rilbe DataDIA AB E-post: [email protected] Telefon: 08-761 06 55 Mobil: 0733-44 24 64
