Subject | Re: [firebird-support] two/more "non-conflicting" updates? |
---|---|
Author | Ann Harrison |
Post date | 2004-01-04T16:52:59Z |
David wrote
Suppose client #1 rolls
back his transaction. How does the system know that it should change
the value from 7
to 6?
consistent transaction
will return an error in either case. But you're right, the secret is
keeping update transactions
short.
Regards,
Ann
>D> Suppose the update by client #1 is increasing the value of a numericIn a sense they're non-conflicting, but in another sense they're not.
>D> field by 1 (e.g. 5 becomes 6). And the client #2's update is of the
>D> same operation. In other words, the two updates are non-conflicting.
>
Suppose client #1 rolls
back his transaction. How does the system know that it should change
the value from 7
to 6?
>Nando Dessena wrote:You wouldn't get an error from a wait/read-committed transaction. A
>...
>Then, assuming the transactions are short-lived enough, I believe it
>would be very rare to get an error in no_wait mode, and impossible in
>wait mode.
>
consistent transaction
will return an error in either case. But you're right, the secret is
keeping update transactions
short.
Regards,
Ann