Subject | RE: [IBDI] Special kind of triggers? |
---|---|
Author | Dmitry Yemanov |
Post date | 2001-07-12T08:00:45Z |
Alexander and Ann,
to receive a notification as immediate as possible, events may not satisfy
your requests. From my experience, the time of the event delivery may vary
considerably, depending on some conditions. Let me give you just one
example. IB 6.0.1 was running on NT 4.0 server, clients were working on a
variety of Windows platforms and were modifying the data across 100Mbit LAN
over TCP/IP. The database had a lot of procedures updating the tables and
posting appropriate events. In the most cases the clients receive events in
milliseconds, but a couple of concrete events had been received still in
milliseconds by all NT clients and in *tens of seconds* by all Win98
clients. I remember also an information about a situation, when event has
been never received by the client communicating with the server across the
Internet. I haven't investigate the source code to localize this problem,
but it should be done sometime.
Cheers,
Dmitry
> At 12:43 PM 7/10/2001 +0000, Alexander V.Nevsky wrote:Yes, this is the best way, but there may be hidden difficulties. If you want
> > Hi, All. Can somebody from FB team answer is it possible
> to have in
> >FB ability to create special kind of triggers that fires on commit
> >only? I think about logging of changes outside of .gdb
>
> Normally, people use events for that sort of logging. The
> event fires only after the transaction commits.
to receive a notification as immediate as possible, events may not satisfy
your requests. From my experience, the time of the event delivery may vary
considerably, depending on some conditions. Let me give you just one
example. IB 6.0.1 was running on NT 4.0 server, clients were working on a
variety of Windows platforms and were modifying the data across 100Mbit LAN
over TCP/IP. The database had a lot of procedures updating the tables and
posting appropriate events. In the most cases the clients receive events in
milliseconds, but a couple of concrete events had been received still in
milliseconds by all NT clients and in *tens of seconds* by all Win98
clients. I remember also an information about a situation, when event has
been never received by the client communicating with the server across the
Internet. I haven't investigate the source code to localize this problem,
but it should be done sometime.
Cheers,
Dmitry