Re: Multiple concurrent isolated transactions on ADO Dataset

I still don't see the problem

for you're scenario, you take out the data you need from the database change
it and write it back. You just don't need to write back temporary results
while you're in a business transaction.

Person A want's to change the seat:
- he checks the current status of seats and sees 89 is not taken yet, so
he changes his local data: he removes him self from 171 and places him on
89 - just in the small dataset that contains the seat reservation relevant
to him
- when he comits, after his coffe break and phone calls, optimistic
concurrency is applied
- he won't be rejected on removing him from seat 171 because the
customer only talked to one travel agent
- he will maybe fail to change to seat 89 (be cause first come first
serve or optmistic concurrency detected modifications on that "row" - or how
ever this is done...) then the whole sysem transaction is rolled back which
consists of removing him from 171 and placing him on 89 (but it runs only
for milliseconds)

if it's web app there might not even be a dataset involved
- the guy just says he want to change the booking, sees a list of
available seats, clicks on one and the system will change the booking - if
the list he saw was 2 hours old chances are high that the system detects
concurrency vialoations (optimistic once in that case) and again the system
transaction is rolled back...

maybe you should try to get your datasets smaller and fill them on demand
with data. I mean, it doesn't make sense to fill the dataset with all flight
reservation informations for all the flights of the next 12 months when a
customer is only interested in one particular flight. Then you're master
dataset is suddenly the database and the datasets you have with the chances
contain more than the changes but not all the data but only the one relevant
to the user or the use cases that user is working in.

Since it's likely that this client is human it's natural that he will have a
cetain limitation on data he can work on in one particular use case or maybe
a series of use cases. So, I still think you can get rid of the big caching
dataset.

cheers

chris

"Mark Roberts" <Mark_Roberts@xxxxxxxxxxxx> schrieb im Newsbeitrag
news:u6PumnvWFHA.3864@xxxxxxxxxxxxxxxxxxxxxxx
> Well, i have a situation where it's fairly critical for changes made to
> the data to be only visible to the person making the changes (until
> commit), whilst the unchanged version should be visible to all other
> users... until they make changes.
>
> When the changes are committed, it's a first come, first served. I.e. it's
> possible for person B to have their changes rejected as a result of person
> A's changes.
>
> I'm using a command pattern to implement the updates, inserts and deletes,
> and stacking them up for the final commit...
>
> Scenario:
> Imagine a plane with 200 seats, and person A is allocated to seat 171, but
> wants to be allocated to seat 89... the changes necessary re-allocate him
> would be to deallocate from 171 and allocate to 89. Say the user was
> interupted on the phone between deallocating from 171 and re-allocating to
> 89, and at that same time, another user allocated someone to 89 and
> another to 171... Person A would now be without a seat on the plane.
>
> Sorry for silly example :)
>
> Mark.
>
>
>
> "christian kuendig" <xxxx> wrote in message
> news:uSFMEdvWFHA.1044@xxxxxxxxxxxxxxxxxxxxxxx
>> You don't need to keep the connection open
>>
>> Why don't you use optimistic or pesimistic offline locking? (the
>> optmisitc one is more or less included in the dataadapter)
>>
>> "Mark Roberts" <Mark_Roberts@xxxxxxxxxxxx> schrieb im Newsbeitrag
>> news:u%23B7CWvWFHA.3760@xxxxxxxxxxxxxxxxxxxxxxx
>>> Hi
>>>
>>> No, it's not really possible to operate directly on the database
>>> unfortunately - the changes that the user can make will be over a long
>>> time (i.e. could be 1-2 hours, maybe more)
>>>
>>> I don't really want to attempt to keep a database transaction open for
>>> that long :)
>>>
>>> Regards,
>>>
>>> Mark.
>>>
>>>
>>>
>>> "Rogas69" <rogas69@xxxxxxxxxxxxxxxx> wrote in message
>>> news:OpQNrSvWFHA.1796@xxxxxxxxxxxxxxxxxxxxxxx
>>>> What database do you use? You describe scenario that is easily handled
>>>> by practically any database. You said that amount of data is huge -
>>>> creating per-user copies will kill your server - think about memory
>>>> allocation and copying. Is it not possible to operate directly on
>>>> database?
>>>>
>>>> Peter
>>>>
>>>> "Mark Roberts" <Mark_Roberts@xxxxxxxxxxxx> wrote in message
>>>> news:exDNqHvWFHA.612@xxxxxxxxxxxxxxxxxxxxxxx
>>>>> Hi,
>>>>>
>>>>> I have been thinking about a problem for some time, and i'd like to
>>>>> throw it open to the newsgroups in case someone has come across a
>>>>> similar situation.
>>>>>
>>>>> I have a C# .Net 2.0 web application with a very large server-side
>>>>> dataset (i.e. can only be 1 in memory).
>>>>>
>>>>> I have multiple users that wish to make changes to the dataset, and
>>>>> each user's should be batched up and only visible to themselves until
>>>>> they're committed, at which point they become part of the main
>>>>> dataset.
>>>>>
>>>>> I am thinking of have one dataset per user as well as the main
>>>>> dataset, and recording changes into the user-based dataset.
>>>>>
>>>>> This is ok, but it adds a lot of complexity when querying the data,
>>>>> since queries are essentially across two datasets.
>>>>>
>>>>> Ideally, i'd like to have multiple transaction layers on the main
>>>>> dataset, one for each user that isolates their changes, but permits
>>>>> dataset searching to work in the context of their transaction until
>>>>> it's committed or rolled back.
>>>>>
>>>>>
>>>>> Many thanks!
>>>>> Mark.
>>>>>
>>>>
>>>>
>>>
>>>
>>
>>
>
>


.

Relevant Pages

  • Re: SQL Query problem take ages to run
    ... SQL Server will allocate space in TempDB, ... database, and the target database transaction log. ... If you allow SQL Server to perform the allocations as needed, ...
    (microsoft.public.sqlserver.programming)
  • Re: Multiple concurrent isolated transactions on ADO Dataset
    ... and stacking them up for the final commit... ... wants to be allocated to seat 89... ... would be to deallocate from 171 and allocate to 89. ... it's not really possible to operate directly on the database ...
    (microsoft.public.dotnet.framework.adonet)
  • Re: AFL solved
    ... A customer books a single seat. ... allocate the leftmost seat in the longest unbroken sequence of empty ... Kafka wasn't an author; ...
    (comp.lang.c)
  • Re: BoardFirst Transaction Receipt
    ... >> For transaction details see the information below. ... > be the first to board your plane and Get the seat of your choice, Avoid ...
    (alt.vacation.las-vegas)
  • Re: Change system time and date by script
    ... If you want the best seat in the house, ... You'll have to move the cat. ... I am doing some development work on a database for a client. ...
    (microsoft.public.windowsxp.basics)