Hi,

I experienced data disappear. Four tables with a 1-n relationship are
heavily updated and are target of some big queries.

Cenario:
Table A with ~3.000 rows sometimes rollback updates
Table B with ~5.000 rows sometimes rollback updates and inserts
Table C with ~60.000 rows have several rollbacked inserts
Table D with ~2.500.000 rows is havely rollbacked

The workflow:
1) Users inserts a row at table A
2) Users inserts one or more rows at B
3) Users inserts lots of rows at C
4) Users inserts lots of rows at D
5) Users updates D
6) Users updates B and C
7) Users updates A and prints a report with all content from the four
tables concerning with the row from A
8) Minutes or hours later users checks for that report and updates A
9) Days later users print a big report from all month data

Problem:
At steps 8) and 9) users tell some rows from some tables are vanished!

Details:
All update, delete and insert are performed by Stored Procedures.
Some selects are performed by no indexed views.
Some selects use more than 12 joins!
Reports are printed before the data vanished.
Time to data disapear (at users discretion) varies from minutes to days.
Each insert, each update, each delete for EACH ROW, is a explicit
transaction.
I´m using MSSQLServer 2000 last sp and Windows Server 2000.
The server and users lies at the same network, the users uses a client
application (not web).
The user´s app connects with BDE but i coded a newer version with ADO
but this not worked.
All worked fine for sometime but sundenly this problem appears.
Users tell the loss of data is harder now the loss is growing.
The LAN posses 2 hub is slowdown from 100MB to 10MB all network (the
number of users is constantly growing).
All server memory is used, some locks take too long some deadlocks
errors arise.
I put a trigger to make sure any insert/update at the problematic tables
inputs a row at a dummy table (since i have the report i can search for
the row) no missing row are found at the dummy table.
Some triggers make a complex validations at some inserts/tables and
roolbacks the transaction, but when tested these triggers works fine.
I run checkdb, rebuild indexes checked realtionships, stored procedures,
application, triggers, etc.

Conclusions:
We need a better server with a lot more memory and a rebuild at the
network.
This problem can be from a deadlock for sources like
memory+network+diskIO and the SQLServer cannot dont handle it.
The locks for sources are handled by windows but maybe the OS failed to
comunicate deadlocks problems to SQLserver.

Workaround:
I activated de server log to write any deadlock, timeout is -1(infinite)
a new server (and switches) are arriving shortly.
I hope this solves the problem. Someone experienced data loss like it?

Jean

*** Sent via Developersdex http://www.developersdex.com ***
Don't just participate in USENET...get rewarded for it!