Generate bank account balance in Oracle table - oracle

Question

I have an accounts table and a movements table in an Oracle 11g database. They work the way you would expect your bank account to work. A simplified version of them would be
CREATE TABLE accounts (
id NUMERIC(20) NOT NULL -- PK
);
CREATE TABLE movements (
id NUMERIC(20) NOT NULL, -- PK
account_id NUMERIC(20) NOT NULL, -- FK to accounts table
stamp TIMESTAMP NOT NULL, -- Movement creation timestamp
amount NUMERIC(20) NOT NULL,
balance NUMERIC(20) NOT NULL
);
You have an account, and some movements are secuentially created, each with a given amount. For example, I would expect the following data to be in the movements table:
| id | account_id | stamp | amount | balance |
-------------------------------------------------------------
| 1 | 1 | 2016-12-29 00:00:01 | 50.00 | 50.00 |
| 2 | 1 | 2016-12-29 00:00:02 | 80.00 | 130.00 |
| 3 | 1 | 2016-12-29 00:00:03 | -15.00 | 115.00 |
-------------------------------------------------------------
My problem is, how do I keep the balance column updated?
I'm doing the inserts inside a Stored Procedure (INSERT INTO movements ... SELECT FROM ...), so it can be done either inside the same query, in a later UPDATE, or with pure PLSQL.
I can think of two methods:
An UPDATE after the insert, something like (an idea, not tested):
UPDATE movements um
SET balance = (um.amount + (SELECT m.balance
FROM movements m
WHERE m.account_id = um.account_id
AND rownum = 1
ORDER BY stamp DESC)) -- last balance from same account?
WHERE stamp > :someDate; -- To limit the updated records
My problem with this is, does it executes in order? From the first movement to the last? Or oracle might run this without specific order, generating the scenario where, for example, the third movement gets updated before the second, so the balance from the second is still outdated?
Cursors: I could define a cursor and run a loop on the ordered list of movements, reading the previous balance of the account in each iteration, and calculating the current balance, setting it with an UPDATE.
This way I would be certain that the balances are updated in order, but I've always avoided cursors because of the performance issues. This stored procedure will work with hundreds of records each time, and the movements table will store millions of records. Will the performance become an issue this way?
My final question would be, considering performance, what is the best way to generate the balance column data?
Edit - Clarification on movements creation
I think I wasn't too clear about this part. At the moment of my SP execution, I'm creating several movements of several different accounts, that's why I mention that the movements creation is done with something like
-- Some actions
INSERT INTO movements (account_id, stamp, amount, balance)
SELECT ... FROM several_tables_with_joins;
-- More actions
That's why I mention that the balance could be generated either in the same query, in a later UPDATE or some other method like the Trigger mentioned in one of the comments.

Answer 1

" considering performance, what is the best way to generate the balance column data"
Usually the ongoing maintenance of summed columns after every transaction incurs a heavier cost than simply calculating them on demand. However, account balance is a special case, because we do need to know it after every transaction, to check, say, whether the account has gone into the red or exceeded an overdraft limit.
The key insight is: before we process a new movement we already know know the current balance. It's the value of BALANCE for the latest MOVEMENT record.
Ah, but how do we know which MOVEMENT record is the latest? There are various different solutions to this, but the simplest would be an ugly is_latest flag. This not only provides a simple way to get the most recent MOVEMENT record it provides a lockable target, which is important in a multi-user environment. We need to ensure that only one transaction is manipulating the balance at any given time.
So, your stored procedure will look something like:
create or replace procedure new_movement
( p_account_id in movements.account_id%type
, p_amount in movements.amount%type )
is
cursor c_curr_bal (p_acct_id movements.account_id%type) is
select balance
from movements
where account_id = p_acct_id
and is_latest = 'Y'
for update of is_latest;
l_balance movements.balance%type;
new_rec movements%rowtype;
begin
open c_curr_bal(p_account_id);
fetch c_curr_bal into l_balance;
new_rec.id := movements_seq.nextval;
new_rec.account_id := p_account_id;
new_rec.stamp := systimestamp;
new_rec.amount := p_amount;
new_rec.balance := l_balance + p_amount;
new_rec.is_latest := 'Y';
update movements
set is_latest = null
where current of c_curr_bal;
insert into movements
values new_rec;
close c_curr_bal;
commit; -- need to free the lock
end new_movement;
/
An alternative to the is_latest flag would be to maintain the current balance as a column on the ACCOUNTS table. The logic would be the same, just SELECT the ACCOUNTS table FOR UPDATE OF CURRENT_BALANCE instead.

Answer 2

I think I would keep the BALANCE in the ACCOUNTS table. Then when you insert your MOVEMENTS record, you update the corresponding ACCOUNT record.