Question

OK, so I know there are tons of articles stating I shouldn’t use DOUBLE to store money on a MySQL database, or I’ll end up with tricky precision bugs. The point is I am not designing a new database, I am ask to find way to optimise an existing system. The newer version contains 783 DOUBLE typed columns, most of them used to store money or formula to compute money amount.

So my first opinion on the subject was I should highly recommend a conversion from DOUBLE to DECIMAL in the next version, because the MySQL doc and everybody say so. But then I couldn’t find any good argument to justify this recommandation, for three reasons :

• We do not perform any calculation on the database. All operations are done in Java using BigDecimal, and MySQL is just used as a plain storage for results.
• The 15 digits precision a DOUBLE offers is plenty enough since we store mainly amounts with 2 decimal digits, and occasionaly small numbers wit 8 decimal digits for formula arguments.
• We have a 6 years record in production with no known issue of bug due to a loss of precision on the MySQL side.

Even by performing operations on a 18 millons rows table, like SUM and complex multiplications, I couldn’t perform a bug of lack of precision. And we don’t actually do this sort of things in production. I can show the precision lost by doing something like

SELECT columnName * 1.000000000000000 FROM tableName;

But I can’t figure out a way to turn it into a bug at the 2nd decimal digit. Most of the real issues I found on the internet are 2005 and older forum entries, and I couldn’t reproduce any of them on a 5.0.51 MySQL server.

So as long as we do not perform any SQL arithmetic operations, which we do not plan to do, are there any issue we should expect from only storing and retreiving a money amount in a DOUBLE column ?

Answer 1

Actually it’s quite different. DOUBLE causes rounding issues. And if you do something like 0.1 + 0.2 it gives you something like 0.30000000000000004. I personally would not trust financial data that uses floating point math. The impact may be small, but who knows. I would rather have what I know is reliable data than data that were approximated, especially when you are dealing with money values.