Best Practice: Put all fields in a db column or lump some into a serialized hash in a TEXT column?

Rails
1

On my previous countless Rails apps, I’ve always put each data field
as its own column in the DB, but on the last 2 projects, I’ve started
to create a catch-all column that is a MySQL TEXT that I serialize and
treat as a hash for a bunch of general stuff, from arrays, hashes,
etc.

This has worked well on the last 2 apps whose requirements were not
firm, so there are lots of changes on the fly.

So when there is a data field for a model that I know I’m going to
want to do a find_by…, order, or some other database operation on, I
obviously put it into its own SQL column, but otherwise, I’ve been
lumping everything else into that general serialized hash and setting
up methods in the model to get/set them as if they were normal
attributes.

I’m curious to hear some feedback on this technique, e.g. “that bad
programming practice because…” or “it has a performance impact
because…” or “hey, me too”

2

Can’t say as I’m in that ‘hey, me too’ group…

I’d classify that as short-sightedness…
Rails has migrations, form builders, plenty of tools to make agile
development possible. You’ve squirreled away part of your data into a
form that is inaccessible except through your application, or someone
else writing code, possibly in another tool, that attempts to mimic
yours.

What happens when a client calls asking ‘Hey, why can’t I search on X? I
see it on the forms.’

And how about support for 3rd party report writers, ad-hoc querying, or
data warehousing?

-1 for submitting to the demands of the moment. But that’s just my
opinion.

3

want to do a find_by…, order, or some other database operation on, I
obviously put it into its own SQL column, but otherwise, I’ve been
lumping everything else into that general serialized hash and setting
up methods in the model to get/set them as if they were normal
attributes.

I’m curious to hear some feedback on this technique, e.g. “that bad
programming practice because…” or “it has a performance impact
because…” or “hey, me too”

I’ve done it, but I think in general it’s probably a bad idea. If
you’re system changes that frequently the odds that a new business
requirement of “oh hey that field we added last week that you put into
the serialized blob we now need to search against” is going to bite
you. Or, someone will want to interface into the database with some
other language (ie. not Ruby) and they’ll have problems reading that
serialized data.

Where I have done it and I think works all right is for “user
preferences”… little bits of mostly boolean flags or color choices
that you’ll never search on and will vary quite a bit as the
preferences grow…

-philip