28

According to PostgreSQL's docs, there's no performance difference between VARCHAR, VARCHAR(n) and TEXT.

Should I add an arbitrary length limit to a name or address column?

Edit: Not a dupe of:

I know the CHAR type is a relic of the past and I'm interested not only in performance but other pros and cons like Erwin stated in his amazing answer.

38

The answer is no.
Don't add a length modifier to varchar if you can avoid it. Most of the time, you don't actually need a length restriction anyway. Just use text for all character data. Make that varchar (no length modifier) if you need to stay compatible with RDBMS which don't have text.

Performance is almost the same - text is a bit faster in rare situations, and you save the cycles for the check on the length.

If you actually need to enforce a maximum length, still use text and add a check constraint for that:

ALTER TABLE tbl ADD CONSTRAINT tbl_col_len CHECK (length(col) < 51);

You can modify or drop such a constraint at any time without having to mess with the table definition and all depending objects (views, functions, foreign keys, ...)

With length modifiers you just run into problems like this or this or this ...

PostgreSQL 9.1 introduced a new feature to alleviate the pain somewhat. I quote the release notes here:

Allow ALTER TABLE ... SET DATA TYPE to avoid table rewrites in appropriate cases (Noah Misch, Robert Haas)

For example, converting a varchar column to text no longer requires a rewrite of the table. However, increasing the length constraint on a varchar column still requires a table rewrite.

  • I think this answer would be a lot better if it was simply "no don't add arbitrary limits to a real database ever." I feel like a lot of this answer needs corrections and further information, but that it's totally off topic and would distract from your conclusion which I totally agree with. – Evan Carroll Mar 18 '18 at 2:22
  • Yes, all based on Postgres versions before 9.1 - 6 years ago. A bit dusty by now, but the basic advice is still good. – Erwin Brandstetter Mar 25 '18 at 21:30
  • Is it a good or bad idea to add a check constraint for every text column for the purpose of a sanity check and ensuring a bug in the client doesn't use up all the database's disk space by inserting a very large text? – Code Jan 2 at 19:21
  • @Code: It's a viable option. If you have many columns with the same constraint, consider domains. Or varchar(n) after all, for simplicity - if the downsides don't typically affect you. (The limit is not arbitrary in your case if you want to enforce an actual maximum length.) – Erwin Brandstetter Jan 2 at 19:51
11

If you see the length limit as a kind of check constraint to make sure you validate the data, then yes add one. Actually you might want to not use a length definition but a real check constraint instead, to make changing the limit faster.

To change (increase) a length limit you need to run an ALTER TABLE which might take a long time to finish (due to a possible re-write of the table) during which an exclusive table lock is necessary.

Changing (i.e. dropping and re-creating) a check constraint is a very brief operation and only requires reading the table's data, it will not change any rows. So that is going to be a lot quicker (which in turn means the exclusive table lock will be held for a much shorter amount of time).

During operation there is no difference whatsovever between a text, a varchar or a varchar(5000) column.

  • out of plain curiosity, why do you think this length check cannot be done on a client application while capturing data? – PirateApp Mar 13 '18 at 12:35
  • 4
    @PirateApp: because very often there will be more than one application or some external data source (think nightly batch imports). And almost always the database (and the data) live longer than one application. – a_horse_with_no_name Mar 13 '18 at 12:38
1

It looks like there might be some performance difference if VARCHAR is regularly used to store very large strings, since "long strings are compressed by the system automatically" and "very long values are also stored in background tables." Theoretically this would mean that a high volume of requests for a very long string field would be slower than for a short string field. You'll probably never run into this problem, since names and addresses aren't going to be very long.

However, depending on how you're using these strings outside your database, you might want to add a practical limit to prevent abuse of the system. For example, if you're displaying the name and address on a form somewhere, you might not be able to display a whole paragraph of text in the "name" field, so it would make sense to limit the name column to something like 500 characters.

  • 1
    AFAIK there is no difference in TOASTing varchar and text fields. – dezso Jul 16 '12 at 18:33
  • VARCHAR is purely syntactic sugar for TEXT in Postgres, there is zero difference in storage handling; the compression vs background table storage you mention is done based on the actual length of the data in the column and not on the column metadata. TEXT columns are stored internally as a varlena C struct (which is a variable length array with the first 4 bytes storing the length on create/update) and it is this struct that is optimized based on its length. – cowbert Aug 27 '17 at 20:31
1

The question is specifically whether adding an arbitrary length limit to VARCHAR columns?

To that, the answer is simply "no". There is nothing that can justify adding an arbitrary limit like you would in inferior databases that support varchar(max) or use conventions like varchar(255). However, if the spec addresses a limit, I think the answer becomes much more complex especially on modern versions of PostgreSQL. And, for that, I would lean towards YES.

In my opinion, the limit is a wise-choice if the spec requires it. Especially for more reasonable workloads. If for no other reason then to preserve meta-data.

From my answer here, index performance for CHAR vs VARCHAR (Postgres), where I address the value of meta-data.

If I found a spec that had variable-length text-keys that were meaningful and that I trusted to have a constant max-length, I would use varchar too. However, I can't think of anything that fits that criteria.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.