Working with Combined Knowledge Varieties inside a Discipline Utilizing Rockset

So… you suppose all of your information in a specific discipline are a string sort, however if you attempt to run your question, you get some errors. Doing extra investigation, it seems to be like you may have some int and undefined sorts as effectively. Bummer…

Despair not! We will really work round this (with out information prep 😉). To recap, in our first weblog, we created an integration with MongoDB on Rockset, so Rockset can learn and [update] the info coming in MongoDB. As soon as the info is in Rockset, we will run SQL on schemaless and unstructured information.

The info:

Embedded content material: https://gist.github.com/nfarah86/ef1cc9da88e56226c4c46fd0e3c8e16e

We have an interest within the release_date discipline: "release_date": "1991-06-07".

The question:

Rockset has a perform referred to as DATE_PARSE(), which lets you flip a string formatted date right into a date object. If you wish to order motion pictures by simply the 12 months, you need to use EXTRACT().

Primarily, when you flip your string formatted date right into a date object, you’ll be able to then extract the 12 months.

At first look, this appears fairly straightforward to resolve— in the event you needed to order all of the film titles by the discharge 12 months, you’ll be able to write one thing like this:

SELECT
      t.title, t.release_date
FROM
    commons.TwtichMovies t
ORDER BY
    EXTRACT(
         YEAR
         FROM
         DATE_PARSE(t.release_date, '%Y-%m-%d')
    ) DESC
;

When working this question, we get a timestamp parsing error:
Error [Query]
Timestamp parse error:

This might imply you’re working with different information sorts that aren’t strings. To examine, you’ll be able to write one thing like this:

SELECT
      t.title, TYPEOF(t.release_date)
FROM
    commons.TwtichMovies t
WHERE 
    TYPEOF(t.release_date) != 'string'
;

That is what we get again:

Now, that we all know what’s inflicting the error, we will re-write the question to discard something that’s not a string sort— proper 🤗?

SELECT
      t.title, t.release_date
FROM
    commons.TwtichMovies t
WHERE
    TYPEOF(t.release_date) = 'string';
ORDER BY
     EXTRACT(
          YEAR
          FROM
              DATE_PARSE(t.release_date, '%Y-%m-%d')
     )DESC
;

WRONG 🥺! This really returns a timestamp parsing error as effectively:

Error [Query]
Timestamp parse error

You are most likely saying to your self, “what the heck.” One case we didn’t take into accounts earlier is that there might be empty strings 🤯- If we run the next question:

SELECT DATE_PARSE('', '%Y-%m-%d');

We get the identical timestamp parsing error again:

Error [Query]
Timestamp parse error

Aha.

How will we really write this question to keep away from the timestamp parsing errors? Right here, we will really examine the LENGTH() of the string and filter out every thing that doesn’t meet the size requirement— so one thing like this:

WHERE LENGTH(t.release_date) = 10

We will additionally TRY_CAST() t.release_date to a string. If the sector worth can’t be was a string, a null worth is returned (i.e. it gained’t error out). Placing this all collectively, we will technically write one thing like this:

SELECT
    t.title,
    t.release_date
FROM
    commons.TwtichMovies t
WHERE
    TRY_CAST(t.release_date AS string) is just not null
    AND LENGTH(TRY_CAST(t.release_date AS string)) = 10
ORDER BY
    EXTRACT(
         YEAR
         FROM
         DATE_PARSE(t.release_date, '%Y-%m-%d')
    )
;

Voila! it really works!

Through the stream, I really wrote a extra difficult model of this question. The above question and the question within the stream are equal. We additionally wrote queries that combination! You may catch the complete breakdown of the session under:

Embedded content material: https://youtu.be/PGpEsg7Qw7A

TLDR: you’ll find all of the sources it’s good to get began on Rockset within the developer nook.