SQL: Select records based on comparison of two most recent associated records - mysql

Question

Let's say we have a person table and survey table. survey is a set of attributes collected from a person at some point in time. Let's say survey has columns address and marriage_status
How do I select all persons whose address or marriage status has changed in the last survey?
Here's how I would write it if MySQL were able to magically interpret my intention:
SELECT *
FROM person
JOIN
(SELECT *
FROM survey
GROUP BY survey.person_id
ORDER BY survey.timestamp DESC
LIMIT 2 EACH) -- of course this part doesn't actually work. Trying to get last 2 records per person
surveys
ON surveys.person_id = person.id
WHERE surveys[0].address != surveys[1].address
OR surveys[0].marriage_status != surveys[1].marriage_status;
OR
SELECT *
FROM person
JOIN
(SELECT MOST RECENT survey FOR EACH person) latest_survey
ON latest_survey.person_id = person.id
JOIN
(SELECT SECOND MOST RECENT survey FOR EACH person) previous_survey
ON previous_survey.person_id = person.id
WHERE latest_survey.address != previous_survey.address
OR latest_survey.marriage_status != previous_survey.marriage_status;
This seems like a relatively straightforward query, but it's driving me crazy. I suspect I have tunnel vision and I'm not approaching this the right way.
EDIT: I am on MySQL v5. Based on the first couple answers, it seems like this might be the time to migrate to v8 (among other reasons)

Answer 1

So here's how I ended up doing it. It's a little long, but I think it's pretty straightforward? This felt amazing to get working.
(Note that underscores are used as prefixes in table aliases to help keep track of subquery depth)
SELECT person.*
FROM person
JOIN (
-- Join full survey data against each 'most recent' survey timestamp
SELECT s1.*
FROM survey s1
JOIN (
-- get most recent timestamp for each person
SELECT _s1.person_id, MAX(_s1.timestamp) timestamp
FROM survey _s1
GROUP BY person_id
) latest_surveys
ON latest_surveys.person_id = s1.person_id and latest_surveys.timestamp = s1.timestamp
) latest
ON latest.person_id = person.id
JOIN (
-- Join full survey data against each 'SECOND most recent' survey timestamp
select s2.*
from survey s2
JOIN (
-- to get SECOND most recent survey timestamp, do similar query, but exclude latest timestamp
SELECT _s2.person_id, MAX(_s2.timestamp) timestamp
FROM survey _s2
JOIN (
-- get most recent timestamp for each person (again)
SELECT __s2.person_id, MAX(__s2.timestamp) timestamp
FROM survey __s2
GROUP BY person_id
) _latest_surveys
-- Note the *NOT* equal here
ON _latest_surveys.person_id = _s2.person_id and _latest_surveys.timestamp != _s2.timestamp
GROUP BY _s2.person_id
) previous_surveys
ON previous_surveys.person_id = s2.person_id and previous_surveys.timestamp = s2.timestamp
) previous
ON previous.person_id = person.id
WHERE latest.address != previous.address
OR latest.marriage_status != previous.marriage_status;

Answer 2

Analytic functions make your question much more tractable. If you are not yet using MySQL 8+, then now would be a good time to upgrade. Assuming you are using MySQL 8+, we can try:
WITH cte AS (
SELECT *, ROW_NUMBER() OVER (PARTITION BY p.id ORDER BY s.timestamp DESC) rn
FROM person p
INNER JOIN survey s ON p.id = s.person_id
)
SELECT id
FROM cte
GROUP BY id
HAVING
MAX(CASE WHEN rn = 1 THEN address END) <> MAX(CASE WHEN rn = 2 THEN address END) OR
MAX(CASE WHEN rn = 1 THEN marriage_status END) <> MAX(CASE WHEN rn = 2 THEN marriage_status END);
The above query uses a pivot trick to isolate the latest, and second latest, addresses and marriage statuses for each person. It retains person id values for those whose latest and second latest addresses or marriage statuses are not identical.

Answer 3

This might be how you can achieve that:
SELECT *
FROM person
JOIN (
SELECT *,
MAX(survey_date) latest_survey,
SUBSTRING_INDEX(SUBSTRING_INDEX(GROUP_CONCAT(survey_date ORDER BY person_id, survey_date ASC),',',-2),',',1) previous_survey,
SUBSTRING_INDEX(GROUP_CONCAT(address ORDER BY person_id, survey_date ASC),',',-1) curadd,
SUBSTRING_INDEX(SUBSTRING_INDEX(GROUP_CONCAT(address ORDER BY person_id, survey_date ASC),',',-2),',',1) prevadd,
SUBSTRING_INDEX(GROUP_CONCAT(marriage_status ORDER BY person_id, survey_date ASC),',',-1) curms,
SUBSTRING_INDEX(SUBSTRING_INDEX(GROUP_CONCAT(marriage_status ORDER BY person_id, survey_date ASC),',',-2),',',1) prevms
FROM survey GROUP BY person_id
HAVING curadd != prevadd OR curms != prevms) A
ON person.id=A.person_id;
Using GROUP_CONCAT and SUBSTRING_INDEX to combine the data value then separate it again and using those to compare at the end. I know there are a bunch of ways to achieve without all these, like your second example is something that I think can be done but when I think about it, it's going to be a very long query. This query however, since you're not using MySQL 8+ is much shorter but the performance of this query is a concern especially on a large table.

Answer 4

It is not given, but I hope you have at least MySQL 8 or similar to have ability to use Common Table Expression. It can simplify the complex query.
The trick part is getting survey records #1 and #2 for each user. I will do it this way: see cte1 and cte2 definition
WITH
cte1 AS (
SELECT MAX(x1.id) AS id, x1.person_id
FROM survey x1
GROUP BY x1.person_id),
cte2 AS (
SELECT MAX(x2.id) AS id, x2.person_id
FROM survey x2
JOIN cte1 ON cte1.person_id = x2.person_id
AND cte1.id > x2.id
GROUP BY x2.person_id)
SELECT
p.*,
s1.address, s2.address address2,
s1.marriage_status, s2.marriage_status marriage_status2
FROM person AS p
JOIN (
cte1 JOIN survey s1 ON s1.id = cte1.id
) ON cte1.person_id = p.id
JOIN (
cte2 JOIN survey s2 ON s2.id = cte2.id
) ON cte2.person_id = p.id
WHERE
(s1.address <> s2.address)
OR (s1.marriage_status <> s2.marriage_status)
https://www.db-fiddle.com/f/hLwdHiZin4MkdUZ4aBz67H/2
Update: Thanks to Ian, I replaced MIN to MAX to get recent records