Fix all usecases where deadlocks can happen (for MySQL versions above 5 and MariaDB)

[petersg83]

What does it do?

The same query than the one used for MySQL 5 (using a temporary table) is now used also for MySQL 8 and MariaDB

Why is it needed?

Deadlock issues found with the custom code for MySQL 5 were also found on the code used for MySQL 8 and MariaDB.
The custom code for MySQL 5 was recently fixed in #15034.
Since the custom code for MySQL 5 also works for MySQL 8 and MariaDB, it becomes now the default code for MySQL 5, MySQL 8 and MariaDB

How to test it?

Follow the steps described in #14937 by using MySQL 8 or MariaDB

Related issue(s)/PR(s)

fix #14937

[codecov]

Codecov Report

Base: 59.88% // Head: 59.88% // Increases project coverage by +0.00% 🎉

Coverage data is based on head (aef49c5) compared to base (2cdd96f).
Patch coverage: 4.16% of modified lines in pull request are covered.

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #15143   +/-   ##
=======================================
  Coverage   59.88%   59.88%           
=======================================
  Files        1348     1348           
  Lines       32800    32799    -1     
  Branches     6255     6255           
=======================================
  Hits        19642    19642           
+ Misses      11303    11302    -1     
  Partials     1855     1855           

Flag	Coverage Δ
back	50.09% <4.16%> (+<0.01%)	⬆️
front	64.33% <ø> (ø)
unit_back	50.09% <4.16%> (+<0.01%)	⬆️
unit_front	64.33% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...e/database/lib/entity-manager/regular-relations.js	10.09% <4.16%> (+0.27%)	⬆️
...er/services/schema-builder/content-type-builder.js	8.52% <0.00%> (-0.14%)	⬇️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[nathan-pichon]

I have added some comments, but I have not yet been able to test the branch.

[nathan-pichon]

I guess we can split this in multiple functions to improve the readability, but looks good enough to be merged.

[Marc-Roig]

Agree, and I find it a little bit difficult to follow when having multiple variables to keep track.

Maybe having a more abstract method such as the following could help:

if (hasInverseOrderColumn(attribute) && !isEmpty(inverseRelIds)) {
  update.push("?? = b.inv_order", inverseOrderColumnName);
  select.push(
    "ROW_NUMBER() OVER (PARTITION BY ?? ORDER BY ??) AS inv_order",
    inverseJoinColumn.name,
    inverseOrderColumnName
  );
  where.push(
    `?? IN (${inverseRelIds.map(() => "?").join(", ")})`,
    inverseJoinColumn.name,
    ...inverseRelIds
  );
}

So then you would do:

      await db.connection
        .raw(
          `UPDATE ?? as a
              SET ${update.getQuery().join(', ')}
              FROM (
                SELECT ${select.getQuery()join(', ')}
                FROM ??
                WHERE ${where.getQuery()join(' OR ')}
              ) AS b
              WHERE b.id = a.id`,
          [joinTableName, ...update.getBindings(), ...select.getBindings(), joinTableName, ...where.getBindings()]
        )
        .transacting(trx);

[Marc-Roig]

I have not managed to replicate the issue using MySQL 8 nor MariaDB . I followed the steps described in #14937 . Could we look at this together some point?

It also interest me to see if the transactions PR #14389 introduces deadlocks too.