Database connection is not correctly destroyed after knex 0.12.2

[kirrg001]

Hey 👋

I have recently updated knex from 0.12.2 to 0.12.7 and noticed that the database connection is not closed when calling connection.destroy. The result is, that a process doesn't exit.

Here is a snippet:

....
this.connection = knex(...);

return this.connection.transaction(function(txn) {
   return doSomething(txn)
       .finally(function() {
           return self.connection.destroy()
                .then(function () {
                    debug('Destroyed connection');
                });
       });
})

NOTE: debug('Destroyed connection') get's called.

I have tested all knex versions and it starts in 0.12.3.

The code works for me with 0.12.2, but does not with 0.12.3.
Is there anything i can change in my code to get it working with > 0.12.2?

Otherwise, I went through the changes, maybe related to the pool update #1702?

Thank you so much!
Kate

[wubzz]

Based on the changelog I would guess #1731 causes this in some way. Some changes to acquireConnection was done so that connections are properly returned to the pool.

Edit: Testing with postgres, I'm unable to reproduce the issue as described, while switching between versions 0.12.2 -> 0.12.7. Viewing the PG server status, the connections appear to close properly and returns to the pool in both versions. The process does not exit in either version.

const db = require('knex')({...});

db
.transaction((tr) => tr('user').select())
.then(() => db.destroy())
.then(() => console.log('Destroyed'));
//No open connections in server status

[kirrg001]

I have tested with MySQL only. Tested multiple mysql versions.

[kirrg001]

I debugged.

This timeout https://github.com/tgriesser/knex/blob/master/src/client.js#L234 is blocking the process to exit.

[kirrg001]

https://github.com/tgriesser/knex/blob/master/src/client.js#L243

If an error occurs, shouldn't the timeout be cleared?
This solves my issue. I can raise a PR.

[wubzz]

@kirrg001 Ah yes, makes sense the timeout would cause the process not to exit, but it also makes sense why @tgriesser added it. In order to properly release connections back to the pool, the two handlers must be seperate for TimeoutError and pool.acquire, and so a timeout check is inevitable.

I agree clearTimeout should be called when an error occurs as well.

Does your app's process "eventually" exit, or does it idle beyond acquireConnectionTimeout with its default value of 60000 ms?

Unfortunately I don't have an answer to this problem right now. Will try look into this some more.

[kirrg001]

Does your app's process "eventually" exit, or does it idle beyond acquireConnectionTimeout with its default value of 60000 ms?

It exits after the timeout.

My use case is a bit complex and 3 database connections are created, one of these connections can throw an error at the moment if the database does not exist, knex tries to acquire a connection to this target database and fails. That's why clearing the timeout solves this issue for me. Thanks for your fast support here.

[wubzz]

@kirrg001 I see, I thought your app remained idle eventhough all attempts of acquiring connections were successful, but I understand now that's not the case.

I'll merge this in right now, definitely a bug. Thanks for the report!

[wubzz]

@kirrg001 FYI: Pushed 0.12.8

[kirrg001]

Thanks so much 🐩