[YarickR2]

This is optimizing edge cases. Handshake is done once per execution , and it's constant time , so "connect + select 1" should be measured as "connect", then "select 1". Other than that, 5ms latency is too high anyway for modules doing hundreds of small requests, so you better have read replicas real close to your code, and shard your masters.

[mattrobenolt]

Maybe in your case, but not always. But I do explicitly call out that I intentionally wanted a test of a "cold start". This is extremely relevant for short lived applications and processes. Think PHP, and serverless environments, etc.

The other tests are measuring already a warmed up connection.

There's also reason why I intentionally coupled "connect + select 1" as the test, because I wanted to make it as close of a comparison as possible. If it was simply a "connect", our HTTP API would be even more favorable since connecting doesn't do authentication or anything like that like the mysql protocol does.

[tookledoodle]

You call out PHP in both the blog post and this comment, which is interesting, because persistent pooling has been a part of PHP (and advised) for years.

Which leaves serverless and scripts (your other example from the blog post). Which, let’s be honest, are both edge cases at this point in time. Maybe that’ll change, but today it’s true.

Twenty year SRE here backing up the person you’re dismissing: you’re optimizing an edge case. Literally step one of operationalizing every system in existence is burying your DB behind a pooler. 100ms off a connect call in a script is not useful. The serverless improvement has some potential, but one would be forgiven for asking why you’d use an environment which doesn’t let you speak network protocols you’d like to speak.

[orware]

I'd be curious to hear about experiences with persistent pooling usage within PHP, since I don't feel that I've heard about it being advised as much myself over the years, but that could perhaps be due to old misperceptions about it.

I know there is the "Persistent Database Connections" section of the PHP manual and the mysqli extension within PHP supports connection pooling / persistent connections, but in my own experiences I've rarely seen them utilized, especially by the bigger open source projects out there such as WordPress, which has an 8 year old enhancement topic on the subject: https://core.trac.wordpress.org/ticket/31018. Putting your database behind a pooler, like ProxySQL let's say, is another option as the level of sophistication for a company/application increases, but most typical PHP setups I've used don't have that immediately available.

I've generally been under the impression that most projects/applications don't use the built-in pooling features for some of the reasons discussed in the link above, leading to those applications being more impacted by lengthier connection times due to a new connection being created at the beginning each request and then closed at the end of the request.

Now I'm inclined to experiment a bit with the built-in mysqli pooling feature though since it would seem a worthwhile feature for developers to experiment with more if it would lessen the connection time impact for each PHP request, particularly for databases that are further away and require secure connections.

Shaving off 100ms for a connection would be significant for most PHP users if they are currently having to open fresh connections on each request, especially if they were previously used to connection times of < 1ms when connecting to a local MySQL database.

[aarondf]

> because persistent pooling has been a part of PHP (and advised) for years.

It may be advised, but I can assure you that it's not very common! I would guess that the vast, vast majority of PHP applications are _not_ pooling their connections. Especially when it comes to PHP hosted on Lambda, which is surprisingly a non-trivial amount of applications at this point.

[mattrobenolt]

I'm not a PHP expert, so I don't know the landscape there fully. I do know our customer complaints and can say people care about cold start times in the PHP space and others.

So while it may be an edge case for you, it's not for others. It also doesn't discredit any of the other testing that doesn't focus on cold starts.

Edit since you edited yours after I posted:

I'm not going to argue the merits of what platforms people choose and it's not really our position as PlanetScale to do that. We serve our customers.

[tookledoodle]

No, it objectively remains an edge case, no matter whether it is for me or you or the person you’re replying to or whatever. Your customers doing inefficient things does not change the fundamentals of the situation. And the other testing that didn’t focus on cold starts looked like basically a non-result looking at the graphs, so I didn’t bother talking about it. I can, though, with the understanding that it will strengthen my point.

[kamikaz1k]

> fundamentals of the situation

On what grounds? Why are you able to unilaterally decide what the situation is?

[mattrobenolt]

Sorry about that.

[jeffbee]

TCP connect is not constant time. It stands a surprisingly large chance of failing and waiting TWO SECONDS to start again.