DigitalOcean DO managed database -worth migrating to?

Anyone using Digital Oceans managed SQL? How is it under real load?

I have a test cluster up for a test site, it works...

Right now I'm still doing the monolith all on one server and just upsizing as it gets tight...
presently on a linode dedicated 32Gb ram with zero complaints. Everything is just about fitting in, but today triggered a 80% alarm on ram... traffic is picking up again.

I switched from dynamic to static when I moved to the dedicated cpu, running 120 php procs. I should probabbly if I'm reading my queue length right need more.

Anyhow, I've been toying with moving the db off. Although you get a higher chance of downtime with two nodes, that would ( touch wood ) still be quite low and bearable for a forum site.

So I was toying with moving to DO for the managed mysql. Partly so I don't have to worry about it; sod the cost. Partly also for the 'real' private network, although linode is meant to be bringing that this year.

However my mysql instance needs around 14gb or ram for the innodb tables/index, which is right on the max DO managed db size, unless I'm misunderstanding modern mysql innodb sizing. I also can't see anyway to 'tune' innodb sizing, buffer size etc. Do I trust they look in on this... I could possibly reduce this memory impact, if I get search offloaded to elastic. So 3 nodes

I know it's a heck of a price, but I'm not of the mind to do a mysql cluster. Because If I did that I'd be daft not to do a php a:b pair too. But thats overkil for what it is...

Thoughts?

---

ps I'm also a bit wary of DO's ddos reactions of null routing, but that might be bad internet blood.

 

more commands and data to parse

I'm now hovering around 75% of ram so won't push much more on to it.

Code:

Every 2.0s: ...  Wed Sep 16 14:50:08 2020

Processes active: 18, idle: 82, Requests:
 2770091, slow: 0, Traffic: 5.9req/sec
pool:                 www
process manager:      static
start time:           09/Sep/2020:22:36:2
7 +0000
start since:          576821
accepted conn:        2770138
listen queue:         0
max listen queue:     51
listen queue len:     511
idle processes:       84
active processes:     16
total processes:      100
max active processes: 207
max children reached: 0
slow requests:        0

It's quite quiet at the moment. Kind of wish there was a way to capture the peak, nixstats quite handy for that in a way.

I've used them, looked at the numbers, then went for a butt dyno.

Setting up on upcloud. Would of gone vultr; possibly, but supply is nonexisteant and thats a concern if I wanted to experiment and to consolidate.

There are more costs in upcloud $ for $, e.g. for a private network, they have a private utility for all vps's on account, to add another it's $5. A firewall is an addiitonal 5$, backups are a bit steeper, but choices like weekly, monthly or year retentions. There is also a scary zero balance termination bit in the terms. Hope the credit card never fails

So to the question of is a managed db worth it... Given I'd be almost doubling my hosting cost, just to bump the database to a standalone 'manged', maybe. It would be closer tripple with failover, which is what I was attracted to, too save setting/managing it. Thats quite a sum and certianly a pause for thought, it would easiy cover the cost of a 'on call' sys admin.

Although this is my first multibox setup in a very long time; I've been spoilt by the likes of heroku and vertical box scaling... first under centmin too, so it's raised questions for me about csf config That and reminded me I'm not a sysadmin; still a dangerous one, a rookie; I can break things and not always know how to fix them

Code:

#csf.allow
tcp|in|d=3306|s=10.x.x.x

I was looking for a way to; for the db nodes, ignore all from eth0; public, allow from eth1; private, then allow port 3306 on eth1 etc rather than each ip on it's own... although being explicit isn't a problem.

But then I see all the allow from cloudflare which will blow past any port restrictions from their ips.

My TCP_IN rule for permit is empty on both web and sql sets of nodes. So kinf oa TCP_IN for each eth really, but can't see anything in the csf docs that leaps out.

Fun days