Notes from when we went to a DigitalOcean conference and I pulled aside one of the main database people at DigitalOcean to ask him some questions. Fantastic people, fantastic company. 10/10 would host again.

read more
- [ ] Is hatch right for us? 
- [x] Are we going to hit limits on number of tables in our database? Adding about 1500 tables per year. 
- [ ] We haven’t been using DigitalOcean’s monitoring because we are using Debian 7, which it doesn’t support. Can we probably safely upgrade to Debian 8?
- [x] At what point would we need to set up load balancing? Resiliency, then we might want two web servers and get one droplet for the database. 
- [x] At what point would we want to add Memcache and what would it help with?
- [x] How do we monitor whether our CPU is too high and need to upgrade? 
- [x] How do we monitor when certain things are bottlenecking, in particular Apache, MySQL, and even Linux? 
- [x] How do you know when MySQL is slowing you done, slow query log? CPU utilization? 


Limits of MySQL equivalent of MaxClients. Could look at connection pooling. Only matters when SQL is taking a while, could pool 


Private networking. SSL connection, private networking. SSH port foe wat


csampaio@digitalocean.com
Should mention the Hatch program. 




,
Currently SSH 


Maximize RAM for MySQL. Maxbuffersize for MYSQL would increase the amount of RAM being used. 

Doing mysql dump may be doing a lot of IO

Watch IO and other things when doing backup. 

Get certain tables to prefetch in memory and keep them in RAM. 

IO Stat and should not be going back for any reads. 

NYC2 has good. Apache is probably waiting for database. Should have a CPU load. 

Spaces bucket

Debian 8, should be okay. 

Memcache is only for very simple stuff, and that doesn’t change very often. Could look at redis. 


Mike soloman, engineering principles
Persona xtra backups
Swagger
May want to start doing monitoring
Shared MySQL
Load balancing and private networking
May need to add memcache once you get to a lot more users. 
Distributed datastore is for very big
Linux performance governor
Linux hyper threading
Kubernetes 

Load balancing for the API/web service 

Separating the API from the web service, so people can visit the pages but not get to the API
 


She says MySQL is very good.