It would be interesting to know, related to the performance section, what kind of discs they are using in the baremetal Postgres servers. I could imagine they are some advanced SSD disks. When I was working at CERN I witnessed changing the discs for the storage for Ceph
In the ordering queues if the query taking the most amount of resources is constantly at the end of queue. How would it run if the queue keeps on getting populated?
It would be interesting to know, related to the performance section, what kind of discs they are using in the baremetal Postgres servers. I could imagine they are some advanced SSD disks. When I was working at CERN I witnessed changing the discs for the storage for Ceph
yes, indeed. But I couldn't find these details.
I'll post if I find something on their blog, thank you.
Learned some exciting features of postgres
awesome, same here.
I learned some new things while writing this post.
You touched a lot of ground here. I like the following points:
TCP Vegas is a proactive algorithm that uses delayed instead of dropped packets as a metric for throttling.
Ordering Queries based on query complexity and priority. If the system is under load - drop complex and non priority queries.
Separation of concern. Postgres is only responsible for storing and queries while smart proxy sits in the front.
thanks
Very interesting! I particular like the how they ensured availability. Never heard of the term 'chaos testing' good to know that such concept exists.
Thanks for another great post, NK!
you're welcome
Thanks for this cool post NK.
One thing that would be a great addition to all your posts: explicitly mention sources and link to them when they're available.
I think a lot of people - including myself - would like to find more details in some specific cases.
thanks for the feedback.
I put them at the bottom of each post: https://newsletter.systemdesign.one/i/140362895/references
I'll add a "references" header to call them out explicitly in future posts.
My bad, I completely missed it. Indeed some more visibility could be useful then!
I think some other people also complained about it, so I updated each post with "References" header.
Let me see if it improves the situation with future posts. Otherwise I need to find a better format for the newsletter.
In the ordering queues if the query taking the most amount of resources is constantly at the end of queue. How would it run if the queue keeps on getting populated?
Where did you get this information? Is there an original post from cloudflare itself?
it's in the references section at the bottom of the post.