In the standard setup of Citus, it has one coordinator, so it is the single point of failure and the bottle neck of the system.
To solve these disadvantages, I built a multi-coordinator using Streaming Replication, which is the built-in replication feature of PostgreSQL.
I found an interesting postgres cluster management system using Pacemaker. It is created by GoCardless. They provide a vagrant box, so we can play and learn it easily.
At the end of 2014, I was shocked at the progress of blind deblurring algorithm, so I studied the recent researches in this field.
I read many papers and made a system based on this paper: “Blind deconvolution using alternating maximum a posteriori estimation with heavy-tailed priors”.