3 ms·

Two reasons: * I expect a larger load in the future, * I want multiple nodes not just for speed, but mostly for data replication.

by jwr 10y ago

ruslan_talpa 10y ago
* you are solving problems you don't have * you can have replicas with postgres also even if you do end up getting the load you are hoping for (though i doubt you will have 1.5 million qps), reads are easily scalable with replicas, and writes are also possible with sharding or tools from CitusData
- ruslan_talpa 10y ago
  I don't have of the top of my head other examples but i know PG is used by yandex.ru which is like the google of russia, and they migrated from oracle and they are very happy, so if it works for them i think it will work for you :)
  - rixed 10y ago
    Don't they use clickhouse instead?
    - ruslan_talpa 10y ago
      they probably use a lot of tools for a lot of reasons but they do use postgres https://www.pgcon.org/2016/schedule/attachments/426_2016.05.19%20Yandex.Mail%20success%20story.pdf https://www.pgcon.org/2016/schedule/attachments/426_2016.05.... https://www.youtube.com/watch?v=-SS4R1sFH3c https://www.youtube.com/watch?v=-SS4R1sFH3c
- williamstein 10y ago
  This is precisely the very painful lesson I learned - don't solve problems you don't have... (Blog author)
- jazoom 10y ago
  He/she was also not having problems with RethinkDB, so why "solve" that by using Postgres? Disclaimer: I'm also happy running RethinkDB in a cluster of 3. It has been rock solid stable for over 2 years.
true_religion 10y ago
It's much easier to write in a new database in the future, than it is to deal with greater complexity initially. I find that in the first phases, I often don't know what kind of load is going to be where, so I simply use a relational database since they give excelent data security, and okayish performance for most things under low load.
- davidbanham 10y ago
  I think the inverse is actually the case here. If you need changefeeds, use Rethink up front rather than write your own implementation. When you hit a problem with its performance, then think real hard about your domain and see if it's possible to replicate the safety guarantees it gives you on another DB.

Two reasons: * I expect a larger load in the future, * I want multiple nodes not just for speed, but mostly for data replication.

by jwr 10y ago

Two reasons: * I expect a larger load in the future, * I want multiple nodes not just for speed, but mostly for data replication.

ruslan_talpa 10y ago

* you are solving problems you don't have * you can have replicas with postgres also even if you do end up getting the load you are hoping for (though i doubt you will have 1.5 million qps), reads are easily scalable with replicas, and writes are also possible with sharding or tools from CitusData

ruslan_talpa 10y ago
I don't have of the top of my head other examples but i know PG is used by yandex.ru which is like the google of russia, and they migrated from oracle and they are very happy, so if it works for them i think it will work for you :)
- rixed 10y ago
  Don't they use clickhouse instead?
  - ruslan_talpa 10y ago
    they probably use a lot of tools for a lot of reasons but they do use postgres https://www.pgcon.org/2016/schedule/attachments/426_2016.05.19%20Yandex.Mail%20success%20story.pdf https://www.pgcon.org/2016/schedule/attachments/426_2016.05.... https://www.youtube.com/watch?v=-SS4R1sFH3c https://www.youtube.com/watch?v=-SS4R1sFH3c
williamstein 10y ago
This is precisely the very painful lesson I learned - don't solve problems you don't have... (Blog author)
jazoom 10y ago
He/she was also not having problems with RethinkDB, so why "solve" that by using Postgres? Disclaimer: I'm also happy running RethinkDB in a cluster of 3. It has been rock solid stable for over 2 years.

true_religion 10y ago

It's much easier to write in a new database in the future, than it is to deal with greater complexity initially. I find that in the first phases, I often don't know what kind of load is going to be where, so I simply use a relational database since they give excelent data security, and okayish performance for most things under low load.

davidbanham 10y ago
I think the inverse is actually the case here. If you need changefeeds, use Rethink up front rather than write your own implementation. When you hit a problem with its performance, then think real hard about your domain and see if it's possible to replicate the safety guarantees it gives you on another DB.