DocStore: Document Database for MySQL at Facebook [pdf]

[ngrilly]

https://www.percona.com/live/mysql-conference-2015/sites/default/files/slides/Facebook%20DocStore%20Percona%202015.pdf

[denma]

Things like MySQL and PHP have bad reputation nowadays but Facebook shows impressive use cases and improves the technology more and more.

[nemothekid → denma]

While tools like MySQL and PHP aren't as bad as the community makes them out to be - Facebook is a terrible example how "impressive" these tools can be considering that Facebook almost entirely rewrote both of those tools (https://github.com/webscalesql/webscalesql-5.6, https://github.com/facebook/hhvm)

[istvan__ → denma]

You can do many things with PHP and MySQL but the real question: is it worth it? I would be interested comparing the TCO (including development time) with platforms like this. I think MySQL is lagging behind Postgres nowadays so I would much rather use that even for storing JSON, even though I do not think that storing JSON in an SQL engine is a particularly good idea, it might work. Our #1 complain against MySQL + JSON was the insert and update time that was significantly higher than the normal insert/update for obvious reasons. If the workload is not update heavy this approach might even work. :)

[aristidb]

If you're not stuck with a huge MySQL legacy installation, Postgresql 9.4's JSONB + json_path_ops index queries seem neater. Or am I missing something?

[istvan__ → aristidb]

I guess Facebook hired a lot of MySQL people and that investment is bigger than the technological one. Also, you are right, Postgres is ahead of MySQL in terms of JSON support and many more.

[frik → aristidb]

MySQL 5.6+ / MariaDB 10 with its InnoDB engine is very popular and suitable for web scale.

PostgreSQL is a good open source alternative for GIS and enterprise needs (Oracle, DB2, MSSQL). Unnerving is a tiny fraction of the Postgres community that acts as trolls.

[sanderjd → frik]

Why exactly is "MariaDB 10 with its InnoDB engine" "suitable for web scale" while PostgreSQL is merely a good "alternative for GIS and enterprise needs"?

I think you may have some interesting information to add, but your comment currently reads like aped talking points, with a dash of accusing someone asking a reasonable question of trolling.

[frik → sanderjd]

The question at the end wasn't there, when I wrote comment. The default postgresql config is rather conservative. It is more an open source Oracle with many of its (former) unique features. PL/pgSQL resembles Oracle's PL/SQL procedural language. [1]

Most people simply don't understand MySQL. MySQL is unique in as it support many database engines [4], it's like a common SQL layer on top of dozens of database engines. MySQL had a bad reputation because of its old MyISAM engine [2], the former default engine (one of many). The current default database engine is InnoDB [3] which is very fast and is what Facebook and many other huge web sites use.

Instead of repeating myself and others, MySQL+memcached is quite popular, check out the highscalability website [5].

[1] http://en.wikipedia.org/wiki/PostgreSQL , http://stackoverflow.com/questions/12622524/postgresql-9-2-1... , http://stackoverflow.com/questions/110927/would-you-recommen... , etc.

[2] http://en.wikipedia.org/wiki/MyISAM

[3] http://en.wikipedia.org/wiki/InnoDB

[4] http://en.wikipedia.org/wiki/Comparison_of_MySQL_database_en...

[5] https://www.google.com/?gws_rd=ssl#q=site:highscalability.co...

[aristidb → frik]

The question at the end was always there.

[pokstad]

The queries remind me of UnQL. It was introduced by the Couchbase and SQLite teams as a way to standardize NoSQL queries: http://unql.sqlite.org/index.html/doc/tip/doc/syntax/all.wik...

[ingenthr → pokstad]

At least for Couchbase thus was replaced by N1QL: http://docs.couchbase.com/developer/n1ql-dp4/n1ql-intro.html

[ingenthr → pokstad]

At least for Couchbase, this was replaced by N1QL

[joseakle]

This reminds me of FriendFeed´s Schemaless MySQL.

https://backchannel.org/blog/friendfeed-schemaless-mysql

They were acquired by FB.

[hanlec → joseakle]

I came here to post just this. Bret Taylor, one of the co-founders of FriendFeed, was CTO @ Facebook after the acquisition.

Also besides FBSON, most of what I've read in those slides looks like PostgreSQL's JSON support:

[1]: JSON type http://www.postgresql.org/docs/9.4/static/datatype-json.html

[2]: JSON functions http://www.postgresql.org/docs/9.4/static/functions-json.htm...

[snowmaker]

This is awesome. We would find it super useful.

Does anyone know if Facebook is planning to open source it, or if not, whether something similar exists?

[ngrilly → snowmaker]

I understand this is or will be a part of WebScaleSQL [1]. But WebScaleSQL does not provide binaries. You have to build it yourself.

An alternative is the JSON Labs Release of MySQL 5.7.7 [2]. The differences are explained in the slides, at the end.

Another great alternative is using PostgreSQL that supports JSON columns stored in a binary format since version 9.4 [3].

That said, I really like the approach of Facebook with DocStore, because of its use of "document paths", instead of functions.

[1] http://webscalesql.org/

[2] http://mysqlserverteam.com/json-labs-release-overview/

[3] http://www.postgresql.org/docs/9.4/static/datatype-json.html

[mandeepj → snowmaker]

This looks similar to mongoDB