Skip to content

Latest commit

 

History

History
99 lines (76 loc) · 4.44 KB

README.md

File metadata and controls

99 lines (76 loc) · 4.44 KB
Raw

CrashDB

CrashDB is an ephemeral in-memory NoSQL database for the confident developer.

Features

  • CrashDB is webscale.
  • Data is not persisted between process restarts.
  • Using the wrong HTTP method will crash the database.
  • Getting a key that isn't in the database will crash the database.
  • Using malformed JSON will crash the database.
  • Any system error will crash the database.

For privacy and security, CrashDB does not provide methods that reveal what keys are currently stored in the database.

That's a paddlin'

Build

$ make

CrashDB accepts no command-line options or arguments.

Use

Health check:

$ curl localhost:8080/ping

Get a value from the DB:

$ curl -X GET --data '{"Key": "mykey"}' localhost:8080/

Put a value in the DB:

$ curl -X POST --data '{"Key": "mykey", "Value": "myvalue"}' localhost:8080/

The keys must be strings, but CrashDB supports arbitrary values that can be encoded in JSON:

$ curl -X POST --data '{"Key": "mykey", "Value": {"foo": ["bar"]}}' localhost:8080/

But why?

You have to learn how to improve and debug things somehow.

CrashDB has no shortage of problems. Discover them, fix them, and by improving CrashDB, improve yourself as a developer.

The author recommends doing any or all of the following. Some of those points are aimed at making any single instance of CrashDB less terrible, and some of them are aimed at treating CrashDB like a distributed system.

  • Run sadness.sh. Then Google delve to find out what you're looking at. Become sad that this was not a part of your life until now. (The author notes that this only really works on Linux, as he sadly discovered while in the last throes of pretending that OSX was a decent operating system for doing development. You can bend Docker to your will to make this work on OSX. Find out how!)
  • Improve the error messages to get a better idea of what went wrong. Collect those error messages somewhere so you can search them.
  • Use an issue tracker - any issue tracker - to keep track of what you think is wrong with this thing. A folder with two folders called "open" and "closed" and some text files under version control is a perfectly good issue tracker for beginners.
  • Decide whether to fix the race conditions (you did run sadness.sh, right?) by embracing channels or just using a fucking mutex like everyone in the world apparently does. Reflect on Erlang/Elixir and debate whether you would have this problem there.
  • Find your own things to care about and improve CrashDB in those directions. Realize after weeks of work that some of those things were bad ideas and all that work needs to be thrown away. Be OK with that.
  • Question some of the original design decisions. Is it really necessary to serialize and deserialize the values all the time? Does this add unacceptable overhead for intricate keys? Is it a potential DOS vector? Can you craft key/value pairs that result in arbitrarily bad performance?
  • Profile CrashDB. Get your own torture tests with sample queries that exhibit some of the problems you want to fix, or some of the performance characteristics of the program. Profile before and after fixes to see their impact.
  • Make CrashDB persistent. Figure out how and where to store its data between restarts or crashes. Do you need to worry about data getting corrupted?
  • Add monitoring. CrashDB is a service, so the RED method seems appropriate. Decide whether to use "push" monitoring (à la Graphite) or "pull" monitoring (à la Prometheus).
  • Support alternative serialization formats. JSON is fine for some things (the more you use numbers, the less fine it is), but consider binary formats like Protocol buffers. Use schemas to autogenerate a client for CrashDB for at least one language.
  • Run CrashDB in Docker. Now run two of them at once and setup a master/slave replication chain. Direct writes to the master and reads to the slave. Do service discovery for the outside world. Run three instances at once and have the slaves elect a new master if the original one becomes sad. Find out why Zookeeper is a thing when you try to do any of this or query the cluster from the outside.