rackerlabs / cache-busters

cache_busters

cache_busters is an automatic cache invalidation system. It works by listening on a database's (such as PostgreSQL or MySQL) replication protocol, and invalidating caches (such as memcached or redis) based on dirtied data.

Basic architecture:

• Can be configured via either a simple configuration file (for individual deployments) or via a REST API (for multi-tenanted deployments).

• A coordinator is responsible for making sure that at all times there is a live cache_busters node which is handling invalidation for a given database. It's ok if there are multiple running.

• In a multi-tenanted deployment, configuration data is stored in a database.

• Inside each cache_busters node, connections to a cache are stored in a variable sized pool and invalidated by LRU.

• It never tries to repopulate a cache, only flush.

• An example configuration file (need to figure out the REST equivalent):

  caches:

  • "127.0.0.1"

  database:

  host: "127.0.0.1" user: "root"

  on_update:

  test_table:

  • "{id}-{name}"

  animals:

  • "{id}-{name}"

TODO

• Figure out what sort of cooperation we need to pretend to be a slave.
• In multi-tenanted mode how does the networking for all this work out.