DESIGN

cache loader

v

config -> main (cache) -> watcher

	/– filters
v v
-------------> walker -> event handlers
-------------> walker -> event handlers
-------------> walker -> event handlers

. . .

Remarks:

main: glues each part together. Loads the cache, maintains a walker thread

pool, and instantiates the watcher.

config: loads all the configurations, like thread pool size, etc.

cache loaders: imports different types of catalog into the cache, mapping the

input catalog format to the in-memory cache data structure.

watcher: registers itself with the kernel notification service (inotify,

etc.). When a change occurs, dispatch it to a walker to handle it.

walkers: each walker is a single thread that handles a range of

directories. Range size for a single walker can be configured. When a change has been detected, invoke the corresponding event handlers.

filters: filters out the files/directories the user is not interested in.

So uninteresting objects will be skipped automatically without invoking the event handlers.

event handlers: multiple event handlers can be registered, each handles a

certain kind of event (file creation, deletion, attribute change, etc.).

object: we don’t need to lock each object.

Because the design guarantees that an object is only accessed by a single walker at a time.