A simple, persistent variable library
Switch branches/tags
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
project
src
.ensime
.gitignore
LICENSE
README.markdown

README.markdown

Clipping: Simple, Transparent Data Storage

Clipping is a simple, flexible framework for declaring containers whose values should make a best effort to be persistent between runs of your software.

Sometimes, you want a giant object serialization and persistence framework with ACID properties, the ability to stream data, and versioning. If you want those things, Clipping is not for you.

Clipping does not enforce any explicit semantics on the strategy for storage. Currently, its implementations use standard Java serialization strategies. Clipping's architecture is such that it is trivial to switch between various storage mediums, and to provide any sort of (simple) persistence semantics you might deem fit.

BankSimple uses Clipping to manage monotonically increasing date-based cursors on streaming data which may contain repetition.

Using A Clipping

First, declare a clipping instance. In this case, we're using a reentrant ConcurrentDiskVar:

val decisionStore = new ConcurrentDiskVar[Boolean](false, "ex1", "/store")

You may then use Clipping's read or apply method to read the value. The decisionStore will attempt to read the value from storage; if it fails, it will provide the default. In this case, the following code:

if( decisionStore() ) { println("yes") } else { println("no") }

attempts to read the file "/store/ex1.clipping" and obtain a boolean value. If that operation succeeds, its value will be returned. If that operation fails, the default value will be used.

Once the clipping instance has consulted its backing store, it will not consult it on reads during its lifetime. To write a new value to a clipping, use either the write or << methods, which are identical in functionality:

decisionStore << true
if( decisionStore() ) { println("yes") } else { println("no") }

This code will always print "yes" because the value cached in the decisionStore will immediately update, assuming there are no other threads with references to decisionStore altering its value.

Finally, sometimes it is useful to predicate a value update operation on a logical expression. Clipping supports a writeIf method that allows one to predicate a write in such a fashion:

val dateStore = new SynchronizedDiskVar[DateTime](DateTime.now, "ex2")
val timeToWrite = DateTime.now
dateStore.writeIf((d) => d.isBefore(timeToWrite)) { timeToWrite }

In this case, the second argument to writeIf is only evaluated and stored only if timeToWrite occurs after the currently stored value.

Current Implementations

Clipping's implementation separates the actual encoding and storage (a PersistingStrategy) of values from the strategy used to order and execute the writes and reads (a ManagementStrategy). Currently the following complete implementations exist:

  1. SynchronizedDiskVar: This implementation writes values to disk using standard Java serialization. When writes occur, they occur within the calling thread. This strategy is reentrant and has a slightly lower overhead than ConcurrentDiskVar, but if many threads are writing to the variable it will cause unnecessary lock contention.
  2. ConcurrentDiskVar: Like SynchronizedDiskVar, it writes values to disk using Java serialization. Unlike SynchronizedDiskVar, it uses a ReadWriteLock and tasks a separate thread with writes. This is probably the preferred implementation for most cases.

Both strategies use OnDiskPersistingStrategy, which in turn uses the standard write-then-move technique for writes.

Future Work

  1. We'd like more storage strategies. Redis seems like a likely candidate.
  2. We'd like more serialization methods.
  3. We'd also like a pony.