neurodrone / crdt Goto Github PK

I am working on Synchronisation for online, offline apps.
CRDTS are a core part of that and great that this lib is here.

Google's vandium implements CRDT's btw for a database called Syncbase.
I am sure you know the project. Yesterday was announced that the project is closing down.
The CRDT related problems they solves are impressive.
Have you looked a this ?

In file g_counter.go line 22, the identity of replica is ident: uuid.NewV4().String() but uuid.NewV4() returns a tuple with of format (UUID, error).

As I saw this reference is used in other places as well!

Currently, the implemented LWW-e-set CvRDT ensures that Adds always wins over Removes. The ability to control this should rely on the value of BiasType key that is passed in while creating a new object of type LWWSet.

We save all the added timestamps for each LWW-e-set element within an array. This makes it easier to add multiple entries that have the same timestamps.

Ideally, only one entry across multiple duplicate timestamps should exist.

https://github.com/neurodrone/crdt/blob/master/lww_e_set.go#L68-L74 the results for each of the switch cases should be swapped.

Instead of resolving conflicts in the read-path, we can try to resolve them in the write-path.

By adding a [element, time, contains?] version we can add or remove an element from the set depending on the timestamp for which op is higher.

When times for conflicting ops (both add and remove) are same then we pick the value for contains? depending on the biasValue that is passed in.

In order to make persistence an easier problem to solve we should be able to serialize every data-structure currently present in the repository. The data-structures should implement json.Marshaler interface to suffice a baseline scenario.