Droplet Archiver

The Droplet Archiver is a simple Node.JS based application that can be used as a central storage point for droplet packages within Cloud Foundry.

In a stock Cloud Foundry application, droplets are stored and handled by the Cloud Controller, however this role can be more storage intensive over time and the built in mechanisms don't allow for replication of application bits.

At PaaS.io, we've been working on breaking up the Cloud Controller into multiple components, one of which is the droplet archiver.

Features

Simple and lightweight process
Optimized for concurrency and stream
Self-replicates droplets to other configured nodes

Configuration

The configuration is a simple JSON file named config.json:

{
  "port" : 3000,
  "droplet_path" : "./droplets",
  "credentials" : {
    "username" : "user",
    "password" : "pass"
  },
  "sync_hosts" : [ "127.0.0.1:3000", "127.0.0.1:4000" ]
}

The configuration is pretty much self explanitory. The sync_hosts is an array of other host:port servers that should have droplets replicated to. It is ok to put its self in the list, as it will recognize it already has the droplet and continue on.

Usage within Cloud Foundry

This is mainly TBD at this point. The Cloud Controller needs to be updated in a few places, mainly in app/models/app_manager.rb in new_message and download_app_uri, as well as app/controllers/staging_controller.rb in upload_droplet_uri.

Endpoints

The application responds to three simple endpoints:

GET /droplets/:appid/:sha1 - Download a droplet
POST /droplets/:appid/:sha1 - Upload a droplet
POST /sync/:appid/:sha1 - Used internally to sync droplets between nodes

Whenever a droplet is uploaded, the file will be validated against the SHA1 given and will be deleted if it doesn't match.

Replication

Droplets are lazily replicated between other hosts after they are uploaded. This is done after the request is already responded to. It uses a separate endpoint as a simple means to ensure a replication storm isn't created. If a host already has the specified droplet, it will simply return. With this, a node can have itself in the sync_hosts list and be ok. It will make a request to iself, but will just quickly return.

Currently, if it fails to replicate to a node it will just continue on. It has no reconciling function to come back, or to ensure consistency across all nodes. This will be coming soon.

droplet-archiver

README.md