Simple Object Storage

This Simple Object Storage (SOS) project is a HTTP-based object-storage system which allows files to be uploaded, and later retrieved via HTTP.

Files can be replicated across a number of hosts to ensure redundancy, and increased availability in the event of hardware failure.

• The design of the system.
• Scaling to large numbers of objects.
• How replication works.
• The APIs we present, both internal and private.

Installation

There are two ways to install this project from source, which depend on the version of the go version you're using.

If you just need the binaries you can find them upon the project release page.

Source Installation go <= 1.11

If you're using go before 1.11 then the following command should fetch/update the project and install it upon your system:

 $ go get -u github.com/skx/sos

Source installation go >= 1.12

If you're using a more recent version of go (which is highly recommended), you need to clone to a directory which is not present upon your GOPATH:

git clone https://github.com/skx/sos
cd sos
go install

Overview

You can read the design overview for more details, but the core idea behind the implmentation relies upon the notion of a "blob server" - which is a very simple service which provides only the following simple primitives:

• Store a particular chunk of binary data with a specific name.
• Given a name retrieve the chunk of binary data associated with it.
• Return a list of all known names.

The public API is built upon the top of that primitive, and both are launched via the same command sos, by specifying the sub-command to use:

 $ ./sos blob-server ...
 $ ./sos api-server ...

Here the first command launches a blob-server, which is the back-end for storage, and the second command launches the public API server - which is what your code/users should operate against.

If you launch sos with no arguments you'll see brief details of the available subcommands.

Quick Start

In an ideal deployment at least two hosts would be used:

• One host would run the public-server.
  • This allows uploads to be made, and later retrieved.
• Each of the two hosts would also run a blob-server.
  • The blob-servers provide the actual storage of the uploaded-objects.
  • The contents of these are replicated out of band.

We can simulate a deployment upon a single host for the purposes of testing. You'll just need to make sure you have four terminals open to run the appropriate daemons.

First of all you'll want to launch a pair of blob-servers:

$ sos blob-server -store data1 -port 4001
$ sos blob-server -store data2 -port 4002

NOTE: The storage-paths (./data1 and ./data2 in the example above) is where the uploaded-content will be stored. These directories will be created if missing.

In production usage you'd generally record the names of the blob-servers in a configuration file, either /etc/sos.conf, or ~/.sos.conf, however they may also be specified upon the command line.

We'll then start the public/API-server ensuring that it knows about the blob-servers to store content in:

$ sos api-server -blob-server http://localhost:4001,http://localhost:4002
Launching API-server
..

Now you, or your code, can connect to the server and start uploading/downloading objects. By default the following ports will be used by the sos-server:

service	port
upload service	9991
download service	9992

Providing you've started all three daemons you can now perform a test upload with curl:

$ curl -X POST --data-binary @/etc/passwd http://localhost:9991/upload
{"id":"cd5bd649c4dc46b0bbdf8c94ee53c1198780e430","size":2306,"status":"OK"}

If all goes well you'll receive a JSON-response as shown, and you can use the ID which is returned to retrieve your object:

$ curl http://localhost:9992/fetch/cd5bd649c4dc46b0bbdf8c94ee53c1198780e430
..
$

NOTE: The download service runs on a different port. This is so that you can make policy decisions about uploads/downloads via your local firewall.

At the point you run the upload the contents will only be present on one of the blob-servers, chosen at random. To ensure your data is replicated you need to (regularly) launch the replication utility:

$ sos replicate -blob-server http://localhost:4001,http://localhost:4002 --verbose
group - server
   default - http://localhost:4001
   default - http://localhost:4002
Syncing group: default
   Group member: http://localhost:4001
   Group member: http://localhost:4002
   Object cd5bd649c4dc46b0bbdf8c94ee53c1198780e430 is missing on http://localhost:4001
     Mirroring cd5bd649c4dc46b0bbdf8c94ee53c1198780e430 from http://localhost:4002 to http://localhost:4001
        Fetching :http://localhost:4002/blob/cd5bd649c4dc46b0bbdf8c94ee53c1198780e430
        Uploading :http://localhost:4001/blob/cd5bd649c4dc46b0bbdf8c94ee53c1198780e430

Meta-Data

When uploading objects it is often useful to store meta-data, such as the original name of the uploaded object, the owner, or some similar data. For that reason any header you add to your upload with an X-prefix will be stored and returned on download.

As a special case the header X-Mime-Type can be used to set the returned Content-Type header too.

For example uploading an image might look like this:

$ curl -X POST -H "X-Orig-Filename: steve.jpg" \
               -H "X-MIME-Type: image/jpeg" \
               --data-binary @/home/skx/Images/tmp/steve.jpg \
        http://localhost:9991/upload
{"id":"20b30df22469e6d7617c7da6a457d4e384945a06","status":"OK","size":17599}

Downloading will result in the headers being set:

$ curl -v http://localhost:9992/fetch/20b30df22469e6d7617c7da6a457d4e384945a06 >/dev/null
..
< HTTP/1.1 200 OK
< X-Orig-Filename: steve.jpg
< Date: Fri, 27 May 2016 06:17:39 GMT
< Content-Type: image/jpeg
< Transfer-Encoding: chunked
<
{ [data not shown]

Production Usage

• The API service must be visible to clients, to allow downloads to be made.

  • Because the download service runs on port 9992 it is assumed that corporate firewalls would deny access.
  • We assume you'll configure an Apache/nginx/similar reverse-proxy to access the files via a host like http://objects.example.com/.

• It is assumed you might wish to restrict uploads to particular clients, rather than allow the world to make uploads. The simplest way of doing this is to use your firewall to filter access to port 9991.

• The blob-servers must be reachable by the host(s) running the API-service, but they should not be publicly visible.

  • If your blob-servers are exposed to the internet remote users could use the API to spider and download all your content.

• None of the servers need to be launched as root, because they don't bind to privileged ports, or require special access.

  • NOTE: issue #6 improved the security of the blob-server by invoking chroot(). However chroot() will fail if the server is not launched as root, which is harmless.

• You can also read about scaling when your data is too large to fit upon a single blob-server:

  • Read about scaling SoS

Future Changes?

It would be possible to switch to using chunked storage, for example breaking up each file that is uploaded into 128Mb sections and treating them as distinct. The reason that is not done at the moment is because it relies upon state:

• The public server needs to be able to know that the file with a given ID is comprised of the following chunks of data:
  • a5d606958533634fed7e6d5a79d6a5617252021f
  • 038deb6940db2d0e7b9ee9bba70f3501a0667989
  • a7914eb6ff984f97c5f6f365d3d93961be2e8617
  • ...
• That data must be always kept up to date and accessible.

At the moment the API-server is stateless, so tracking that data is not possible. It possible to imagine using redis, or some other external database to record the data, but that increases the complexity of deployment.

Github Setup

This repository is configured to run tests upon every commit, and when pull-requests are created/updated. The testing is carried out via .github/run-tests.sh which is used by the github-action-tester action.

Releases are automated in a similar fashion via .github/build, and the github-action-publish-binaries action.

Questions?

Questions/Changes are most welcome; just report an issue.

Steve