gos3rve

Expose local file system via S3 API ( S3 Proxy/connector/server )

TLDR;

git clone git@github.com:shvechkov/gos3rve.git && cd gos3rve && go build 

./gos3rve  -p 8080 -user_id you@email.com -key_id s3KeyID -key_val s3SecretKey

Get full list of flags/usage by running -h flag:

Usage of ./gos3rve:
  -config string
    	configuration file  (default "./config.xml")
  -dir_buckets string
    	dir to store buckets (default "./buckets/")
  -dir_uploads string
    	temp dir to store upload parts (default "./uploads/")
  -help
    	Show usage
  -key_id string
    	Access Key ID (default "muB07ZERr4")
  -key_val string
    	Secret Access Key (default "U8J89Z6XZCwXBWv1lP8tbzK35AaiR7Fz")
  -p int
    	Port to listen on (default 8080)
  -region string
    	S3 region (default "us-east-1")
  -user_id string
    	AWS S3 user ID (default "c5dbe9e2-4d44-404a-96f9-bd1dc1163a4a")
  -user_name string
    	AWS S3 user name (default "s3user@amazon.com")

Configuration values can be passed in configuration file (using –config flag) Here is the sample config file:

<root>
    <AccessKeyId>any</AccessKeyId>
    <SecretAccessKey>key</SecretAccessKey>
    <Region>us-east-1</Region>
    <UploadsPath>./uploads</UploadsPath>
    <BucketsPath>./buckets</BucketsPath>
    <Port>8080</Port>
</root>

supported S3 operations

tested with s3cmd

AWS API	supported	s3cmd
ListObjectsV2	yes	ls
CreateBucket	yes	mb
DeleteBucket	yes	rb
PutObject	yes	put
GetObject	yes	get
DeleteObject	yes	del

How to build

Install golang on your platform and execute :

git clone git@github.com:shvechkov/gos3rve.git && cd gos3rve && go build

For creating static executable (which runs on all Linux platforms w. same arch) run:

CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build -a -tags netgo -ldflags '-w'

Run as a service

sudo cp gos3rve.service  /etc/systemd/system/
sudo systemctl start gos3rve
sudo systemctl enable gos3rve

TBD

parts of multipart uploads should go into separate temp dir (to prevent end user from seeing partially uploaded objects/to maintain atomicity). If multipart-part upload fails we should clean stale parts .. This can be done asynchronously by GC thread which wil monitor temp uploads dir
implement mv /renames, implement other missing APIs (?)
add option to allow only unique uploads and/or use FS reflinks, if supported ( xfs/btrs/zfs), to clone existing object instead of creating new onces

Disclaimer

This is a toy project - quick and dirty code created for testing and educational purposes only. Use at your own risk. Look at ceph or seaweedfs if you need production grade scalable solution