s3fsr is yet another file system interface to S3.

This means you can mount your S3 buckets as directories (like ~/s3) and then use ls/cp/mv to copy/move directories and files between your S3 buckets and your other file systems.

Most usefully, s3fsr understands four popular styles of S3 directory notation:

• s3sync library's etag marker objects
• S3 Organizer plugin's _$folder$ suffixed marker objects
• AWS Console's folder/ marker objects
• Plain "common prefix" directories (no marker objects, just inferred by Amazon's S3 API by having children objects, based on / as a delimiter)

This means you should be able to browse most any S3 bucket hierarchically without seeing odd names, duplicates entries, or missing directories.

When explicitly creating directories (e.g. with mkdir), the AWS Console folder/ marker directory convention is used.

s3fsr [<bucket-name>] <mount-point>

For example, to mount the bucket mybucket to the directory ~/s3:

1. mkdir ~/s3
2. s3fsr mybucket ~/s3
3. ls ~/s3 -- see all the directories/files inside of mybucket

To mount all of your buckets to the directory ~/s3:

1. mkdir ~/s3
2. s3fsr ~/s3
3. ls ~/s3 -- see all of the buckets for your Amazon account

When you're done:

1. Use Ctrl-C to kill the ruby process, this will also unmount the directory

s3fsr uses the AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables, so you need to set them with your Amazon key information.

• Install Ruby for your OS
• Install FUSE and fusefs for your OS
  • FUSE is very OS-specific, so the Ruby fusefs library is not a gem
  • E.g. on Ubuntu, installing the libfusefs-ruby package will install FUSE and the Ruby fusefs library
• Install libopenssl-ruby
• gem install s3fsr

• To avoid indexing daemons from scanning your S3 mount, you might try chmod og-rx ~/s3
  • I'm pretty sure this worked for me...feedback appreciated

s3fsr is very handy, but it's target audience is a developer poking at S3 during their everyday development tasks. Before using it in production for any sort of automation processes, please consider that:

• Ruby's FuseFS is inherently single-threaded (AFAICT), so if one process is saving/loading a file, any other process that accesses the s3fsr-mounted directory will block.
• Ruby's FuseFS has no streaming support (again AFAICT), so all data is passed around in memory as Strings.
• s3fsr itself uses a very naive way of accessing files (explicitly, saving sub1/sub2/filea.txt implicitly involves fetching the names of the files in sub1 and sub2) that is not optimized for the case of only writing new files and not ever reading back the list of existing files.

• File content is never cached (though file size is)
• Directory listings are always cached
• Directory listings can be explicitly reset using touch, e.g. touch ~/s3/subdir

This gives very good CLI performance, e.g. for quick ls/cd commands, while ensuring the data itself is always fresh.

• 1.0 - Initial release
• 1.1 - Fix file size to not make extra per-file HEAD requests
• 1.2 - Fix directories with >1000 files
• 1.3 - Killing s3fsr now also unmounts the directory
• 1.4 - Fix directories that are only from common prefixes, move to Gemcutter
• 1.5 - Fix touching a file that does not yet exist, add dependent gems
• 1.6 - Fix for directory marker files ending with /, e.g. dir/
• 1.7 - Fix deletion of AWS Console-style directory marker files, create new directories with AWS Console-style
• 1.8 - Fix marker handling and horrible performance for large, prefix-based directories.
• 1.9 - Fix file/directory names being numbers causing TypeErrors
• 1.10 - Syntax fixes for Ruby 1.9
• 1.11 - Another fix for Ruby 1.9

• Add switch to pass allow_other so that other users can use your mount (e.g. if you want root to do the mounting on boot)
• Nothing is streamed, so if you have files larger than your available RAM, s3fsr won't work
• Given the recent speedups, timing out the cache every ~5 minutes or so seems reasonable--probably via a command line parameter