This tool is specifically designed for syncing static web sites managed via FTP to S3.

• Posix only.
• Efficient:
  • Need to work with millions of files (but not 10s of millions)
  • Can't wait very long. This implies using multiple threads.
  • Not practical to compute md5s
  • known usage patterns mean we can simply use modification times.
• Addresses an insane requirement of emulating insane rewrites by making extra data copies. Whimper.
• Option local index avoids listing buckets, which is important for large buckets.
• Retry on failed AWS operations.

On each syncronization:

• Get {path->mtime} mapping for s3 and for local file system.

  This is done in 2 threads.

  Optionally, we can keep an index, based on the previous file scan and avoid scanning s3.

• Compute diff.
• Apply diff to s3
  • Configurable number of workers apply changes one by one.
  • Workers fed from queue.
• As an optimization to try to avoid creation of giant dicts, when building dicts, if a key is already in the other dict, to the comparison right away.
• To decide if files have changed, we compare file-system modification times and S3 object modification times. This is awkward as they don't match up precicely. We end up having to add a fudge factor to file-system modification times to account for the fact that thay don't line up, as well as for clock skew.

Note on AWS keys

You pass keys via AWS instance roles (if running in AWS), .boto files, keyring, or .boto files.

• Added support for cloudfront invalidations.

• Fixed: directories with weird file names broke index generation.

• Fixed: index.html files included dot files.

• Fixed: Content-Type wasn't set for generated index.html files.

  (Also tweaked html layout to force new index.html files to be sent.

• Added generation of index.html files in S3 for directories without them on the file system.
• Removed simple prefix rewrites. We didn't need them.
• Fixed: the restore script didn't remove extra files from the destination directory.

Added a simple restore script for restoring files from S3. It can restore an entire directory or update a directory, syncing with S3 based on file size.

Added missing retry on failed adds or deletes.

Added support for using a local index file to avoid lengthy bucket scans.

Added lock-file support to avoid simultaneous syncs.

Added support for bucket prefixes (mainly for secondary use cases).

Added a -D option to disable deleting keys.

Implement simple prefix rewrites that duplicate keys matching certain prefixes to the same keys but with different prefixes.

Fix: needed to use encoded file names when reading data from file

system. (We were storing them decoded and boto was using a different encoding when trying to read them.)

Decode file paths using the configured encoding, which defaults to latin-1.

Refactored the way time stamps are compared. Iterating over s3 buckets doesnt' return user-defined meta data (and it woould be too expensive to fetch it on a case-by-case bases), so we can't capture the original mtimes (which has a race condition the way we did it anyway). Instead, we now compare file-system modification times with S3 object modification times, using a fudge factor to account for the fact that they're not computed the same way, and for clock skew.

Initial release