Skip to content

Latest commit

 

History

History
51 lines (35 loc) · 2.39 KB

README.md

File metadata and controls

51 lines (35 loc) · 2.39 KB
Raw

Maildir Deduplicate

Deduplicates maildir contents using hardlinks. Assumes immutable files.

As long as processes delivering/reading messages do not change the contents, deduplication works without data corruption, and without any support from MTA/mail clients. Maildir specification mandates that files must be modified only under tmp directories. All other operations should be either creating hard links or unlinking. However, not all programs follow this principle.

Messages are deduplicated on file level (instead of block level), meaning, only exactly the same messages will be deduplicated.

There is no built-in mechanism to undo deduplication. If enough space is available, probably the easiest approach is to copy the whole maildir and delete the original one (and repeat this for all deduplicated directories). When finished, delete the dedup folder.

Installation

No dependencies. Edit settings.py to add pattern for maildir folders.

Assumes maildir subfolder and creates dedup folder for storing hard links using hashes. For example, FOLDERS = ["/storage/gmail/*"] assumes following folder structure:

/storage
  \- gmail
    \- something - write permission to this folder and everything under this is necessary
      \- maildir
        \- label
          \- cur
          \- new
          \- tmp - skipped
        \- label
          \- cur
          ...
          ...
      \- dedup - automatically created by maildir_dedup.py

Deduplication process goes as follows:

  • Scan all files under cur/new folders. See maildir(5) for more information.
  • Calculate SHA512 hash of file contents (everything, including headers). Convert hash to hexdigest.
  • Create folder dedup/hash[0]/hash[1]/hash[2] to avoid folders with too many files.
  • Check whether dedup/hash[0]/hash[1]/hash[2]/hash exists.
    • If yes, check whether both original file and hash file point to same inode.
      • If yes, skip.
      • If no, delete the original file and create a new hard link from the hash file to the original file.
    • If no, create new hard link from the original file to the hash filename.

TODO

Handle file deletion. Currently, if all copies of the message has been deleted, hard link under dedup/ still persists. os.stat returns st_nlink, which is the number of hard links to inode. If st_nlink is >1, hash file is still in use.