Go to file
2019-09-23 16:31:22 +01:00
.github/workflows Use actions-rs 2019-09-23 13:00:10 +01:00
.vscode add vscode tasks 2018-12-22 16:08:49 +00:00
src Versions should increase 2019-09-23 16:30:40 +01:00
tests Remove unnecessary todo 2019-09-23 16:31:22 +01:00
.gitignore Ignore intellij files 2018-08-17 17:58:42 +01:00
Cargo.lock Add git hooks 2019-09-07 16:12:18 +01:00
Cargo.toml Add git hooks 2019-09-07 16:12:18 +01:00
LICENSE Initial commit 2018-08-14 19:59:05 +01:00
README.md Update README.md 2019-08-31 11:05:56 +01:00
rust-toolchain add toolchain 2018-12-20 16:47:22 +00:00
rustfmt.toml rustfmt config 2018-10-04 13:42:39 +01:00

Motivation: All the backup systems are either slow or crashing or both on my backup.

Tried duply:

Works but is very slow
--------------[ Backup Statistics ]--------------
StartTime 1547198362.85 (Fri Jan 11 09:19:22 2019)
EndTime 1547209509.04 (Fri Jan 11 12:25:09 2019)
ElapsedTime 11146.19 (3 hours 5 minutes 46.19 seconds)
SourceFiles 3065438
SourceFileSize 585041709586 (545 GB)
NewFiles 0
NewFileSize 0 (0 bytes)
DeletedFiles 0
ChangedFiles 0
ChangedFileSize 0 (0 bytes)
ChangedDeltaSize 0 (0 bytes)
DeltaEntries 0
RawDeltaSize 0 (0 bytes)
TotalDestinationSizeChange 111 (111 bytes)
Errors 0
-------------------------------------------------

--- Finished state OK at 12:25:15.000 - Runtime 03:06:43.000 ---

Tried restic:

  • crashes with OOM
  • corrupts repo if you suspend one backup process and launch another from a different computer

Goals for bakare:

  • fast
  • using max bandwidth
  • use max cpu
  • use max disk I/O
  • memory usage limit
  • encryption by default - asymmetric, creates a keypair for you
  • deduplication of file data
  • fuzzy find by file name in stored files
  • failure to process one file should not affect any other files
  • intermittent network failures should not make the whole process fail (test with random packet drop)
  • system suspend/resume should not make the repo become corrupted, even in the presence of other active backup processes running on other computers, targeting same repo - this is where restic fails

Nice to have:

  • daemon that listens for file events and updates a list of files to be backed up on the next backup run - or a continous backup mode - the daemon uploads the file whenever it sees the change
  • peer2peer mode - people storing encrypted backups for each other
  • relay mode, where daemon works on one or more central points with local storage (e.g. NAS) and various computers sync with that central location. Then though the central locaiton uploads everything to the other location, typically the cloud.

Implementation:

  • test with randomly created dirs and files, with property based tests and fuzzer
  • see if we can use salsa for recomputaiton
  • index corruption tests - mutate random byte and see if everything is readable
  • network packet drop tests