Embedded persistent key-value storage optimized for random-access workload and huge-capacity HDD

3 releases

0.9.2 Jan 16, 2019
0.9.1 Oct 26, 2018
0.9.0 Oct 17, 2018

#30 in Database implementations

Download history 18/week @ 2018-12-20 15/week @ 2018-12-27 176/week @ 2019-01-03 153/week @ 2019-01-10 126/week @ 2019-01-17 32/week @ 2019-01-24 137/week @ 2019-01-31 64/week @ 2019-02-07 55/week @ 2019-02-14 106/week @ 2019-02-21 157/week @ 2019-02-28 313/week @ 2019-03-07 365/week @ 2019-03-14 125/week @ 2019-03-21 191/week @ 2019-03-28

311 downloads per month
Used in 8 crates

MIT license


cannyls cannyls Documentation Build Status License: MIT

CannyLS is an embedded and persistent key-value storage optimized for random-access workload and huge-capacity HDD.

CannyLS mainly has following features:

  • A local storage for storing objects that called as "lump":
    • Basically, a lump is a simple key-value entry
    • The distinctive properties are that the key is fixed length (128 bits) and suited for storing a relatively large size value (e.g., several MB)
  • Provides simple functionalities:
  • Optimized for random-access workload on huge-capacity HDD (up to 512 TB):
  • Aiming to provide predictable and stable read/write latency:
    • There are (nearly) strict upper bounds about the number of disk accesses issued when executing operations
      • One disk access when PUT and DELETE, and two when PUT
    • There are no background processings like compaction and stop-the-world GC which may block normal operations for a long time
    • For eliminating overhead and uncertainty, CannyLS has no caching layer:
      • It uses Direct I/O for bypassing OS layer caching (e.g., page cache)
      • If you need any caching layer, it is your responsibility to implement it
  • Detailed metrics are exposed using Prometheus

See Wiki for more details about CannyLS.



~67K SLoC