#concurrent #dashboard #tasks #progress #blocking

prodash

A dashboard for visualizing progress of asynchronous and possibly blocking tasks

17 stable releases

✓ Uses Rust 2018 edition

new 3.1.1 Mar 25, 2020
3.0.2 Mar 25, 2020
2.1.0 Mar 24, 2020
1.2.0 Mar 7, 2020
1.1.5 Feb 23, 2020

#12 in Visualization

Download history 84/week @ 2020-02-16 139/week @ 2020-02-23 36/week @ 2020-03-01 55/week @ 2020-03-08 80/week @ 2020-03-15

188 downloads per month
Used in 3 crates (via criner)

MIT license

66KB
1.5K SLoC

Build Status

prodash is a dashboard for displaying progress of concurrent applications.

It's easy to integrate thanks to a pragmatic API, and comes with a terminal user interface by default.

asciicast

How to use…

Be sure to read the documentation at https://docs.rs/prodash, it contains various examples on how to get started.

Or run the demo application like so cd prodash && cargo run --example dashboard.

Features

  • fast insertions and updates for transparent progress tracking of highly concurrent programs
  • a messages buffer for information about success and failure
  • a terminal user interface for visualization, with keyboard controls and dynamic re-sizing
  • unicode and multi-width character support

Limitations

  • it does copy quite some state each time it displays progress information and messages
  • The underlying sync data structure, dashmap, does not document every use of unsafe
    • I also evaluated evmap, which has 25% less uses of unsafe, but a more complex interface.
    • Thus far it seemed 'ok' to use, who knows… we are getting mutable pieces of a hashmap from multiple threads, however, we never hand out multiple handles to the same child which should make actual concurrent access to the same key impossible.
  • If there are more than 2^16 tasks
    • then
      • running concurrently on a single level of the tree, they start overwriting each other
      • over its lifetime, even though they do not run concurrently, eventually new tasks will seem like old tasks (their ID wrapped around)
    • why
      • on drop, they never decrement a child count used to generate a new ID
    • fix
      • make the id bigger, like u32
      • we should do that once there is a performance test

Lessons Learned

  • drop() is not garantueed to be called when the future returns Ready and is in the futures::executor::ThreadPool
    • Workaround: drop and cleanup explicitly, prone to forgetting it.
    • This is also why futures::future::abortable() works (by stopping the polling), but doesn't as cleanup is not performed, even though it clearly would be preferred.
    • fix
      • Use a join handle and await it - this will drop the future properly
  • select() might not work with complex futures - these should then be boxed() if Unpin isn't implemented.

Dependencies

~0.6–1.7MB
~34K SLoC