Ever found yourself in a remote ssh session, doing some lengthy operations, and every now and then you feel the need to hit [enter] just to ensure you didn't lose the connection? Ever wondered where your processing is in, and when will it finish? Ever needed to pause the progress bar for a while, return to the python REPL for a manual inspection or fixing an item, and then resume the process like it never happened? I did...
I've made this cool progress bar thinking about all that, the Alive-Progress bar! :)
I like to think of it as a new kind of progress bar for python, as it has among other things:
- a cool live spinner, which makes it clear the process did not hang and your terminal/connection is healthy;
- a visual feedback of the current speed/throughput, as the spinner runs faster or slower according to the actual processing speed;
- an efficient multi-threaded bar, which updates itself at a fraction of the actual speed (1,000,000 iterations per second equates to roughly 60fps refresh rate) to keep CPU usage low and avoid terminal spamming;
- an expected time of arrival (eta), that shows the remaining processing time in a friendly way, not anything like
eta: 1584s, it will nicely show
eta: 0:26:24as you would expect (but anything less than a minute is indeed
print()hook, which allows print statements in the midst of an alive-bar context, nicely cleaning the output of any garbage, and even enriching with the current count when it occurred;
- after your processing has finished, a nice receipt is printed with the statistics of that run, including the elapsed time and observed throughput;
- it tracks the actual count in regard of the expected count, so it will look different if you send in more or less than expected;
- it automatically detects if there's really an allocated tty, and if there isn't, only the final receipt is printed, so you can safely include the alive-bar in all and any code and rest assure your log file won't get 60fps garbage;
- you can pause the alive-bar! I think that's an unprecedented feature for a progress-bar, it's incredible to orchestrate operations that require manual interaction on some items;
- it is customizable, with a growing smorgasbord of different bar and spinner styles, as well as several factories to easily generate yours!
Just install with pip:
$ pip install alive-progress
Open a context manager like this:
from alive_progress import alive_bar items = range(1000) # retrieve your set of items with alive_bar(len(items)) as bar: # declare your expected total for item in items: # iterate as usual # process each item bar() # call after consuming one item
And it's alive! 😲
In general lines, just retrieve the items, enter the
alive_bar(total) context manager, and iterate/process as usual, calling
bar() once per item.
itemscan be any iterable, and usually will be some queryset;
- the first argument of the
alive_baris the expected total, it could be a
qs.count()for querysets, a
len(items)if the iterable supports it, or anything that returns an integer;
bar()call is what makes the bar go forward -- you usually call it in every iteration after consuming an item, but you can get creative! For example you could call it only when you find something you want, or call it more than once in the same iteration, depending on what you want to monitor. Just adjust the total accordingly to get a useful eta;
bar()call also returns the current count if needed, and enables to pass situational messages to the bar.
So, you could even use it like:
with alive_bar(3) as bar: corpus = read_big_file() bar('file read, tokenizing') tokens = tokenize(corpus) bar('tokens ok, processing') process(tokens) bar()
total argument is optional. Providing it makes the bar enter the definite mode, the one used for well-bounded tasks.
If you do not provide it, the bar enters the unknown mode. In this mode, the whole progress-bar is animated like the cool spinners, as it's not possible to determine the percentage of completion.
Note that the cool spinners are still present, and each animation runs independently of each other, rendering a unique show in your terminal.
Then you have the (new) manual mode, where you get to manually control the bar!
Just pass a
manual=True argument to
alive_bar(), and send a progress percentage (a float between 0 and 1) to the
bar() call to put the alive-bar in wherever position you want! Call it as frequently as you need.
The frames per second will be computed according to the sent progress and the actual elapsed time.
In this mode, you can also provide the total if you have it, and get all the same position and throughput statistics as the definite mode. The position is dynamically calculated only when needed to increase efficiency.
If you don't provide the total, a simpler
%/s will be used, with no position nor eta.
Wondering what styles does it have bundled? It's
showtime! ;) (Click it to see in motion)
I've made these styles to test all combinations of parameters of the factories, but I think some of them ended up very very cool! Use them, or create your own.
While in an alive progress bar context, you have two ways to output messages:
bar('message'), which besides incrementing the counter, also sets/overwrites a situational message within the bar line, usually to display something about the phase the processing is in, or some hint about the items being processed;
print('message'), which prints an enriched message that includes the current position of the alive bar, thus leaving behind a log and continuing the bar below it.
Both methods always clear the line appropriately to remove any garbage of previous messages on screen. (Click it to see in motion)
All of the components are individually customizable, both globally and per use!
And you can mix and match them! (Click it to see in motion)
Create your own animations
Make your own spinners and bars!
There's builtin support for frames, scrolling, bouncing, delayed and compound spinners! Get creative! (Click it to see in motion)
The Pause mechanism
To use the pause mechanism, you must use a generator to yield the objects you want to interact with. The bar object includes another context manager to do that, just do
with bar.pause(): yield obj.
Let's use an example, suppose you need to reconcile transactions. You need to iterate over thousands of them, detect somehow the faulty ones, and fix them. They could be broken or not synced or invalid or anything else, several different problems.
Typically you would have to let the process run, appending to a list each inconsistency found, and waiting, potentially a long time, until the end to be able to do anything. You could mitigate this by processing in chunks, but that has its own implications.
With the Alive-Progress bar and the Pause mechanism, you can inspect these transactions in real-time! You wait only until the next one is found! You would do something like this:
def reconcile_transactions(): qs = Transaction.objects.filter() # django example, or in sqlalchemy: session.query(Transaction).filter() with alive_bar(qs.count()) as bar: for transaction in qs: if not validate(transaction): with bar.pause(): yield transaction bar()
Then you could use it in ipython or your preferred REPL, just instantiate the generator and call
next. The progress bar will run as usual, but as soon as an inconsistency is found, the bar pauses itself and you get the prompt back:
In : gen = reconcile_transactions() In : t = next(gen, None) |█████████████████████ | 105/200 [52%] in 5s (18.8/s, eta: 4s) In : t Out: Transaction<#123>
Debug and fix that transaction any way you want, and when you're done, continue the process with the same
next as before... The bar returns like nothing happened!! How cool is that? :)
In : t = next(gen, None) |█████████████████████ | ▁▃▅ 105/200 [52%] in 5s (18.8/s, eta: 4s)
Forcing animations on not-interactive consoles (like Pycharm's)
Pycharm's python console do not report itself as "interactive", so I've included a
force_tty argument to be able to use the alive-progress bar in it.
So, just start it as:
with alive_bar(1000, force_tty=1) as bar: for i in range(1000): time.sleep(.01) bar()
Do note that this console is heavily instrumented and has much more overhead, so the outcome is not as fluid as you would expect.
create an unknown mode for bars (without a known total and eta) implement a pausing mechanism change spinner styles change bar styles include a global configuration system create generators for scrolling, bouncing, delayed and compound spinners create an exhibition of spinners and bars, to see them all in motion include theme support in configuration
- include colors in spinners and bars
- try some adaptive algorithm for ETA, like moving average or exponential smoothing
- any other ideas welcome!
- This whole project was implemented in functional style;
- It uses wholeheartedly Closures and Generators, they're in almost all modules here;
- It does not declare even a single class;
- It does not have any dependencies.
- 1.1.0: new manual mode
- 1.0.1: pycharm console support, improve compatibility with python streams
- 1.0.0: first public release, already very complete and mature
This software is licensed under the MIT License. See the LICENSE file in the top distribution directory for the full license text.