Dgsh – Directed graph shell
www2.dmst.aueb.gr153 points by pabs3 2 days ago
153 points by pabs3 2 days ago
This would have been great 10-20 years ago, or even at the coining of Unix pipes. By today's standards, however, the syntax feels clunky and dated. I'd like to see contemporary shells like nushell and elvish copy these ideas, with attribution of course, in a more modern way. That is the best way I can see to honor this stagnant project: https://github.com/dspinellis/dgsh
I went through two iterations before adopting the current syntax. Truth is neither me nor Doug McIlroy, the inventor of Unix pipes, who kindly and generously provided feedback during dgsh's development, had something better to propose.
What syntax would you propose?
Greetings, Diomidis.
I would suggest a familiar notation like "[a, b] -> c" in a dedicated dag block:
dag text_stats {
tee -> [ split_words, count_chars ]
# word-based frequencies
split_words -> tee_words
tee_words -> ngram2 -> save_digram
tee_words -> ngram3 -> save_trigram
tee_words -> ranked_frequency -> save_words
# character-based frequencies
count_chars -> add_percentage
chars_to_lines -> ranked_frequency -> add_percentage -> save_chars
}
run text_stats < input.txt
or
dag commit_graph {
git_log -> filter_recent -> sort -n -> [ uniq_committers, sort_by_email ]
uniq_committers -> [ last_commit, first_commit, committer_positions ]
[ last_commit, first_commit ] -> cat -> tr '\n' ' ' -> days_between
[ committer_positions, sort_by_email ] -> join_by_email -> sort -k2n -> [ make_bitmap_header, plot_per_day ]
[ uniq_committers, days_between ] -> emit_dims -> plot_per_day
make_bitmap_header -> cat
plot_per_day -> morphconv -> [ to_png_large, to_png_small ]
}
run commit_graph
The translations above are computer-assisted and may contain mistakes, but you get the idea.
The closeness of this syntax to graphviz dot is very interesting.
having dgsh output a graphvis file in dry-run mode would be a neat feature.
Thank you for the suggestion. This would mean that you'd also then create some mapping from each name (like git_log) to its implementation, right?
Yes, using shell functions:
git_log() {
git log --pretty=tformat:'%at %ae'
}
And, more importantly, assign a name to a process, so that it can appear multiple times in the graph.
Murex has had this capability for years. (https://github.com/lmorg/murex)
I’m on my phone at the moment and cooking so cannot type any examples, but if I get time, I’ll throw together some comparisons later tonight
I could not find any mention of DAGs or directed acyclic graphs in the documentation.
Yeah it’s not technically DAG since it uses iteration, but then dgsh will use iteration under the hood too.
However Murex does support CSP-style concurrency. So while there’s no syntax sugar for writing graphs, you can very easily create adhoc pipes and pass them around instead of using stdout / stderr.
So it wouldn’t actually take much to refine that with some DAG-friendly syntax.
In fact maybe that can be my next project…
I'm curious: what do you mean by "dgsh will use iteration under the hood too"? Dgsh does several things under the hood, but I wouldn't characterize any of them as iteration.
Yes you’re right. My apologies. I was glancing at the examples while cooking, specifically the git example (https://www2.dmst.aueb.gr/dds/sw/dgsh/#commit-stats) thinking that it was iterating over the lines output from git, but clearly that’s not even how bash would work. That will teach me for commenting without giving something my full attention first doh!
Looking properly at this, I can see no iteration is needed. Which actually makes the Murex implementation even easier because Murex already has tee pipes just like dgsh. It’s just not (yet) particularly well documented.
Admiring your multi-tasking!
Haha thank you but it’s really not that good otherwise I’d have grokked the iteration thing before making a fool out of myself!
would you be able to share or point to some examples? i am curious.
Sorry, it’s been a very busy couple of days so not had 5 minutes to test anything.
Stay tuned though, What I’m going to do is write a blog post about it. It’s an interesting enough topic to deserve one
awesome. i am looking forward to that. that blog on the murex website could use some attention anyways :-)
Frankly, I find that anything more than some preparatory `exec {my_fd}< <(commands ...)` is an unmaintainable mess, so bash is plenty for any program that should be implemented in bash.
Manually playing around with fds is definitely unmaintainable. My hope is that a clean syntax can help create maintainable complex pipelines.
Related. Others?
Dgsh – Directed Graph Shell - https://news.ycombinator.com/item?id=21700014 - Dec 2019 (11 comments)
Dgsh – Directed graph shell - https://news.ycombinator.com/item?id=13352659 - Jan 2017 (51 comments)
This is very interesting, but I'm wondering how it compares to just using a dynamic language like Python or Ruby for the same tasks. Curious how the line count to express the same tasks would come out.
From a glance, it looks like very similar tradeoffs vs bash. Much harder to read in a medium-large application, but much more ergonomic IO and process control.
I.e. much faster to use dgsh for a basic processing DAG, much more painful to use dgsh for a large ETL pipeline.
Python with something like Prefect isn't something you'd use a REPL to bang out a one-off on, but it'd be more maintainable. dgsh would let you use a REPL to bang out a quick and dirty DAG.
I've found creating pipelines with Python to be messy and intuitive. Other than creating a DSL to express them I can't see how DAGs can be expressed naturally with Python's syntax.
Even creating tools in Python that can be connected together in a Unix shell pipeline isn't trivial. By default if a downstream program stops processing Python's output you get an unsightly broken pipe exception, so you need to execute signal.signal(signal.SIGPIPE, signal.SIG_DFL) to avoid this.
I respect Python but the upgrade to Python 3 showed that data processing workloads that can be handled by standard Unix tooling should stay there.
The upgrade was a nightmare for so many organizations. It shouldn't be that way but it was.
There is a lot of stuff for Python which follows the "express computation as a dag" approach, especially Apache Airflow
Apache Airflow solves a very different problem. Its DAGs are static dependencies between sequentially executed processing steps, whereas the DAGs of dgsh express live direct data flows.
Yeah, there are also the boxes and lines tools like
which have their own subculture. You could solve the same problems they do with pandas and scikit-learn but people who use those tools would never use pandas and scikit-learn and vice versa.
Circa 2015 I was thinking those tools all had the architectural flaw that they pass relational rows over the lines as opposed to JSON objects (or equivalent) which means you had to realize joins as highly complex graphs where things that seem like local concerns to me require a global structure and where what seems like a little change to management changes the whole graph in a big way.
I found the people who were buying up that sort of tools didn’t give a damn because they thought customers demanded the speed of columnar execution which our way couldn’t deliver.
I made a prototype that gave the right answers every time and then went to work for a place which had some luck selling their own version that didn’t always give the right answers because: they didn’t know what algebra it supported, didn’t believe something like that had an algebra, and didn’t properly tear the pipeline down at the end.
Do you mean to say that two non-dependant tasks in an Airflow DAG aren't able to concurrently execute? Thats not my experience. I'm also confused by the use of 'static' in this context.
That's the point: non-dependant tasks can run concurrently in Airflow. In sh/BAsh/dgsh dependant tasks can also run concurrently, as in tar cf - . | xz.
spawning shell commands and the equivalent of piping is surprisingly hard in python. It's almost easier to do in C
There are probably libraries that could help, but then you need to install dependencies which is sad in python for other reasons
We use snakemake a lot in bioinformatics to take advantage of parallelism in workflows while staying close to Python: https://github.com/snakemake/snakemake
Others use nextflow but that requires learning Groovy and it's less intuitive.
Does the build work on Mac OSX?
A solution to the One Billion Row Challenge (1brc.dev) written in dgsh would be a interesting as a benchmark.
Nice benchmark! This is a (not at all efficient) awk one-liner.
awk -F\; ' $2 > max[$1] { max[$1] = $2 } !($1 in min) || $2 < min[$1] { min[$1] = $2 } { sum[$1] += $2; count[$1]++} END { for (n in sum) printf("%s=%.1f/%.1f/%.1f, ", n, min[n], sum[n] / count[n], max[n])}'
Can't see how dgsh could be applied to it.
Interesting. What are the benefits of thinking of data pipelines in terms of a DAG? Why cant it be cyclical with exit conditions?
This is the nerdiest thing I've ever seen, and I absolutely adore it.
As someone who loves graphs and Neo4j I wish I had though of this.
Hello. In English this makes me think of the phrase “dog shit”. Not sure if that’s intentional or not.
Second English speaker here who didn't make that connection at all
English is my third language and I can confirm I didn't even think about this
> In English this makes me think of the phrase “dog shit”.
In English, this makes me think of the phrase "dig shell". I guess we just have different things on our minds...
:-p
That's what I think when I hear "bash".
That's literally the word for shit in Norwegian: bæsj https://upload.wikimedia.org/wikipedia/commons/6/69/Nb-b%C3%...
The word “dash” is a word for shit in English (as in dashboard - literally the board on a buggy or wagon to deflect the horse droppings). That doesn’t keep a shell from being named that. Of course, dash also means to move quickly so it’s not the only meaning. Moving quickly seems to be the inspiration for the shell’s name.
Well, bash as a shell is ok. Writing complex scripts on the other hand...
Now I can't unconnect this, I hope OP was aware because now he wont forget too.