I wonder so often about many new CLI tools whose primary selling point is their speed over other tools. Yet I personally have not encountered any case where a tool like jq feels incredibly slow, and I would feel the urge to find something else.
What do people do all day that existing tools are no longer enough? Or is it that kind of "my new terminal opens 107ms faster now, and I don't notice it, but I simply feel better because I know"?
I process TB-size ndjson files. I want to use jq to do some simple transformations between stages of the processing pipeline (e.g. rename a field), but it so slow that I write a single-use node or rust script instead.
> Now I'm really curious. What field are you in that ndjson files of that size are common?
I'm not OP,but structured JSON logs can easily result in humongous ndjson files, even with a modest fleet of servers over a not-very-long period of time.
Replying here because the other comment is too deeply nested to reply.
Even if it's once off, some people handle a lot of once-offs, that's exactly where you need good CLI tooling to support it.
Sure jq isn't exactly super slow, but I also have avoided it in pipelines where I just need faster throughput.
rg was insanely useful in a project I once got where they had about 5GB of source files, a lot of them auto-generated. And you needed to find stuff in there. People were using Notepad++ and waiting minutes for a query to find something in the haystack. rg returned results in seconds.
The use case could be e.g. exactly processing an old trove of logs into something more easily indexed and queryable, and you might want to use jq as part of that processing pipeline
Fair, but for a once-off thing performance isn't usually a major factor.
The comment I was replying to implied this was something more regular.
EDIT: why is this being downvoted? I didn't think I was rude. The person I responded to made a good point, I was just clarifying that it wasn't quite the situation I was asking about.
At scale, low performance can very easily mean "longer than the lifetime of the universe to execute." The question isn't how quickly something will get done, but whether it can be done at all.
Good point. I said it above, but I'll repeat it here that I shouldn't have discounted how frequent once offs can be. I've worked in support before so I really should've known better
Certain people/businesses deal with one-off things every day. Even for something truly one-off, if one tool is too slow it might still be the difference between being able to do it once or not at all.
I would love, _love_ to know more about your data formats, your tools, what the JSON looks like, basically as much as you're willing to share. :)
For about a month now I've been working on a suite of tools for dealing with JSON specifically written for the imagined audience of "for people who like CLIs or TUIs and have to deal with PILES AND PILES of JSON and care deeply about performance".
For me, I've been writing them just because it's an "itch". I like writing high performance/efficient software, and there's a few gaps that it bugged me they existed, that I knew I could fill.
I'm having fun and will be happy when I finish, regardless, but it would be so cool if it happened to solve a problem for someone else.
I maintain some tools for the videogame World of Warships. The developer has a file called GameParams.bin which is Python-pickled data (their scripting language is Python).
Working with this is pretty painful, so I convert the Pickled structure to other formats including JSON.
The file has always been prettified around ~500MB but as of recently expands to about 3GB I think because they’ve added extra regional parameters.
The file inflates to a large size because Pickle refcounts objects for deduping, whereas obviously that’s lost in JSON.
I care about speed and tools not choking on the large inputs so I use jaq for querying and instruction LLMs operating on the data to do the same.
> The query language is deliberately less expressive than jq's. jsongrep is a search tool, not a transformation tool-- it finds values but doesn't compute new ones. There are no filters, no arithmetic, no string interpolation.
Mind me asking what sorts of TB json files you work with? Seems excessively immense.
Command-line Tools can be 235x Faster than your Hadoop Cluster (2014)
Conclusion: Hopefully this has illustrated some points about using and abusing tools like Hadoop for data processing tasks that can better be accomplished on a single machine with simple shell commands and tools.
This article is good for new programmers to understand why certain solutions are better at scale, there is no silver bullet. And also, this is from 2014, and the dataset is < 4GB. No reason to use hadoop.
The discussion we had here was involving TB of data, so I'm curious how this is faster with CLIs rather than parallel processing...
JQ is very convenient, even if your files are more than 100GB.
I often need to extract one field from huge JSON line files, I just pipe jq to it to get results. It's slower, but implementing proper data processing will take more time.
If you work at a hyperscaler, service log volume borders on the insane, and while there is a whole pile of tooling around logs, often there's no real substitute for pulling a couple of terabytes locally and going to town on them.
> often there's no real substitute for pulling a couple of terabytes locally and going to town on them.
Fully agree. I already know the locations of the logs on-disk, and ripgrep - or at worst, grep with LC_ALL=C - is much, much faster than any aggregation tool.
If I need to compare different machines, or do complex projections, then sure, external tooling is probably easier. But for the case of “I know roughly when a problem occurred / a text pattern to match,” reading the local file is faster.
I'll write a one-off shell pipeline to inspect something on 10^5 servers - it will be sent to each of those servers and run once or a handful of times, and the results will be transmitted back and that's that. Kind of a map-reduce shell thing, for ops type tasks.
Sometimes those will actually need to process through a bunch of data unexpectedly.
Sometimes those will be run on a loop - once per second, N per minute (etc), and the results will be used to monitor a situation until a bug is fixed or a spike in load is resolved or a proper monitoring program/metric can be deployed.
Sometimes those are to investigate a pegged CPU and the amortized lower runtime across all the tasks on the CPU is noticable.
We run our machines hot and part of the reason we can do that is being in the habit of choosing lower cost (in cycles) tooling whenever we can. If i can spend a little time and effort learning a tool that saves a bunch of cpu in aggregate, its a win. When the whole company does it, we can spend a lot less on hardware than it costs in engineer time to make these decisions.
Another way of putting it is: its a type of frugality (not cheapness, just spending wisely). If you save a dollar once, its nothing. If you have a habit of saving a dollar every time the opportunity arises, it adds up quickly. By having a habit of choosing more performant tools, you're less likely to hit a case where you wish you did use more performant tools, and are practiced at it when the need arises for pure parsimony and it's less painful.
We parse JSON responses for dashboards, alerting, etc. Thousands of nodes, depending on the resolution of your monitoring you could see improvements here.
- They take Rust for performance or FAVORITE_LANG for credentials
- Claude implements small subset of features
- Benchmark subset
- Claim win, profit on showcase
Note: this particular project doesn't have many visible tells, but there's pattern of overdocumentation (17% comment-to-code ratio, >1000 words in README, Claude-like comment patterns), so it might be a guided process.
I still think that the project follows the "subset is faster than set" trend.
You don't know something is slow until you encounter a use case where the speed becomes noticeable. Then you see the slowness across the board. If you can notice that a command hasn't completed and you are able to fully process a thought about it, it's slow(er than your mind, ergo slow!).
Usually, a perceptive user/technical mind is able to tweak their usage of the tools around their limitations, but if you can find a tool that doesn't have those limitations, it feels far more superior.
The only place where ripgrep hasn't seeped into my workflow for example, is after the pipe and that's just out of (bad?) habit. So much so, sometimes I'll do this foolishly rg "<term>" | grep <second filter>; then proceed to do a metaphoric facepalm on my mind. Let's see if jg can make me go jg <term> | jq <transformation> :)
Well grep is just better sometimes. Like you want to copy some lines and grep at the end of a pipeline is just easier than rg -N to suppress line numbers. Whatever works, no need to facepalm.
If somebody needs performance, they probably shouldn't be calling out to a separate process for json of all things, no?
(Honestly, who even still writes shell scripts? Have a coding agent write the thing in a real scripting language at least, they aren't phased by the boilerplate of constructing pipelines with python or whatever. I haven't written a shell script in over a year now.)
If you’re writing the script to be used by multiple people, or on multiple systems, or for CI runners, or in containers, etc. then there’s no guarantee of having Python (mostly for the container situation, but still), much less of its version. It’s far too easy to accidentally use a feature or syntax that you took for granted, because who would still be using 3.7 today, anyway? I say this from painful recent experience.
Plus, for any script that’s going to be fetching or posting anything over a network, the LLM will almost certainly want to include requests, so now you either have to deal with dependencies, or make it use urllib.
In contrast, there’s an extremely high likelihood of the environment having a POSIX-compatible interpreter, so as long as you don’t use bash-isms (or zsh-isms, etc.), the script will probably work. For network access, the odds of it having curl are also quite high, moreso (especially in containers) than Python.
If you're distributing the script to other people then the benifit of using python and getting stuff like high quality argument parsing for free is even greater.
Prioritizing SEO-ing speed over supporting the same features/syntax (especially without an immediately prominent disclosure of these deficiencies) = marketing bullshit
A faster jq except it can't do what jq does... maybe I can use this as a pre-filter when necessary.
I am not sure if it was simon or pg who might've quoted this but I remembered a quote about that a 2 magnitude order in speed (quantity) is a huge qualititative change in it of itself.
