This article is full of misinformation. Just a few representative things:
- The expansion of pclntab in Go 1.2 dramatically improved startup time and reduced memory footprint, by letting the OS demand-page this critical table that is used any time a stack must be walked (in particular, during garbage collection). See https://golang.org/s/go12symtab for details.
- We (the Go team) did not “recompress” pclntab in Go 1.15. We did not remove pclntab in Go 1.16. Nor do we have plans to do either. Consequently, we never claimed “pclntab has been reduced to zero”, which is presented in the article as if a direct quote.
- If the 73% of the binary diagnosed as “not useful” were really not useful, a reasonable demonstration would be to delete it from the binary and see the binary still run. It clearly would not.
- The big table seems to claim that a 40 MB Go 1.8 binary has grown to a 289 MB Go 1.16 binary. That’s certainly not the case. More is changing from line to line in that table than the Go version.
Overall, the claim of “dark bytes” or “non-useful bytes” strikes me as similar to the claims of “junk DNA”. They’re not dark or non-useful. It turns out that having the necessary metadata for garbage collection and reflection in a statically-compiled language takes up a significant amount of space, which we’ve worked over time at reducing. But the dynamic possibilities in reflection and interface assertions mean that fewer bytes can be dropped than you’d hope. We track binary size work in https://golang.org/issue/6853.
Size fluctuates from release to release, but the overall trendline is flat: Go 1.16 binaries are roughly where Go 1.3 binaries were.
At the moment, it looks like Go 1.17 binaries will get a bit smaller thanks to the new register ABI making executable code smaller (and faster).
¹ Well, not completely. The gofmt code itself was changing from release to release, but not much. Most of the binary is the libraries and runtime, though, so it's still accurate for trends.
² Turns out we shipped the go 1.4 gofmt binary built with the race detector enabled! Oops.
Strip removes the symbol tables and DWARF information.
But still, the sum of all the bytes advertised in symbol tables for the non-DWARF data does not sum up to the stripped size. What's the remainder about?
I am reminded of how early versions of MSWord were embedding pages of heap space in save files that were not relevant to the document being saved, just because it made the saving algorithm simpler. For all we know, the go linker could be embedding random data.
> But still, the sum of all the bytes advertised in symbol tables for the non-DWARF data does not sum up to the stripped size. What's the remainder about?
if you do know, then pray, what is the answer to this question?
Maybe another day I will take the time to write a full-length blog post examining the bytes in a Go binary. Today I have other work planned and still intend to do it.
My points today are only that:
1. Go binary size has not gotten dramatically better or worse in any particular release and is mostly unchanged since Go 1.3.
2. Many claimed facts in the blog post are incorrect.
3. The linker is not "embedding random data" into Go binaries as you conjectured in the comment above.
Stepping back a level, you don't seem to be interested in engaging in good faith at all. I'm not going to reply to any more of your comments.
I have no dog in this fight either way, I'm just very curious about the answer: if something like 30-40% in a Go executable that clocks in at more than a 100 megabytes is not taken up by either symbols, debug information or the pclntab, what exactly is in it? You mentioned "necessary metadata for garbage collection and reflection in a statically-compiled language" in a previous comment. Can you give some more details on what that means?
You can see the true size of the Go pclntab in ELF binaries using "readelf -S" and in Mac binaries using "otool -l". Its not zero.
One thing that did change from Go 1.15 to Go 1.16 is that we broke up the pclntab into a few different pieces. Again, it's all in the section headers. But the pieces are not in the actual binary's symbol table anymore, because they don't need to be. And since the format is different, we would have removed the old "runtime.pclntab" symbol entirely, except some old tools got mad if the symbol was missing. So we left the old symbol table entry present, with a zero length.
Clearly, we could emit many more symbols accurately describing all these specific pieces of the binary. But ironically that would just make the binary even larger. Leaving them out is probably the right call for nearly all use case.
Except perhaps trying to analyze binary size, although even there even symbols don't paint a full picture. OK, the pclntab is large. Why is it large? What are the specific things in it that are large? Symbols don't help there at all. You need to analyze the actual data, add debug prints to the linker, and so on.
That would make an interesting post, and perhaps we will write one like that. But not today.
> OK, the pclntab is large. Why is it large? What are the specific things in it that are large?
Is it reasonably easy to attribute individual entries in pclntab to specific symbols? If so I'd love to add this capability to https://github.com/google/bloaty which already tries to do per-symbol analysis of many other sections (eh_frame, rela.dyn, etc).
It's reasonably easy for a specific Go release, but the details can and often do change from release to release. At some point we may write a more detailed, general-purpose binary size analysis tool that could dump JSON for bloaty to import, but today that tool does not exist.
You must know that breaking down and verifying someone else's analysis is more time consuming than writing your own. Just like dealing with a bug in another person's code.
Given them the benefit of doubt that Go team is cautious about binary size. People have dug in to this. Sure, they could do a better job giving some breakdown, but claiming that they are careless deserves that kind of response.
Given a choice of their time, I would rather have them work on some other Go language problem. Most low hanging fruit has already been had. See [1] [2] [3]
> For all we know, the go linker could be embedding random data
To all the people reading this as a literal accusation instead of hyperbole: plase consider that this reading is only possible under the same bad faith that you're attributing to knz42.
Another thing I noticed in the revised blog post on a second skim, regarding this claim:
> Every time, 70% of a couple hundred megabytes are copied around for no good reason and someone needs to pay ingress/egress networking costs for these file copies.
That 70% includes the ELF/DWARF metadata that is easily removed from the binary using strip. It's true that the DWARF info in particular has gotten larger in recent releases, because we've included more information to make debuggers work better. I don't think it has grown quite as rapidly as the table indicates - I think some of the rows already have some of the debug metadata removed in "Raw size".
Regardless, I would hope that anyone sensitive to networking costs at this level would be shipping around stripped binaries, so the growth in accurate DWARF info should not be relevant to this post at all.
That is, the right comparison is to the "Stripped" column in the big table. If you subtract out the strippable overheads and you take the "Dark + pclntab" as an accurate representation of Go-specific overhead (debatable but not today), then the situation has actually improved markedly since Go 1.12, which would have been current in April 2019 when the first post was written.
Whereas in Go 1.12 the measured "actual program" was only about 40% of the stripped binary, in Go 1.16 that fraction has risen to closer to 55%.
This is a marked-up copy of the table from the dr-knz.net revised post that at time of writing has not yet made it to cockroachlabs.com: https://swtch.com/tmp/cockroach-blog.png
I think the numbers in the table may be suspect in other ways, so I am NOT claiming that from Go 1.12 to Go 1.16 there has actually been a 15% reduction in "Go metadata overhead". I honestly don't know one way or the other without spending a lot more time looking into this.
But supposing we accept for sake of argument that the byte counts in the table are valid, they do not support the text or the title of the post. In fact, they tell the opposite story: the stripped CockroachDB binary in question has gotten smaller since April 2019, and less of the binary is occupied by what the post calls "non-useful" or "Go-internal" bytes.
> I would hope that anyone sensitive to networking costs at this level would be shipping around stripped binaries, so the growth in accurate DWARF info should not be relevant to this post at all.
Good point. I removed that part from the conclusion.
> If you subtract out the strippable overheads and you take the "Dark + pclntab" as an accurate representation of Go-specific overhead [...] then the situation has actually improved markedly since Go 1.12 [...] Whereas in Go 1.12 the measured "actual program" was only about 40% of the stripped binary, in Go 1.16 that fraction has risen to closer to 55%.
Ok, that is fair. I will attempt to produce a new version of these tables with this clarification.
> the stripped CockroachDB binary in question has gotten smaller since April 2019, and less of the binary is occupied by what the post calls "non-useful" or "Go-internal" bytes.
There's an explanation for that, which is that the crdb code was also reduced in that time frame.
The article has been updated to avoid calling the bytes "not useful" and to address the narrow, specific points I raised, but it is still generally suspect.
Over on Lobsters, zeebo took the time (thanks!) to build the same version of CockroachDB with various historical Go toolchains, varying only the toolchain, and found that if anything the Go executables are getting smaller over time, in some cases significantly so.
The title of the article ("My Go executable files are still getting larger") appears to be literally untrue, at least read as a critique of Go itself. If they are getting larger, it's because new code is being added, not because the Go runtime or compiler is degrading in some way over time.
> The title of the article ("My Go executable files are still getting larger") appears to be literally untrue, at least read as a critique of Go itself. If they are getting larger, it's because new code is being added, not because the Go runtime or compiler is degrading in some way over time.
Yes this is a fair assessment, although I find it surprising (and enlightening) that you refer to “a critique of Go”. At no moment was the intent to critique Go specifically; the entire analysis is made of observation of the results of combining Go with specific (and varying) amounts of source code.
In any case, based on this discussion I have decided to amend the title and emphasize in the conclusion that the absolute size of the code+data for a fixed amount of source code has decreased between go 1.15 and 1.16.
Yeah, I do not understand those numbers. I just compiled one of my Go projects with version 1.13.8 and version 1.16.3. The size was 8.3 MB (6.5 MB after stripping), and 8.0 MB (5.8 MB after stripping) respectively.
You do NOT write an article talking about "dark bytes" and a 70% "not useful" piece of binary for an Open Source project.
Specially one such as Golang where you can easily reach out to the creators at https://groups.google.com/g/golang-dev to ask for help understanding what your experiment tools can't explain...
Eg.
"""
Hey guys I am doing some digging with "such and such" tools and found that 70% of my binary size has no explanation. BTW this binary is Cockroach DB, an well known program in the Go ecosystem.
How can I find out what are those bytes for?
Is there any documentation about it, other than the source code itself?
Or maybe some tooling that explains it?
Or some other set of tools I should be using instead or on top of my current setup?
"""
I mean... there is no defense possible after writing such an article IMHO. Why didn't you ask first?
Having said that, when you deviate from the ubiquitous C standards for calling conventions and binary format layout, that comes with a price. I totally get that Go had to take such path, because Go binaries embed a non trivial runtime dealing, at the very least, with goroutines & GC support. But the price is there nonetheless: As you are not using a standard, you cannot rely on pre-existing tooling or documentation. You have to write your own.
It is understandable that writing such binary dissection documentation or tooling (or both) might not be a top priority compared to other more pressing matters for the Go Language. But this kind of thing becomes technical debt and as such, the longer you take to pay it, the more expensive it becomes. For instance:
- This happened, and might have damaged the Go Language reputation, at least temporarily for some people.
- I am surprised how could the Go team (or community) achieve improvements at all in the binary size work without tools to properly dissect binaries or detailed documentation. I am guessing it relied on the original programmers of those bits to have a very clear idea of the binary layout uploaded in their brains, so they did not need to re-grok all the sources again... with time that becomes more and more difficult, people forget stuff, move to other projects, etc.
I think it would be wise to prioritize this somehow. If the Go team cannot take it for now, maybe they can promote it as a good Go community project to be done ASAP, and promise to help the takers with all info and reviews needed.
Or maybe the article writers should take up the challenge, to fully redeem themselves ;-p
> The expansion of pclntab in Go 1.2 dramatically improved startup time and reduced memory footprint [...]
yes this is acknowledged in the OP
> We (the Go team) did not “recompress” pclntab in Go 1.15.
There's now less redundancy in the funcdata, so in my book less redundancy = more compression.
> We did not remove pclntab in Go 1.16.
Correct; it is not "removed"; instead the advertised size of the symbol has been reduced to zero. Maybe the data is still there, but it's not accounted any more.
Changed in the text.
(The correction is already present in the original version of the analysis, and the cockroach labs copy should be updated soon)
> we never claimed “pclntab has been reduced to zero”, which is presented in the article as if a direct quote.
Correct, there was indeed no such claim. Removed the quotes and rephrased that paragraph.
> If the 73% of the binary diagnosed as “not useful” were really not useful, a reasonable demonstration would be to delete it from the binary and see the binary still run. It clearly would not.
1) the phrase "non-useful" was a mistake. There is no definite proof it is non-useful, as you pointed out. Corrected in the text.
2) see discussion below - the demonstration is more complicated than that, as removing 100 bytes where just 1 byte is necessary will break the executable in the same way as removing 100 necessary bytes.
I think the proper next step here is to acknowledge that the problem is not "usefulness" but rather accountability.
> The big table seems to claim that a 40 MB Go 1.8 binary has grown to a 289 MB Go 1.16 binary. That’s certainly not the case. More is changing from line to line in that table than the Go version.
Correct. Added a note to emphasize this fact.
> Overall, the claim of “dark bytes” or “non-useful bytes” strikes me as similar to the claims of “junk DNA”. They’re not dark or non-useful.
Let's forget about "non-useful", this was a mistake and will be corrected. The word 'dark' is still relevant however. The adequate comparison is not "junk DNA", but instead "dark silicon":
In this case, what is the point the blog post is trying to make?
The title of the post is "Go Executable Files Are Still Getting Larger". Upon further reading and conversation here it seems this is possibly not true, nor what the post is about. If we believe Russ's comments, Go executable sizes haven't increased much in general. Perhaps the reason you're seeing increases in Cockroach DB is because you keep writing more code for Cockroach DB?
Now the point has shifted to this notion of "dark bytes". So the article is about ... how the way you previously diagnosed the contents of binaries doesn't work anymore? That's fine and legitimate, but it seems like the point is over-extrapolated to become a criticism of the Go team.
> Go executable sizes haven't increased much in general.
Russ's example was just the "gofmt" program.
> Perhaps the reason you're seeing increases in Cockroach DB is because you keep writing more code for Cockroach DB?
If that was the only reason, then the % overhead would remain constant-ish. But it is increasing. So there is a non-linear factor for _some_ go programs (like cockroachdb) and it's still unclear what that factor is.
It's not clear that the overhead is due to Go itself producing bigger binaries over time though. If you recompiled all the different CockroachDB versions with Go 1.8 (if that was feasible), it's quite probable that the tables you would end up with would look fairly similar to the ones you're actually showing.
If there is superlinear growth in binary sizes as the project grows – for example, if some part is O(n^2) in the number of interfaces – then that's certainly interesting. If you demonstrated that such superlinear growth is happening, and wrote an article based on that, people wouldn't be so critical.
If Go binaries are getting bigger because Go produces bigger binaries for the same source code over time, then that's also interesting. If you demonstrated that Go binaries are getting more and more bloated over time for the same source code, and wrote an article based on that, people wouldn't be critical.
But as it is, you kind of just complained that CockroachDB is getting bigger, tried to blame it partly on the Go compiler producing more bloated code over time, partly on a mystical "dark area" which you don't understand, you mentioned superlinear growth only in the comment section, and you didn't actually gather data or do experiments to prove or disprove any of the things you're claiming as a cause. That's why people are complaining.
> tried to blame it partly on the Go compiler producing more bloated code over time
Where? The argument is _precisely_ that the growth is occurring in non-code areas.
> partly on a mystical "dark area" which you don't understand
The _observation_ is that the growth is happening in an area of the file that's not accounted for in the symtable. That's what makes it "dark". It's not mythical: it's _there_ and you can observe it just as well as anyone else.
> you mentioned superlinear growth only in the comment section
it's in the reported measurements.
> and you didn't actually gather data or do experiments to prove or disprove any of the things you're claiming as a cause
The analysis is stating observations and reporting that the size is increasingly due to non-accounted data. That observation is substantiated by measurements. There's no claim of cause in the text!
> Where? The argument is _precisely_ that the growth is occurring in non-code areas.
But how is this important? If the thing you're optimizing for is "total Go binary size", then all that matters is the total size of binary! How bytes are organized internally is irrelevant to this metric.
You should redo the analysis where you compile an old version of Cockroach DB (say v1.0.0) with Go versions 1.8 through 1.16, and then see what the numbers say. Your current analysis, which doesn't account for growth in the code base at all, or tries to account for it by deep-diving into the internal organization of the binary, is not sound.
> all that matters is the total size of binary! How bytes are organized internally is irrelevant to this metric.
Not quite so if the task is to work on reducing the metric.
When the size is attributed to data/code that's linked to the source code, then we know how to reduce the final file size (by removing data/code from the source code, or reducing them).
When the size is non-attributed and/or non-explained (“dark”) we are lacking a control to make the size smaller over time.
You keep saying it’s unexplained as if it’s intentionally kept secret. You pretend you have no control over it, but if you reduced your own source code, you would find that the “dark” space shrunk.
The Go source code is available to you. Russ has pointed out there’s no existing tool to break down those “dark” bytes but that they do serve a purpose, but perhaps you could work on that tool instead of complaining that it’s not covered by the symbol table.
Hmm, I think I understand where we have a misunderstanding. I - and presumably many others - interpreted the article to make the claim that newer Go versions are producing more bloated Go executables. There are multiple parts of the article which can be read that way. But if you're not doing that, and you're just trying to investigate why the CockroachDB binary is getting bigger over time, then that's a different matter.
I'm not going to respond point by point because those points are kind of moot if my accusations were based on an incorrect reading.
CockroachDB has always had a reputation of being slow, 10x-20x slower than the same operation being made in Postgres, with this and the issues about binary size, was a GC language like Go the right choice for CK? Would you have pick something else today if starting a new?
My intuition is that later versions of crdb are more like 1/3rd the efficiency of Postgres per core. GC is some of that but I don’t think it’s all that much.
Everything has trade offs. Go is not the easiest language in which to write highly efficient code. Cockroach generates some code to help here. Certainly at this point there’s pain around tracking memory usage so as to not OOM and there’s pain around just controlling scheduling priorities. But then again, had it been C++ or Rust perhaps getting to table stakes correctness would have taken so long it wouldn’t have mattered.
Some cost just comes from running distributed replication and concurrency control. That’s unavoidable. Some also comes from lack of optimization. Postgres has been around and has some very optimized things in its execution engineer.
Also, if you run Postgres in SERIALIZABLE, it actually does quite badly, largely because that isolation level was bolted on and the concurrency control isn’t optimized for it. Crdb was core-for-core competitive in serializable on some workloads last time I checked.
For being somewhat familiar with the CockroachDB project, I doubt that that claimed performance difference is linked to the programming language. It's more something about mandatory 3-way (or more) replication upon every write, and several additional layers of protection against hardware failures, network problems etc which postgres do not have.
The term "dark silicon" refers to silicon that's not (currently) in use, such as when you have a workload which only exercises one arithmetic unit of one core even if you have 8 cores with 4 arithmetic units each (only one arithmetic unit is "lit up", 63 are "dark"). There's no reason to believe that what you're calling "dark bytes" isn't actively used while executing the program.
Dark silicon is not used all the time - that's the key point.
In the same way, the functab data in Go is not used all the time either, only when generating stack traces.
Also since that original article from 2011 was published, the phrase "dark silicon" was extended to designate silicon IP which is not directly necessary for a given target application but is embedded for the occasional use by a niche in the same market.
In the same way, all languages (not just Go) embed constructs in generated code that are there only to handle edge cases in certain applications/deployments, and are inactive for 90%+ of executions.
Without swap, you mean without virtual memory or without a swap partition/file?
Because even without a swap partition/file, the whole executable will not block physical memory, but will page in/out as needed. And whole sections of it will never be loaded at all.
Yes, any mmap'd file will be swapped in on demand, when a page is first accessed, it will not all be copied in physical memory at once. In case of memory pressure, pages will be removed from physical memory ("swapped out"), since they can be loaded back from the file again when needed.
Since the executable is mapped read-only, the pages loaded in physical memory can also be shared between multiple instances of the process.
As explained in another comment below, it's a good thing when all the byte usage is represented in the ELF section headers or the symbol tables.
The DWARF data is currently so represented, and so was pclntab prior to 1.16.
Today, the DWARF data is still present; the symbol for pclntab has an advertized size of zero, and the amount of data not represented in the symbol table and ELF section headers has grown.
> If the 73% of the binary diagnosed as “not useful” were really not useful, a reasonable demonstration would be to delete it from the binary and see the binary still run. It clearly would not.
Probably not if all of it was removed at once.
It could be that just 5% of it is useful and removing all of it would produce a non-working binary. What does the experiment prove exactly?
Look you can claim that "having the necessary metadata for garbage collection and reflection in a statically-compiled language takes up a significant amount of space" but without clear evidence of how much space that is, with proper accounting of byte usage, this claim is non-falsifiable and thus of doubtful value.
> Look you can claim that "having the necessary metadata for garbage collection and reflection in a statically-compiled language takes up a significant amount of space" but without clear evidence of how much space that is, with proper accounting of byte usage, this claim is non-falsifiable and thus of doubtful value.
The article made the claim that 70% of space is wasted "dark bytes". The article should prove the claim, which it did not. It's an extraordinary claim that really requires more evidence than just an off-hand napkin calculation and talk about mysterious "dark bytes".
It takes very little time to write up something that's wrong.
It takes much more time to write a detailed rebuttal.
What you're doing here is pretty much the same trick quacks, young-earth creationists, and purveyors of all sorts of pseudo-scientific claims pull whenever they're challenged. Any fool can claim the earth is 6,000 years old. Proving that it's actually several billions years old requires deep insight in multiple branches of science. People stopped doing this after a while as it's so much effort and pointless as they're not listening anyway, so now they pull the "aha, you didn't falsify my claim about this or that bullshit I pulled out of my ass therefore I am right" zinger and think they look Very Smart And Scientific™.
But that's not how it works.
Also just outright disbelieving people like this is rather rude. You're coming off really badly here and your comment has the strong implication that Russ is outright lying. Yikes!
> The article made the claim that 70% of space is wasted "dark bytes"
This is incorrect. The claim is that the bytes are either non-accounted, or motivated by technical choices specific to Go.
> What you're doing here is pretty much the same trick quacks [...]
Look the article has some measurements with numbers which you can readily reproduce on your own computer, and the methodology is even described. The main claim is that "it's unclear what these bytes are about". The previous claim that they were "non-useful" was retracted. The data is there, and there's a question: "What is this data about?"
The text is even doubling down by spelling out "there's no satisfying explanation yet".
> outright disbelieving people like this is rather rude
We're not in the business of "believing" or "disbelieving" here I think? There's data, there's measurements, and there are explanations.
After my comments and that of others, Russ provided a more elaborate, more detailed (and at last, falsifiable in the positive, epistemological sense of the word) explanation deeper in the thread. Now we can make the work of looking into it and check the explanation.
> your comment has the strong implication that Russ is outright lying
Your understanding is flawed then? There was no assumption of lies implied.
> The claim is that the bytes are either non-accounted, or motivated by technical choices specific to Go.
That's not what it says; with "70% of a couple hundred megabytes are copied around for no good reason" written in bold no less:
> That’s right! More than two thirds of the file on disk is taken by… bits of dubious value to the software product.
>
> Moreover, consider that these executable files fly around as container images, and/or are copied between VMs in the cloud, thousands of times per day! Every time, 70% of a couple hundred megabytes are copied around for no good reason and someone needs to pay ingress/egress networking costs for these file copies. That is quite some money being burned for no good reason!
So you did make the claim, but just retracted it, in spite of you saying you never made the claim, a claim which is still in the article linked here. I am now supposed to argue against some revised article published elsewhere? This is a very vexing way to have a conversation.
This is also a trick peddlers of pseudoscience pull by the way. Honestly, you're coming off even worse now and this is reflecting pretty badly on all of CockroachDB to be honest. I don't know what your relationship with CockroachDB is exactly (if any), but it's on their website, and it's not a good look. If I was a manager there then I'd back-pedal, unpublish the entire thing, and issue an apology.
> a claim which is still in the article linked here. I am now supposed to argue against some revised article published elsewhere?
The article linked in this thread is a syndicated copy of an original article published elsewhere, as clearly stated by the attribution section at the bottom. It's reasonable to expect that changes to the original will only be updated in the copy with a delay.
This is nowhere near reasonable. You said that "the article made the claim that 70% of space is wasted dark bytes" was "incorrect" with no further details. Only when pressed and provided the quote where you literally said exactly that did you start talking about some retraction.
But whatever, this is pointless. Russ was right and it's hard to take any of this in good faith. It feels like you're going to great lengths to avoid saying "oops, I was wrong".
> Look you can claim that "having the necessary metadata for garbage collection and reflection in a statically-compiled language takes up a significant amount of space" but without clear evidence of how much space that is, with proper accounting of byte usage, this claim is non-falsifiable and thus of doubtful value.
I respectfully disagree. I believe there is value in pointing out the false claims in a long blog post even when there isn't time to write a full-length rebuttal.
Go has a complete accounting of what all the bytes are. You can read them for yourself in the compiler, as it is open source, and trace through where every byte comes from, if you like. It isn't exactly in a broken down table precisely to the specs of what knz42 wants, but the info is all there. There's nothing hidden here.
Oh, you don't want to do that? That's not a surprise. Neither do I. Definitely not on my agenda today.
But I'm not making wild accusations. If you're going to, you probably shouldn't be surprised when we're not exactly impressed.
The compiler is all open, and as such things go, fairly easy to read. "As such things go" is definitely doing some heavy lifting there, but, still, for a compiler of a major language, it's a relatively easy read.
The vague intimation that there's something hidden and sneaky going on is hard to give much credence to.
> As explained in another comment below, it's a good thing when all the byte usage is represented in the ELF section headers or the symbol tables.
It's an ELF binary; all that's relevant are the program/segment headers and dynamic table. The section headers and non-dynamic symbol table are just there for orthogonality and to make the debugger's life a little easier (and the debugger would much prefer DWARF data and doesn't care about the section headers, tbh)
You are getting ridiculous at this point. Probably some things could be done to improve the binary size, but the maintainers of Go don’t have unlimited time. Russ showed that Go is not growing binary sizes. Your article is misleading and at best poorly worded. When you add more code your binaries get bigger. Is that a surprise to you? If Go binary size is a critical concern for you, you could be helping solve the problem, but you are just complaining.
To promote my own tool, https://github.com/bradfitz/shotizam lets you drill down into why Go binaries are large without having to make up terms like "dark bytes".
Because it's genuinely unknown what those are, figuring out what these bytes are is just a deep dive into the opensource go compiler at worst or just asking around on some go channels for much less effort.
This code, if I'm reading it right, uses the symbol table and ELF section headers.
As explained in OP, the sum of sizes advertised in the symtable and ELF section headers does not add up to the final binary size. The shotizam tool is thus blind to that difference.
> The sum of the sizes reported by go tool nm does not add up to the final size of the Go executable.
> At this time, I do not have a satisfying explanation for this “dark” file usage.
The author's journey of starting with "nm --size", discovering "dark" bytes, and wanting to attribute them properly, is exactly what led me to create and invest so much effort into Bloaty McBloatface: https://github.com/google/bloaty
Bloaty's core principle is that every byte of the file should be attributed to something, so that the sum of the parts always adds up to the total file size. If we can't get detailed symbol, etc. information for a given region of the file, we can at least fall back to describing what section the bytes were in.
Attributing all of the bytes requires parsing much more than just the symbol table. Bloaty parses many different sections of the binary, including unwind information, relocation information, debug info, and the data section itself in an attempt to attribute every part of the binary to the function/data that emitted it. It will even disassemble the binary looking for references to anonymous data (some data won't make it into the symbol table, especially things like string literals).
> I excerpted two symbols from the report. Between these two symbols, Bloaty has found seven distinct kinds of data that contributed to these two symbols. If you wrote a tool that naively just parsed the symbol table, you would only find the first of these seven:"
The author's contention that these "dark" bytes are "non-useful" is not quite fair. There are plenty of things a binary contains that are useful even though they are not literally executable code. For example, making a binary position-independent (which is good for security) requires emitting relocations into the binary so that globals with pointer values can be relocated at program load time, once the base address of the binary is chosen. I don't know if Go does this or not, but it's just one example.
On the other hand, I do agree that the ability to produce slim binaries is an important and often undervalued property of modern compiler toolchains. All else being equal, I much prefer a toolchain that can make the smallest binaries.
Bloaty should work reasonably well for Go binaries, though I have gotten some bug reports about things Bloaty is not yet handling properly for Go: https://github.com/google/bloaty/issues/204 Bloaty is just a side thing for me, so I often don't get as much time as I'd like to fix bugs like this.
> In other words, the Go team decided to make executable files larger to save up on initialization time.
I mean... Im genuinely curious if this is a "we have extra engineering resources and can explore/complain about this" or "we have a client who is running cockroachdb and can't handle a 172mb binary install for a database server".
Is there really someone out there who installs Cockroach (a global distributed auto-sharded database) and thinks twice about 172mb of disk space?
Sure, it'd be nice to have smaller binaries but outside of some embedded applications Go's binaries sizes are well within the nothing-burger range for most compute systems.
The article says they don't really care about initialization time though, which is right.
Remember: cockroachdb is always synchronizing data across the cluster, that 175mb of ingress to start up a DB node, probably pales in comparison to data synchronization/relocations that happen on a cluster. Which is why worrying about ingress/egress costs over binary size is nonsense here too.
The bandwidth you need to run a distributed database cluster could download 172mb binary in milliseconds. If your node initiation time for DB failovers needs anything faster, you're doing something wrong.
There are stakeholders to this problem, Cockroach probably isn't one.
For production, yes. But it also effects startup and download time on developer machines. Want to have multiple versions installed? Now it takes more space. Takes longer to download on 4g while on the road or on crappy corporate/conference WiFi, etc .
In the end this all ends up because it is for all go binaries. I’ve come appreciate attention for leanness because in the end it does add up.
This is effectively what is happening btw; the crdb binary went from 80MB to 200MB in the same time it took to make it twice as fast. The % growth in size is not a problem on its own; it's more the % size attributed to the program vs. the % size attributed to unclear purposes, that's a problem.
If most of the bytes are indeed cruft, can't you just send binary diffs around? I imagine they should change that much between compilations? It seems to be an issue for db developers but not for users.
This isn't an interesting or convincing argument because it revolves around a non-satisfiable metric. No matter the size of an artifact, this line of reasoning can always be used to claim it's too big and needs to be smaller.
Tinygo is severely limited, e.g. it doesn't support encoding/json. Want to deal with any JSON API, or anything that indirectly uses JSON serialization? Forget about it.
A current side project of mine uses golang's wasm. Not a big codebase, but the wasm is 2.7MB brotli'ed. Certainly huge to me (I'm sure it's almost in the lean and mean camp compared to the average website today, though).
Yeah, tinygo was originally intended for embedded, so the good news is that many significant wasm binary size optimizations are still to come, I think.
Today, there are ways to get the binary smaller, but they involve avoiding certain libraries or preferring certain implementations of common libraries over others.
It’s definitely an art that requires conscientious practice these days. A run of the mill go application will not compile down to a tiny wasm with tinygo, without refactoring.
But it’s possible. Like I said, it’s on the outskirts, not downtown. Still within reach at least.
3. Unicode: It replaces invalid utf8 with unicode questions marks (dunno how exactly is the replacement character called) but does not provide any API to detect that it happened. So that can lead to silent corruption of the data.
I think it’s well demonstrated to be slow (ns/op) and wasteful (allocs/op) in benchmarks, especially in bigger payloads.
The HTTP library is similar.
Most 3rd party libs make some trade offs that are pretty reasonable to achieve superior perf, but it’s worth understanding them to make sure it’s best for your use case.
Why not? If you use wasm for a small library need in your web app, then sure, go doesn't make sense. But if your whole app is coded in wasm like a game, then it's probably ok I guess as the app will be heavy to load regardless.
So tomorrow if the Go team doubled your binary size by adding "darkest bytes" area with garbage in it your response would be to question the users's inability to be ok with it and not the Go team's choice and reasoning?
The alternative isn't avoiding fixing it. The alternative is a boring bug report on github with actual stakeholders discussing the best strategies and trade offs.
This is only on HN's homepage because of the langauge flame wars. It's a garbage post.
It's possible, but it is the "who cares" mindset I object to. File sizes are important. Memory usage is important. CPU cycles are important. Many devs with powerful machines don't give a heck and in the end everybody has to pay, in different ways.
> Moreover, consider that these executable files fly around as container images, and/or are copied between VMs in the cloud, thousands of times per day! Every time, 70% of a couple hundred megabytes are copied around for no good reason and someone needs to pay ingress/egress networking costs for these file copies. That is quite some money being burned for no good reason!
Does the author think the Go authors are stupid blubbering idiots who someone missed this huge elephant-sized low-hanging fruit? Binary sizes have been a point of attention for years, and somehow missing 70% wasted space would be staggeringly incompetent.
Reminds me of the time in high school when one of the my classmates ended up with a 17A doorbell in some calculations. I think he used the wrong formula or swapped some numbers. The teacher, quite rightfully, berated him for not actually looking at the result of his calculation and judging if it's roughly in the right ballpark, as 17A is a ludicrous amount of power for a doorbell. Anyone can see that's just widely wrong.
If this story had ended up with 0.7%, sure, I can believe that. 7%? Unlikely and I'd be skeptical, but still possible I suppose. *70%* Yeah nah, that's just as silly as a 17A doorbell.
This huge 70% number should have been a clue to the author themselves too that they've missed something.
> Every time, 70% of a couple hundred megabytes are copied around for no good reason and someone needs to pay ingress/egress networking costs for these file copies. That is quite some money being burned for no good reason!
Meanwhile half of the world is pushing images by nvidia, intel and amd around for their machine learning software:
Intel OneAPI runtime libraries: 4.74GB (or 18.4GB for compilers)
CUDA runtime libraries: 1.92GB (or 4.2GB for compilers)
Most of the time, especially with the docker hub rate limiting changes people are or should be using their cloud provider's mirror or running their own mirror. Actually ingressing the docker image into your network should happen once in most production setups.
After looking into the size of the CockroachDB binary, the magnitude of the plank in the author's eye becomes clear. This iceberg is ridiculously bloated. Much of the space is coming from the static data of geographic projections that, I assume, basically nobody needs. This includes a single init function that is 1.4MB of machine code from 6MB of auto-generated source code. Then there's the entire AWS SDK, with a static definition of every service AWS offers, by name, and in what regions, by name. Nevermind the Azure SDK. There are three implementations of protocol buffers in here: gogo in Go and Google's in both Go and C++. There are at least four SQL parsers in here, including vitess's and another one for crdb in Go.
Last but by no means least there are in total 13MB of autogenerated functions of the colexec package, each of which is over 100KB long, which are autogenerated and share virtually all of their code. These are an obscene waste of code space and undoubtedly de-deuplicating this code would not just reduce code size but also speed up the program, due to icache trashing.
I don't think this information in Go is used for debugging alone. It's also used by mallocgc and other parts of the runtime. This is why you will see "gentraceback" in the profiles of busy Go servers. It's not because your program is printing a lot of stack traces, it's because stack walking is a core part of the GC, and because some logic in the runtime refers to the names of functions to branch around special cases.
Yeah, that a function's name begins with "runtime." is 1 bit of information at most. But now we're talking about the optimal representation, not whether this information should exist.
Go includes support for introspection. If Go used DWARF for introspection, and someone stripped DWARF, their code would break. That's not something that should happen with stripping. As such, the Go authors have made the tradeoff to have two "copies" of the data.
No source needed -- you're right. The author's looking in the symbol table, and in 1.16 the Go linker set the size of runtime.pclntab to 0. If the OP used nm to look at their binary, they'd have seen that pclntab is still there. They author has apparently revised their article to address some of this (yet still draw some incorrect conclusions).
Hey all -- Cockroach Labs blog editor here. Based in part on the feedback we received from this community, we have retracted the post. The blog post link above will take you the retraction, where we share what we've learned from this experience.
This article should be renamed “what is Chesterton‘s fence?” And the author should have realized their mistake right after typing “at this time I don’t know what it is used for”.
I always wonder if this is the flip side of the fast compilation?
It would be nice to be able to decide on those trade-offs ourselves. I mostly write web servers in Go, which (as the article says) are executed rarely, so init time really doesn't matter to me. But I've been looking at writing some desktop apps in Go, and then init time will matter.
More likely the flip side of statically compiling every library in. Even if the Go compiler did the equivalent of -Os you'd still be getting binaries that people were complaining about.
I know for a good long time Go wasn't even stripping out unused functions, e.g., if you used any of the "strings" standard module you got the whole thing compiled in. I don't know if that's still the case, but it would be another source of size problems.
I'm also not sure why you're talking about init time; binary size doesn't necessarily factor in to that very much. The things that make the Go binaries large also tend to lean in favor of faster initialization anyhow. A lot of these binaries won't even necessarily be read in to memory, or have to stay there in the event of swap pressure, since they're already disk-backed memory pages that can be never loaded in the first place, or cheaply evicted by the virtual memory system if space is needed.
At least for CLI-tool-sorta-programs, I find Rust and Go emit vaguely similar sized binaries, a few Mb is typical. The biggest cause is of course static linking.
However, once you exceed the 10-20Mb regime, I seem to find more 20-200Mb Go bins in the wild. This can be misleading, since the Go world is a fan of packing assets into binaries, eg for web GUIs.
What dings Go bins on init time is the Go runtime. Still, it's rarely noticeable unless you are firing thousands of times a second.
> I mostly write web servers in Go, which (as the article says) are executed rarely, so init time really doesn't matter to me.
Presumably a few mb of disk usage also doesn't matter then? I also write web services, but I do care about the init time precisely because I want to be deployments to take as little time as possible so that we can deploy (and rollback) many times per day with relatively simple automation. That said, the bottleneck to fast deployments isn't the binary starting up, but the machine pulling the executable artifact, so the binary sizes do matter to me. That said, very often these executable artifacts are a Docker image, which tend to come with a lot more bloat than one will find in a Go binary, so step 1 is getting your Go binary on a scratch image.
Yeah, it's a trade-off. At the moment the longest part of my deploy is copying the new executable to the server. I'd trade a couple of seconds of init time for a smaller executable because that would result in a faster deploy.
I don't use Docker to deploy, because it's just a single executable file (and a bunch of templates, though I'm looking at embedding those). One of the reasons I'm reluctant to go down the Docker road is because it's going to add more time to my deployment.
I still use Docker images to deploy several of my Go systems. For one, it's nice to have it integrated into other ecosystems where a "Docker image" is just the base level of functionality. The additional security (even if not necessarily perfect) and isolation isn't all bad either.
It's perfectly fine to compile a Go binary and stick it into a Docker container on its own; it is not obligatory for a Docker container to contain a full Linux installation. I've got a couple of Docker files that are pretty much just
FROM scratch
ADD my_binary /
(Usually my bind mounts and the command executed are set up elsewhere for logging and such.)
It is also a convenient way of knowing exactly what your dependencies are... for instance I have several Go containers that also have to include a trusted cert store so they can access remote HTTPS resources correctly. Since you don't need a full Linux install to run a Go binary, it's very easy to use Docker as a complete, human-comprehensible manifest of exactly what is in that container.
> I don't use Docker to deploy, because it's just a single executable file (and a bunch of templates, though I'm looking at embedding those). One of the reasons I'm reluctant to go down the Docker road is because it's going to add more time to my deployment.
Yeah, I don't advocate Docker for its own sake, but my organization deploys everything via Kubernetes because it's simpler than having a bespoke orchestration strategy/toolchain for each project, but if you're a one-project shop then containerization probably doesn't add a lot of value (although I still haven't figured out a streamlined, reliable way to provision/configure VMs).
Unless you are running these in containers, then that time adds up (assuming you are using go-based container runtime implementations, as most people are).
No. The two are mostly orthogonal. Sure, you can take a longer time to compile small binaries, but that's not what Go does. OP is talking about internal data structures Go uses for introspection and GC. The amount of time time spent compiling would have a marginal effect on the size of those structures.
> I always wonder if this is the flip side of the fast compilation?
Go is cheating a bit on this one by heavily caching everything. Building in
fresh container is quite slow (ok, maybe not c++ slow but still much slower
then C).
Always having to build anything due to static linking does not help either.
> I always wonder if this is the flip side of the fast compilation?
I don't think there's an easy either-or answer to questions like this, but broadly language design seems to be about finding a sweet spot while balancing lots of trade offs, so to get fast compiles you're going to make trade-offs in other areas of the language and implementation. I imagine if the compilation time budget was higher there'd be some more space for binary pruning passes.
totally agree. Being able to hit the "run this test" and not having to wait for it to compile is awesome :)
But if there some flags to tell the compiler to compile slowly and produce a smaller executable, that would be awesome for final builds that need to be shipped across the internet.
Probably falls under the umbrella issue filed by Rob Pike…
back in 2013[1]. But that's the umbrella issue, and it's
probably better for the Cockroach people to file a new
one mentioning their findings, but don't quote me on that.
We need more initContainers where the big shared system libraries are in a separate container that is cached locally. I feel like history is repeating.
It won't. If you try to remove this information your program will crash as soon as the GC runs or someone calls systemstack, i.e. pretty much instantly.
> Every time, 70% of a couple hundred megabytes are copied around for no good reason and someone needs to pay ingress/egress networking costs for these file copies.
Just zero them out and they'll compress to nothing. Even better, with a sparse file aware tool like tar, they won't even use disk space.
> We can call this the “dark file usage” of Go binaries, and it occupies between 15% and 33% of the total file size inside CockroachDB.
> Sadly, the removal of pclntab in Go 1.16 actually transferred the payload to the “dark” bytes.
I surely would have expect better from programming language designers/developers than this. Sounds like they just moved the problem from one place to another.
I'll hold my judgement until someone manages to actually find out what this "dark" space is (quote from the article: "At this time, I do not have a satisfying explanation for this “dark” file usage").
- The expansion of pclntab in Go 1.2 dramatically improved startup time and reduced memory footprint, by letting the OS demand-page this critical table that is used any time a stack must be walked (in particular, during garbage collection). See https://golang.org/s/go12symtab for details.
- We (the Go team) did not “recompress” pclntab in Go 1.15. We did not remove pclntab in Go 1.16. Nor do we have plans to do either. Consequently, we never claimed “pclntab has been reduced to zero”, which is presented in the article as if a direct quote.
- If the 73% of the binary diagnosed as “not useful” were really not useful, a reasonable demonstration would be to delete it from the binary and see the binary still run. It clearly would not.
- The big table seems to claim that a 40 MB Go 1.8 binary has grown to a 289 MB Go 1.16 binary. That’s certainly not the case. More is changing from line to line in that table than the Go version.
Overall, the claim of “dark bytes” or “non-useful bytes” strikes me as similar to the claims of “junk DNA”. They’re not dark or non-useful. It turns out that having the necessary metadata for garbage collection and reflection in a statically-compiled language takes up a significant amount of space, which we’ve worked over time at reducing. But the dynamic possibilities in reflection and interface assertions mean that fewer bytes can be dropped than you’d hope. We track binary size work in https://golang.org/issue/6853.
An unfortunate article.