nix-store: implement --query --unsubstitutable #7526

nrdxp · 2022-12-29T04:10:55Z

Resolves #3946 by providing an efficient means of querying cache status of the passed paths.

For every path passed, it will be returned back on standard output only if it is not available in any of the configured substituters.

~~A threadpool is used to make this as quick as possible.~~

Included in the documentation is a sample command showing how users can now trivially capture any and all uncached build & runtime dependencies of a given derivation. This information is highly useful in CI environments, for example.

Only caveat is that we might hit the OS arg limit until #7437 is fixed.

TODO

implement nix-store query
write tests

Checklist for maintainers

Maintainers: tick if completed or explain if not relevant

agreed on idea
agreed on implementation strategy
tests, as appropriate
- functional tests - tests/**.sh
- unit tests - src/*/tests
- integration tests - tests/nixos/*
documentation in the manual
code and comments are self-explanatory
commit message explains why the change was made
new feature or bug fix: updated release notes

nixos-discourse · 2022-12-29T04:19:51Z

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/how-to-correctly-cache-build-time-dependencies-using-nix-haskellforall-com/22726/5

mlyxshi · 2022-12-29T16:04:25Z

Are you interested in this relative issue?
#7527

nrdxp · 2022-12-29T19:57:33Z

Are you interested in this relative issue? #7527

You could use this PR to accomplish your goal by simply processing your derivations through nix-store -qU (see the example I added in the doc for how to get build & runtime dependencies) before sending them to nix copy. As Sandro rightly pointed out though, it will still result in a few redundant dependencies, since if the entire dependency closure does not exist in a single cache, Nix throws an error. To improve the situation we just have to allow Nix to continue instead of erroring in this situation, and you could then just pass a --no-recursive to nix copy and cache only those that don't exist elsewhere.

In general, my recent experience writing a Nix-centric CI has kind of changed my perspective somewhat, and I'm starting to feel like we should take a more "git-like" approach to our cli, rather than replacing the older, more verbose commands outright, we should stabalize them as much as possible and use them as the "plumbing" to the higher level commands.

nrdxp · 2022-12-30T00:04:50Z

Because there isn't really a nice standalone solution for this atm, I went ahead and extracted the logic as a standalone tool for now until this is accepted:
https://github.com/divnix/nix-uncached

edolstra · 2022-12-30T14:58:10Z

I think functionality like this should be added to the new CLI rather than the legacy CLI. It seems to me that what we really want here is a CLI interface to Store::queryMissing() (ideally with JSON output).

Also "uncached" is a vague term. What does it mean for a path to be uncached? Maybe "unsubstitutable" is better.

nrdxp · 2022-12-30T18:46:06Z

I think functionality like this should be added to the new CLI rather than the legacy CLI. It seems to me that what we really want here is a CLI interface to Store::queryMissing() (ideally with JSON output).

I can add it to nix path-info I think, fairly easily. Are these "legacy" commands intended to be completely removed at some point? My first pass I did use JSON, but it seemed a bit superflous. Since nix path-info already has a json flag though, I can probably include this information as an extra boolean in that output when passed.

Even so, nix path-info doesn't seem to have an equivalent atm for nix-store -qR --include-outputs, so I would still need that command to forward to the list to check in nix path-info.

Also, my first implementation did just use Store::queryMissing(), but that isn't exactly what I wanted, since I don't care if the path exists locally or not, I just want to know if it exists in a substituter (so it can be sent to nix copy to add it there, for example). If you want to include the local store as part of the equation you can add --extra-substituters auto to the invocation to make that explicit.

Also "uncached" is a vague term. What does it mean for a path to be uncached? Maybe "unsubstitutable" is better.

Yeah I was thinking that myself, for now I just used the terminology that has been used elsewhere in the community so far, but I was also thinking of using unsubstitutable instead just last night, only thing is that it's quite long, although I guess with a short flag it would still be fine.

nrdxp · 2022-12-31T01:36:40Z

@edolstra, I implemented the same logic for nix path-info, including allowing it the --json to add a new boolean output showing whether a path is cached or not.

However there are a few downsides to the nix path-info implementation compared to the nix-store one so far. In order to explain that, understand that my main usecase for this is a novel CI system which querys information about the paths before they have been built, to decide if they need to be built, etc. I acquire the paths I need from a flake using nix eval which produces a special json we then want to manipulate based on information we are going to query, like the cache status.

Importantly, I only want to run evaluation once, and only once, for a given CI run, so passing flake refs around isn't desirable as they may induce additional evaluation. Instead we query information on the derivations directly. This just doesn't work well with nix path-info.

With the nix-store implementation it works perfectly, I can just query the derivation with nix-store -qR --include-outputs and then filter that through my new nix-store -qU and I've got the information I need without having to build a thing, minus some additional post-processing which can be easily done with jq and nix show-derivation (to remove locally built derivations from the equation, etc).

Specifically these are the quircks I've encountered with the nix path-info implementation so far:

as already mentioned, there is no equivalent to the --include-outputs flag, so its either querying buildtime or runtime deps, not both.
If I pass a derivation (which is all I have), then the command will not return information about it's outpath. I have to explicitly pass the outpath for that, but that is a non-starter for my usecase since I don't have them built yet (and don't want to necessarily). This means there is no way to mimic the behavior of --include-outputs by simply calling nix path-info twice (with and without --derivation) since passing an outpath that doesn't exist just induces an error.

For these reasons I propose we keep both implementations until both are more in-line, and as mentioned in my previous comment, we may want to consider a more "plumbing" and "porceline" approach for flakes 2.0, to keep drift between the old and new commands from becoming a big issue.

I like the low-level control the legacy commands give me when I need that. Say in a complex CI system that focuses on doing as little work as possible, but I like the higher level commands for ease of use when working on my own system. I think there is an argument to be made for keeping both around and taking implementations from the lower-level commands and reusing them for their flake counterparts.

nixos-discourse · 2022-12-31T02:04:38Z

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/what-would-you-like-to-see-improved-in-nix-cli-experience/24012/58

tfc · 2022-12-31T15:05:18Z

src/nix/path-info.cc

+                    clone["substitute"] = !infos.empty();
+
+                    auto jsonList(jsonList_.lock());
+                    jsonList->emplace_back(clone);


in order to really emplace it at the back without unneeded copies i suggest this:

Suggested change

jsonList->emplace_back(clone);

jsonList->emplace_back(std::move(clone));

src/nix-store/nix-store.cc

tfc · 2022-12-31T15:07:15Z

src/nix/path-info.cc

+                }
+
+                std::function<void(nlohmann::json::basic_json::value_type)> checkSubs;
+                checkSubs = [&](const nlohmann::json::basic_json::value_type & value) {


same suggestion with the nested lambda expression here. all the type info is relatively noisy and can be reduced.

tfc · 2022-12-31T15:08:22Z

src/nix/path-info.cc

+                    auto clone = value;
+                    std::string storePathS = value["path"];
+                    auto info = store->queryPathInfo(store->followLinksToStorePath(storePathS));
+                    auto storePath = info->path;


it seems like some const ... & could help the reader understand what really needs to be a (mutable) copy and what not, as well as reduce some redundant copying.

tfc · 2022-12-31T15:11:12Z

src/nix/path-info.cc

+            Sync<std::set<std::string>> results_;
+
+            std::function<void(StorePath)> printPath;
+            printPath = [&](const StorePath & storePath) {


Here, the parameter type from the std::function and the actual assignment differ by having and not having const ... & qualifiers. i suggest using the same nested lambda expression approach here, too.

tfc · 2022-12-31T15:11:58Z

src/nix/path-info.cc

                }

-                std::cout << std::endl;
+                auto results(results_.lock());
+                results->emplace(storePathS);


using std::move will help avoiding a needless copy.

tfc · 2022-12-31T15:14:35Z

src/nix/path-info.cc

+            auto results(results_.lock());
+            for (auto & str : *results) {
+                std::cout << std::endl << str;
+            }


std::endl means to emit a new line character and flush the terminal. But this code suggests that you mean to separate all strings by newlines and then flush only once at the end, so i suggest this:

Suggested change

auto results(results_.lock());

for (auto & str : *results) {

std::cout << std::endl << str;

}

for (const auto & str : *results_.lock()) {

std::cout << '\n' << str;

}

std::cout << std::endl;

nrdxp · 2023-01-01T00:33:45Z

Thanks for the review, I'm moving to draft for now as the implementation needs work. After some initial benchmarks, the threadpool didn't actually help anything and is about the same as using a single thread, even when setting the threadpool limit to something super high like 1000.

nrdxp · 2023-01-10T17:34:48Z

~~Taking out of draft as after some testing it seems the performance issues already existed and were not introduced by me~~: #7575

I will go ahead and continue my effort here, address the review comments, and possibly simplify the implementation by just removing the threadpool unless/until we find a solution to the performance issues. That should probably be handled in a separate PR though anyway.

Resolves NixOS#3946 by providing an efficient means of querying cache status of the passed paths. For every path passed, it will be returned on standard output only if it is not available in any of the configured substituters.

edolstra · 2023-01-20T12:24:53Z

src/nix-store/nix-store.cc

+
+            StorePathSet storePaths;
+            for (auto const & str : opArgs) {
+               storePaths.emplace(store->followLinksToStorePath(str));


The indentation here and below is wrong (should be 4 characters per level).

edolstra · 2023-01-20T12:25:25Z

src/nix-store/nix-store.cc

+            }
+
+            for ( auto & path : storePaths) {
+                if (substitutablePaths.find(path) == substitutablePaths.end())


Suggested change

if (substitutablePaths.find(path) == substitutablePaths.end())

if (!substitutablePaths.count(path))

This does not short-circuit

In what way? On an std::map it should be equally efficient.

But since we're fine with C++20 now, you can also use substitutablePaths.contains(path).

tomberek · 2023-01-20T12:27:10Z

adding to make the cross-link easier to find: #7587

nrdxp · 2023-02-11T17:52:13Z

closing in favour of the aforementioned PR

tomberek · 2023-02-11T18:09:12Z

Miss click

nixos-discourse · 2023-02-13T14:51:15Z

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/2023-01-20-nix-team-meeting-minutes-25/25432/1

nrdxp requested review from fricklerhandwerk, edolstra and thufschmitt as code owners December 29, 2022 04:10

fricklerhandwerk added feature Feature request or proposal cli The old and/or new command line interface labels Dec 30, 2022

tfc reviewed Dec 31, 2022

View reviewed changes

src/nix-store/nix-store.cc Show resolved Hide resolved

tfc reviewed Dec 31, 2022

View reviewed changes

nrdxp marked this pull request as draft January 1, 2023 00:31

nrdxp marked this pull request as ready for review January 10, 2023 17:33

nrdxp force-pushed the uncached branch from 00fba19 to 10b0f62 Compare January 11, 2023 20:45

nrdxp changed the title ~~nix-store: implement --query --uncached~~ nix-store: implement --query --unsubstitutable Jan 11, 2023

nrdxp force-pushed the uncached branch 2 times, most recently from 6d692d5 to 65e57dd Compare January 11, 2023 22:36

nix-store: implement --query --unsubstitutable

76467e1

Resolves NixOS#3946 by providing an efficient means of querying cache status of the passed paths. For every path passed, it will be returned on standard output only if it is not available in any of the configured substituters.

nrdxp force-pushed the uncached branch from 65e57dd to 76467e1 Compare January 11, 2023 23:03

nix-store: query all caches, not just the first

a84d4ad

edolstra reviewed Jan 20, 2023

View reviewed changes

nrdxp closed this Feb 11, 2023

tomberek reopened this Feb 11, 2023

tomberek closed this Feb 11, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

nix-store: implement --query --unsubstitutable #7526

nix-store: implement --query --unsubstitutable #7526

nrdxp commented Dec 29, 2022 •

edited by roberth

Loading

nixos-discourse commented Dec 29, 2022

mlyxshi commented Dec 29, 2022

nrdxp commented Dec 29, 2022 •

edited

Loading

nrdxp commented Dec 30, 2022

edolstra commented Dec 30, 2022

nrdxp commented Dec 30, 2022 •

edited

Loading

nrdxp commented Dec 31, 2022 •

edited

Loading

nixos-discourse commented Dec 31, 2022

tfc Dec 31, 2022

tfc Dec 31, 2022

tfc Dec 31, 2022

tfc Dec 31, 2022

tfc Dec 31, 2022

tfc Dec 31, 2022 •

edited

Loading

nrdxp commented Jan 1, 2023 •

edited

Loading

nrdxp commented Jan 10, 2023 •

edited

Loading

edolstra Jan 20, 2023

edolstra Jan 20, 2023

ncfavier Jan 20, 2023

edolstra Jan 30, 2023

tomberek commented Jan 20, 2023

nrdxp commented Feb 11, 2023

tomberek commented Feb 11, 2023

nixos-discourse commented Feb 13, 2023

	jsonList->emplace_back(clone);
	jsonList->emplace_back(std::move(clone));

	if (substitutablePaths.find(path) == substitutablePaths.end())
	if (!substitutablePaths.count(path))

nix-store: implement --query --unsubstitutable #7526

nix-store: implement --query --unsubstitutable #7526

Conversation

nrdxp commented Dec 29, 2022 • edited by roberth Loading

TODO

Checklist for maintainers

nixos-discourse commented Dec 29, 2022

mlyxshi commented Dec 29, 2022

nrdxp commented Dec 29, 2022 • edited Loading

nrdxp commented Dec 30, 2022

edolstra commented Dec 30, 2022

nrdxp commented Dec 30, 2022 • edited Loading

nrdxp commented Dec 31, 2022 • edited Loading

nixos-discourse commented Dec 31, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tfc Dec 31, 2022 • edited Loading

Choose a reason for hiding this comment

nrdxp commented Jan 1, 2023 • edited Loading

nrdxp commented Jan 10, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tomberek commented Jan 20, 2023

nrdxp commented Feb 11, 2023

tomberek commented Feb 11, 2023

nixos-discourse commented Feb 13, 2023

nrdxp commented Dec 29, 2022 •

edited by roberth

Loading

nrdxp commented Dec 29, 2022 •

edited

Loading

nrdxp commented Dec 30, 2022 •

edited

Loading

nrdxp commented Dec 31, 2022 •

edited

Loading

tfc Dec 31, 2022 •

edited

Loading

nrdxp commented Jan 1, 2023 •

edited

Loading

nrdxp commented Jan 10, 2023 •

edited

Loading