Comments (4)
Personally, I think we actually want to have a custom build Bazel. This would give us even better control over the whole build environment.
Concerning the issues you raised, I think it's ok to prioritize the nix workflow for now, since I don't see a huge drawback with using nix. The few MB for libc++ should be no problem either, since storage is quite inexpensive.
I think it is the spirit of rules_ll to provide the most advanced toolchain possible, so I'm happy to prioritize remote execution over those minor inconveniences.
from rules_ll.
Statically linking libc++ is fine, those few MB are neglectable in comparison to the cache and nix environment size. We should keep dynamical linking in mind if image size becomes an issue in the future.
Yes, we should aim for a custom build Bazel, this aligns well with the rest of the rules_ll project. Could you go into further detail on how you want to handle the patching and building of Bazel?
from rules_ll.
@jaroeichler I initially tried just patching the RPATHs with patchelf, but then bazel refuses to operate. Probably for security reasons.
My current plan is:
- Write a nix package that builds bazel via the non-upstream llvm toolchain from nixpkgs and statically link libc++ into it.
- Distribute that Bazel in a way that is compatible with Bazelisk's custom release mechanism outlined here.
- Fetch the binary in our remote execution images via bazelisk.
If things work as i intend, we'd end up with remote execution images that no longer require libstdc++ or any gcc-toolchain parts. If we can reference these custom Bazel binaries in .bazelversion
this approach should also be portable to non-nix users as long as the LLVM toolchain parts are statically linked.
As an interesting sidenote we could also try to statically link libmusl into that release to create a fat binary that is independent of the host's glibc version. But let's leave this for later when things actually work
from rules_ll.
Ok remote execution works, so we could run tests in CI. But that might be really expensive. A single build with near perfect cache reuse (which we basically always have) still needs ~2GB of artifacts to operate (makes sense, building a single target requires the tools from the ll_toolchain
, which is roughly that size). At 1 commit per day this is ~60GB just for the main branch. This does not include any PR testing etc. This is also a minimum value. For instance, updating LLVM alone which requires a full cache rebuild and a few revisions might be many times larger than that.
We probably still need a fraction of the resources that others would need for a similar setup, but it's still a big setup. We it might be better off hosting our own remote exec cluster.
from rules_ll.
Related Issues (20)
- GPU targets run correctly but tests fail HOT 1
- Module override for circl breaks downstream users
- Investigate the use of aspects for clang-tidy
- Borrowing an Nvidia GPU HOT 2
- Dependency Dashboard
- Readd vale HOT 1
- The `ll init` command is a bit whacky HOT 2
- 😵 GPU examples make my eyes bleed
- Shellcheck pre-commit hook might be broken HOT 1
- Document usage with local CUDA HOT 2
- Migrate CUDA imports to new variants in nixpkgs HOT 4
- Migrate to zlib-ng HOT 2
- We can't ignore example lockfile but also can't commit it
- Improve user experience for the `rbegen` tool HOT 1
- Document new remote execution toolchains HOT 1
- Release blockers
- Consider building our own remote execution service HOT 3
- Remote execution images too hard to customize
- Rework internal file inputs as preparation for module std HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from rules_ll.