Skip to content

Respect jobserver set by Cargo#120

Merged
BusyJay merged 4 commits intotikv:mainfrom
Kobzol:jobserver
Jun 9, 2025
Merged

Respect jobserver set by Cargo#120
BusyJay merged 4 commits intotikv:mainfrom
Kobzol:jobserver

Conversation

@Kobzol
Copy link

@Kobzol Kobzol commented Mar 3, 2025

The jemalloc-sys crate currently does not respect the Make jobserver protocol, and hardcodes -j$NUM_JOBS Make jobs as set by Cargo.

Cargo recommends to use CARGO_MAKEFLAGS instead (https://doc.rust-lang.org/cargo/reference/environment-variables.html), to avoid oversubscribing CPU cores when building Jemalloc C++ code in parallel with other Rust crates.

Here is a small benchmark that I did on hyperqueue with and without jobserver protocol support on AMD Zen 3 with 8 cores enabled:

/pr/it/hyperqueue [main/u]$ hyperfine --warmup 1 --runs 2 --prepare "rm -rf target" "cargo build --release" "JEMALLOC_JOBSERVER=1 cargo build --release" "cargo build" "JEMALLOC_JOBSERVER=1 cargo build"
Benchmark 1: cargo build --release
  Time (mean ± σ):     57.620 s ±  0.010 s    [User: 334.019 s, System: 37.222 s]
  Range (min … max):   57.613 s … 57.626 s    2 runs
 
Benchmark 2: JEMALLOC_JOBSERVER=1 cargo build --release
  Time (mean ± σ):     56.969 s ±  0.017 s    [User: 324.140 s, System: 36.739 s]
  Range (min … max):   56.956 s … 56.981 s    2 runs
 
Benchmark 3: cargo build
  Time (mean ± σ):     45.410 s ±  0.056 s    [User: 193.525 s, System: 37.347 s]
  Range (min … max):   45.371 s … 45.450 s    2 runs
 
Benchmark 4: JEMALLOC_JOBSERVER=1 cargo build
  Time (mean ± σ):     44.039 s ±  0.011 s    [User: 194.451 s, System: 37.357 s]
  Range (min … max):   44.031 s … 44.046 s    2 runs

With jobserver protocol support, the compilation finishes 2-3% faster.

@Kobzol
Add jobserver support
7994e07 
Signed-off-by: Jakub Beránek <[email protected]>
@Kobzol
Copy link
Author

Kobzol commented Mar 3, 2025

Added the signoff.

BusyJay
BusyJay reviewed Jun 4, 2025
View reviewed changes
BusyJay
BusyJay reviewed Jun 9, 2025
View reviewed changes
BusyJay
BusyJay approved these changes Jun 9, 2025
View reviewed changes
Copy link
Member

@BusyJay BusyJay left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you!

@BusyJay BusyJay merged commit 5c5a9f7 into tikv:main Jun 9, 2025
10 checks passed
@Kobzol Kobzol deleted the jobserver branch June 9, 2025 14:01
@goffrie goffrie mentioned this pull request Sep 17, 2025
Open
@eitsupi eitsupi mentioned this pull request Nov 15, 2025
Open
10 tasks
@reneleonhardt
Copy link

reneleonhardt commented Nov 15, 2025 •
edited
Loading

MAKEFLAGS have not been sanitized, so the jobs fail.
-j and -jN and -j N have to be removed when --jobserver-fds is present.

MAKEFLAGS="-j --jobserver-fds=8,9"

#92 (comment)

│ │ running: cd "$SRC_DIR/src/rust/target/x86_64-unknown-linux-gnu/release/build/tikv-jemalloc-sys-cde55cd13423d9e3/out/build" && MAKEFLAGS=" -- OBJECTS=init.o SHLIB=polars.so -j --jobserver-fds=8,9 --jobserver-auth=8,9" "make"
│ │ make[1]: Entering directory '$SRC_DIR/src/rust/target/x86_64-unknown-linux-gnu/release/build/tikv-jemalloc-sys-cde55cd13423d9e3/out/build'
│ │ make[1]: Leaving directory '$SRC_DIR/src/rust/target/x86_64-unknown-linux-gnu/release/build/tikv-jemalloc-sys-cde55cd13423d9e3/out/build'
│ │ --- stderr
│ │ make[1]: *** No rule to make target '-j'. Stop.
│ │ thread 'main' (6599) panicked at $BUILD_PREFIX/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/tikv-jemalloc-sys-0.6.1+5.3.0-1-ge13ca993e8ccb9ba9847cc330696e02839f328f7/build.rs:411:9:
│ │ command did not execute successfully: cd "$SRC_DIR/src/rust/target/x86_64-unknown-linux-gnu/release/build/tikv-jemalloc-sys-cde55cd13423d9e3/out/build" && MAKEFLAGS=" -- OBJECTS=init.o SHLIB=polars.so -j --jobserver-fds=8,9 --jobserver-auth=8,9" "make"

@Kobzol
Copy link
Author

Kobzol commented Nov 24, 2025

Let's discuss in #92.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@BusyJay BusyJay BusyJay approved these changes

Assignees

No one assigned

Labels

dco-signoff: yes

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@Kobzol @reneleonhardt @BusyJay