Forward-merge release/26.06 into main#1293
Merged
Merged
Conversation
This PR cleans #1252, so only the changes related to the OpenMP are present. ## Original description: `run_concurrent` in `cpp/src/pdlp/solve.cu` now dispatches the barrier and dual simplex workers as `#pragma omp task` inside a `#pragma omp taskgroup` instead of raw `std::thread`. PDLP still runs synchronously on the dispatching thread. MIP path (`omp_in_parallel()`): reuses the upstream `solve_mip` OMP team. Barrier and dual simplex now consume slots from the configured `num_cpu_threads` budget instead of spawning extra OS threads outside it. Stand-alone LP path: stands up a local `#pragma omp parallel + single` with the right number of workers. This PR also removes the confuscated `std::future` logic on the barrier on other PR. Authors: - Nicolas L. Guidotti (https://github.com/nguidotti) - Akif ÇÖRDÜK (https://github.com/akifcorduk) Approvers: - Alice Boucher (https://github.com/aliceb-nv) URL: #1291
Contributor
Author
|
FAILURE - Unable to forward-merge due to an error, manual merge is necessary. Do not use the IMPORTANT: When merging this PR, do not use the auto-merger (i.e. the |
(Claude-driven bug fix, potentially fixing the crash in https://github.com/NVIDIA/cuopt/actions/runs/26249434727/job/77262219725?pr=1103. The issue seems rare enough that I wasn't able to reliably reproduce locally to confirm the fix.) `init_logger_t` reconfigures the process-global `cuopt::default_logger()` sinks under `g_guard_mutex`, but `CUOPT_LOG_*` calls go through that same global logger without holding the mutex. Three risky entry points existed: - `call_batch_solve()` ran an OpenMP loop where every worker constructed its own `init_logger_t` inside `solve_lp()`. When workers finished at different times the last guard's destructor reset sinks while siblings were still logging — a likely source of rare `pure virtual method called -> std::terminate -> SIGABRT` aborts. - `solve_lp()` and `solve_qp()` called `print_version_info()` (which uses `CUOPT_LOG_*`) before constructing their `init_logger_t`, so version logging ran through whatever global sink configuration happened to be active at that moment. Fix: - Construct an outer `init_logger_t` at the top of `call_batch_solve()` before any `CUOPT_LOG_*` and before the OpenMP region. Worker-local `init_logger_t` instances now ref-count onto it and never tear down global sinks. - Reorder `solve_lp()` and `solve_qp()` so `init_logger_t` is constructed before `print_version_info()`. Verified by running `test_parser_and_batch_solver` 300x in one process and 2x300 in concurrent processes without any abort signature. The changes to `solve_lp` and `solve_qp` are incidental and probably not related to the original crash. They're preventing a potential similar issue if `solve_lp` or `solve_qp` are called from multiple threads. Authors: - Miles Lubin (https://github.com/mlubin) Approvers: - Rajesh Gandham (https://github.com/rg20) URL: #1286
Switch the wheel build to link gRPC against OpenSSL 3. This is good for modernization, and also necessary to enable FIPS compliance on many current OS platforms. * install_protobuf_grpc.sh: on Rocky/RHEL 8, install epel-release and openssl3-devel. Rocky/RHEL 9 and Ubuntu/Debian already have openssl-devel at version 3 so use that. * build_wheel_libcuopt.sh: Add libssl.so.3 / libcrypto.so.3 to the auditwheel exclude list so the wheel does not bundle them; runtime resolves to the host's OpenSSL 3, ensuring libcrypto and any FIPS provider (system or mounted) stay byte-version-matched. Verified that the resulting binary links require only OPENSSL_3.0.0 symbol versions (159 refs, 0 newer), so the wheel is ABI-compatible with the cuopt container's Ubuntu 22.04 (libssl 3.0.2), RHEL/Rocky 9 (3.0.7), and any OpenSSL 3.x runtime. Hosts must provide libssl.so.3 / libcrypto.so.3 (Ubuntu 22.04+, Debian 12+, RHEL/Rocky/Alma 9+, Fedora 36+ provides). Stock RHEL/Rocky 8 and Ubuntu 20.04 are not covered by this; users on those systems should either use the cuopt container image or install OpenSSL 3 from EPEL or a backport before pip-installing cuopt. Authors: - Trevor McKay (https://github.com/tmckayus) - Ramakrishnap (https://github.com/rgsl888prabhu) Approvers: - Ramakrishnap (https://github.com/rgsl888prabhu) URL: #1188
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Forward-merge triggered by push to release/26.06 that creates a PR to keep main up-to-date. If this PR is unable to be immediately merged due to conflicts, it will remain open for the team to manually merge. See forward-merger docs for more info.