From f33efe06d26f144c05189dfdc4634586951cad34 Mon Sep 17 00:00:00 2001 From: Dilum Aluthge Date: Sat, 24 Jul 2021 22:58:54 -0400 Subject: [PATCH] CI (Buildkite): Add the `package_linux64` and `doctest` builders (#41541) Co-authored-by: Elliot Saba (cherry picked from commit c7097c5793545658aef20d6bb83eb1af7bc81310) --- .buildkite/misc/doctest.yml | 39 +++++ .buildkite/{ => misc}/embedding.yml | 4 +- .../{llvm_passes.yml => misc/llvmpasses.yml} | 8 +- .buildkite/{ => misc}/whitespace.yml | 4 +- .buildkite/pipeline.yml | 19 ++- .buildkite/platforms/linux64.yml | 80 +++++++++++ .buildkite/rr_capture.jl | 134 ++++++++++++++++++ 7 files changed, 273 insertions(+), 15 deletions(-) create mode 100644 .buildkite/misc/doctest.yml rename .buildkite/{ => misc}/embedding.yml (91%) rename .buildkite/{llvm_passes.yml => misc/llvmpasses.yml} (88%) rename .buildkite/{ => misc}/whitespace.yml (87%) create mode 100644 .buildkite/platforms/linux64.yml create mode 100644 .buildkite/rr_capture.jl diff --git a/.buildkite/misc/doctest.yml b/.buildkite/misc/doctest.yml new file mode 100644 index 00000000000000..234e49f5dcf536 --- /dev/null +++ b/.buildkite/misc/doctest.yml @@ -0,0 +1,39 @@ +# These steps should only run on `sandbox.jl` machines, not `docker`-isolated ones +# since we need nestable sandboxing. The rootfs images being used here are built from +# the `.buildkite/rootfs_images/llvm-passes.jl` file. +agents: + queue: "julia" + # Only run on `sandbox.jl` machines (not `docker`-isolated ones) since we need nestable sandboxing + sandbox.jl: "true" + os: "linux" + +steps: + - label: "doctest" + key: doctest + plugins: + - JuliaCI/julia#v1: + version: 1.6 + - staticfloat/sandbox#v1: + rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v3.0/package_linux.x86_64.tar.gz + rootfs_treehash: "d5722d586b93eb307bb6340d275afdbf7578a756" + uid: 1000 + gid: 1000 + workspaces: + # Include `/cache/repos` so that our `git` version introspection works. + - "/cache/repos:/cache/repos" + commands: | + echo "--- Build Julia from source" + make -j 6 + + echo "--- Print Julia version info" + ./julia -e 'using InteractiveUtils; InteractiveUtils.versioninfo()' + + echo "--- Build Julia docs" + make docs + + echo "--- Run Julia doctests" + JULIA_NUM_THREADS=1 make -C doc doctest=true + timeout_in_minutes: 120 + notify: + - github_commit_status: + context: "doctest" diff --git a/.buildkite/embedding.yml b/.buildkite/misc/embedding.yml similarity index 91% rename from .buildkite/embedding.yml rename to .buildkite/misc/embedding.yml index 5cf6b985573f59..31f85633479a41 100644 --- a/.buildkite/embedding.yml +++ b/.buildkite/misc/embedding.yml @@ -14,8 +14,8 @@ steps: - JuliaCI/julia#v1: version: 1.6 - staticfloat/sandbox#v1: - rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v1/llvm-passes.tar.gz - rootfs_treehash: "f3ed53f159e8f13edfba8b20ebdb8ece73c1b8a8" + rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v3.0/package_linux.x86_64.tar.gz + rootfs_treehash: "d5722d586b93eb307bb6340d275afdbf7578a756" uid: 1000 gid: 1000 workspaces: diff --git a/.buildkite/llvm_passes.yml b/.buildkite/misc/llvmpasses.yml similarity index 88% rename from .buildkite/llvm_passes.yml rename to .buildkite/misc/llvmpasses.yml index 7453af82c1e734..7660b3978e9674 100644 --- a/.buildkite/llvm_passes.yml +++ b/.buildkite/misc/llvmpasses.yml @@ -14,8 +14,8 @@ steps: - JuliaCI/julia#v1: version: 1.6 - staticfloat/sandbox#v1: - rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v1/llvm-passes.tar.gz - rootfs_treehash: "f3ed53f159e8f13edfba8b20ebdb8ece73c1b8a8" + rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v3.1/llvm_passes.x86_64.tar.gz + rootfs_treehash: "9dd715500b117a16fcfa419ea0bca0c0ca902cee" workspaces: # Include `/cache/repos` so that our `git` version introspection works. - "/cache/repos:/cache/repos" @@ -36,8 +36,8 @@ steps: - JuliaCI/julia#v1: version: 1.6 - staticfloat/sandbox#v1: - rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v1/llvm-passes.tar.gz - rootfs_treehash: "f3ed53f159e8f13edfba8b20ebdb8ece73c1b8a8" + rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v3.0/package_linux.x86_64.tar.gz + rootfs_treehash: "d5722d586b93eb307bb6340d275afdbf7578a756" uid: 1000 gid: 1000 workspaces: diff --git a/.buildkite/whitespace.yml b/.buildkite/misc/whitespace.yml similarity index 87% rename from .buildkite/whitespace.yml rename to .buildkite/misc/whitespace.yml index bd625e855c5b5b..01b4c79e7822bd 100644 --- a/.buildkite/whitespace.yml +++ b/.buildkite/misc/whitespace.yml @@ -14,8 +14,8 @@ steps: - JuliaCI/julia#v1: version: 1.6 - staticfloat/sandbox#v1: - rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v1/llvm-passes.tar.gz - rootfs_treehash: "f3ed53f159e8f13edfba8b20ebdb8ece73c1b8a8" + rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v3.0/package_linux.x86_64.tar.gz + rootfs_treehash: "d5722d586b93eb307bb6340d275afdbf7578a756" workspaces: - "/cache/repos:/cache/repos" commands: | diff --git a/.buildkite/pipeline.yml b/.buildkite/pipeline.yml index 097b8976eea702..3a3ebc718cbaa9 100644 --- a/.buildkite/pipeline.yml +++ b/.buildkite/pipeline.yml @@ -4,9 +4,9 @@ # # Yes, this is creating another layer of indirection; the flow now looks like: # -# [webui] -> pipeline.yml -> llvm_passes.yml +# [webui] -> pipeline.yml -> misc/whitespace.yml # -# when we could theoretically just have the `webui` launch `llvm_passes.yml`, +# when we could theoretically just have the `webui` launch `misc/whitespace.yml`, # however this raises the bar for contributors to add new (unsigned) steps to # our CI configuration, so I'd rather live with an extra layer of indirection # and only need to touch the webui configuration when we need to alter @@ -14,10 +14,15 @@ steps: - label: ":buildkite: Launch unsigned pipelines" commands: | - # We launch whitespace first, because we want that pipeline to finish as quickly as possible. - # The remaining unsigned pipelines are launched in alphabetical order. - buildkite-agent pipeline upload .buildkite/whitespace.yml - buildkite-agent pipeline upload .buildkite/embedding.yml - buildkite-agent pipeline upload .buildkite/llvm_passes.yml + # First, we launch whitespace, because we want that pipeline to finish as quickly as possible. + buildkite-agent pipeline upload .buildkite/misc/whitespace.yml + + # Next, we launch the miscellaneous pipelines in alphabetical order. + buildkite-agent pipeline upload .buildkite/misc/doctest.yml + buildkite-agent pipeline upload .buildkite/misc/embedding.yml + buildkite-agent pipeline upload .buildkite/misc/llvmpasses.yml + + # Finally, we launch the platform pipelines in alphabetical order. + buildkite-agent pipeline upload .buildkite/platforms/linux64.yml agents: queue: julia diff --git a/.buildkite/platforms/linux64.yml b/.buildkite/platforms/linux64.yml new file mode 100644 index 00000000000000..69befe1064df70 --- /dev/null +++ b/.buildkite/platforms/linux64.yml @@ -0,0 +1,80 @@ +# These steps should only run on `sandbox.jl` machines, not `docker`-isolated ones +# since we need nestable sandboxing. The rootfs images being used here are built from +# the `.buildkite/rootfs_images/llvm-passes.jl` file. +agents: + queue: "julia" + # Only run on `sandbox.jl` machines (not `docker`-isolated ones) since we need nestable sandboxing + sandbox.jl: "true" + os: "linux" + +steps: + - label: "package_linux64" + key: package_linux64 + plugins: + - JuliaCI/julia#v1: + version: 1.6 + - staticfloat/sandbox#v1: + rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v3.0/package_linux.x86_64.tar.gz + rootfs_treehash: "d5722d586b93eb307bb6340d275afdbf7578a756" + uid: 1000 + gid: 1000 + workspaces: + # Include `/cache/repos` so that our `git` version introspection works. + - "/cache/repos:/cache/repos" + commands: | + echo "--- Build Julia from source" + make -j 6 + make release + make install + + echo "--- Print Julia version info" + ./julia -e 'using InteractiveUtils; InteractiveUtils.versioninfo()' + + echo "--- Compress build artifacts" + mv julia-* julia-artifact + rm -rf julia-linux64.tar.gz + tar czf julia-linux64.tar.gz julia-artifact/ + + echo "--- Upload build artifacts" + buildkite-agent artifact upload julia-linux64.tar.gz + timeout_in_minutes: 60 + notify: + - github_commit_status: + context: "package_linux64" + + # - label: "tester_linux64" + # key: tester_linux64 + # depends_on: package_linux64 + # plugins: + # - JuliaCI/julia#v1: + # version: 1.6 + # - staticfloat/sandbox#v1: + # # TODO: use a separate `tester_linux` image, instead of using the `package_linux` image. + # rootfs_url: https://github.com/JuliaCI/rootfs-images/releases/download/v3.0/package_linux.x86_64.tar.gz + # rootfs_treehash: "d5722d586b93eb307bb6340d275afdbf7578a756" + # uid: 1000 + # gid: 1000 + # workspaces: + # # Include `/cache/repos` so that our `git` version introspection works. + # - "/cache/repos:/cache/repos" + # env: + # JULIA_SHELL: "/bin/bash" + # commands: | + # echo "--- Download build artifacts" + # rm -rf julia-linux64.tar.gz + # buildkite-agent artifact download julia-linux64.tar.gz . + # + # echo "--- Extract build artifacts" + # rm -rf julia-artifact/ + # tar xzf julia-linux64.tar.gz julia-artifact/ + # + # echo "--- Print Julia version info" + # julia-artifact/bin/julia -e 'using InteractiveUtils; InteractiveUtils.versioninfo()' + # + # echo "--- Run the Julia test suite" + # unset JULIA_DEPOT_PATH + # julia-artifact/bin/julia .buildkite/rr_capture.jl julia-artifact/bin/julia -e 'Base.runtests(["all"]; ncores = Sys.CPU_THREADS)' + # timeout_in_minutes: 120 + # notify: + # - github_commit_status: + # context: "tester_linux64" diff --git a/.buildkite/rr_capture.jl b/.buildkite/rr_capture.jl new file mode 100644 index 00000000000000..07d57f31ff29cf --- /dev/null +++ b/.buildkite/rr_capture.jl @@ -0,0 +1,134 @@ +using Dates +using Pkg +using Tar + +if Base.VERSION < v"1.6" + throw(ErrorException("The `rr_capture.jl` script requires Julia 1.6 or greater")) +end + +if length(ARGS) < 1 + throw(ErrorException("Usage: rr_capture.jl [command...]")) +end + +const TIMEOUT = 2 * 60 * 60 # timeout in seconds + +# We only use `rr` on the `tester_linux64` builder +const use_rr_if_builder_is = "tester_linux64" + +const run_id = get(ENV, "BUILDKITE_JOB_ID", "unknown") +const shortcommit = get(ENV, "BUILDKITE_COMMIT", "unknown") +const builder = get(ENV, "BUILDKITE_STEP_KEY", use_rr_if_builder_is) +const use_rr = builder == use_rr_if_builder_is + +@info "" run_id shortcommit builder use_rr +@info "" ARGS + +# if !use_rr # TODO: uncomment this line +if true # TODO: delete this line + @info "We will not run the tests under rr" + p = run(`$ARGS`) + exit(p.exitcode) +end + +@info "We will run the tests under rr" + +const num_cores = min(Sys.CPU_THREADS, 8, parse(Int, get(ENV, "JULIA_TEST_NUM_CORES", "8")) + 1) +@info "" num_cores + +proc = nothing + +new_env = copy(ENV) +mktempdir() do dir + Pkg.activate(dir) + Pkg.add("rr_jll") + Pkg.add("Zstd_jll") + + rr_jll = Base.require(Base.PkgId(Base.UUID((0xe86bdf43_55f7_5ea2_9fd0_e7daa2c0f2b4)), "rr_jll")) + zstd_jll = Base.require(Base.PkgId(Base.UUID((0x3161d3a3_bdf6_5164_811a_617609db77b4)), "Zstd_jll")) + rr(func) = Base.invokelatest(rr_jll.rr, func; adjust_LIBPATH=false) + rr() do rr_path + capture_script_path = joinpath(dir, "capture_output.sh") + loader = Sys.WORD_SIZE == 64 ? "/lib64/ld-linux-x86-64.so.2" : "/lib/ld-linux.so.2" + open(capture_script_path, "w") do io + write(io, """ + #!/bin/bash + + $(rr_path) record --nested=detach "\$@" > >(tee -a $(dir)/stdout.log) 2> >(tee -a $(dir)/stderr.log >&2) + """) + end + chmod(capture_script_path, 0o755) + + new_env = copy(ENV) + new_env["_RR_TRACE_DIR"] = joinpath(dir, "rr_traces") + new_env["RR_LOG"]="all:debug" + new_env["RR_LOG_BUFFER"]="100000" + new_env["JULIA_RR"] = capture_script_path + t_start = time() + global proc = run(setenv(`$(rr_path) record --num-cores=$(num_cores) $ARGS`, new_env), (stdin, stdout, stderr); wait=false) + + # Start asynchronous timer that will kill `rr` + @async begin + sleep(TIMEOUT) + + # If we've exceeded the timeout and `rr` is still running, kill it. + if isopen(proc) + println(stderr, "\n\nProcess timed out. Signalling `rr` for force-cleanup!") + kill(proc, Base.SIGTERM) + + # Give `rr` a chance to cleanup + sleep(60) + + if isopen(proc) + println(stderr, "\n\n`rr` failed to cleanup within one minute, killing and exiting immediately!") + kill(proc, Base.SIGKILL) + exit(1) + end + end + end + + # Wait for `rr` to finish, either through naturally finishing its run, or `SIGTERM`. + # If we have to `SIGKILL` + wait(proc) + + # On a non-zero exit code, upload the `rr` trace + if !success(proc) + println(stderr, "`rr` returned $(proc.exitcode), packing and uploading traces...") + + if !isdir(joinpath(dir, "rr_traces")) + println(stderr, "No `rr_traces` directory! Did `rr` itself fail?") + exit(1) + end + + # Clean up non-traces + rm(joinpath(dir, "rr_traces", "latest-trace")) + rm(joinpath(dir, "rr_traces", "cpu_lock")) + + # Create a directory for the pack files to go + pack_dir = joinpath(dir, "pack") + mkdir(pack_dir) + + # Pack all traces + trace_dirs = [joinpath(dir, "rr_traces", f) for f in readdir(joinpath(dir, "rr_traces"))] + filter!(isdir, trace_dirs) + run(ignorestatus(`$(rr_path) pack --pack-dir=$pack_dir $(trace_dirs)`)) + + # Tar it up + mkpath("dumps") + datestr = Dates.format(now(), dateformat"yyyy-mm-dd_HH_MM_SS") + dst_path = "dumps/rr-run_$(run_id)-gitsha_$(shortcommit)-$(datestr).tar.zst" + zstd_jll.zstdmt() do zstdp + tarproc = open(`$zstdp -o $dst_path`, "w") + Tar.create(dir, tarproc) + close(tarproc.in) + end + end + end +end + +# Pass the exit code back up to Buildkite +if proc.termsignal != 0 + ccall(:raise, Cvoid, (Cint,), proc.termsignal) + exit(1) # Just in case the signal did not cause an exit +else + exit(proc.exitcode) +end