CI (Buildkite): Add the package_linux64 and tester_linux64 pipelines

.buildkite/linux64.yml

@@ -0,0 +1,85 @@
+# 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/v1/llvm-passes.tar.gz
+ rootfs_treehash: "f3ed53f159e8f13edfba8b20ebdb8ece73c1b8a8"
+ 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 -j 6 release
+ make install
+
+ echo "--- Print Julia version info"
+ ./julia -e 'using InteractiveUtils; InteractiveUtils.versioninfo()'
+ ./julia -e '@info "" Sys.CPU_THREADS'
+
+ echo "--- Upload build artifacts"
+ julia -e 'println(filter(startswith("julia-"), readdir(pwd())))'
+ julia -e 'println(only(filter(startswith("julia-"), readdir(pwd()))))' # make sure there is only a single directory of the form `julia-*``
+ ls -d julia-*
+ mv julia-* julia-release
+ ls -d julia-*
+ ls -d julia-release
+ rm -rf julia-release.tar.gz
+ tar cfz julia-release.tar.gz julia-release/
+ ls julia-release.tar.gz
+ buildkite-agent artifact upload julia-release.tar.gz
+
+ timeout_in_minutes: 60
+ notify:
+ - github_commit_status:
+ context: "package_linux64"
+
+ - label: "tester_linux64"
+ key: tester_linux64
+ # depends_on: package_linux64 # TODO: uncomment this line
+ plugins:
+ - 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"
+ 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()'
+ ./julia -e '@info "" Sys.CPU_THREADS'
+
+ echo "--- Run the Julia test suite WITHOUT rr" # TODO: delete this line
+ ./julia -e 'Base.runtests(["abstractarray"]; ncores = Sys.CPU_THREADS)' # TODO: delete this line
+
+ echo "--- Run the Julia test suite"
+ # TODO: change "abstractarray" to "all"
+ # TODO: provide the correct value for MY_RUNID
+ # TODO: provide the correct value for MY_SHORTCOMMIT
+ ./julia .buildkite/rr_capture.jl MY_RUNID MY_SHORTCOMMIT ./julia -e 'Base.runtests(["abstractarray"]; ncores = Sys.CPU_THREADS)'
+
+ timeout_in_minutes: 120
+ notify:
+ - github_commit_status:
+ context: "tester_linux64"

.buildkite/pipeline.yml

@@ -17,5 +17,6 @@ steps:
  buildkite-agent pipeline upload .buildkite/whitespace.yml
  buildkite-agent pipeline upload .buildkite/llvm_passes.yml
  buildkite-agent pipeline upload .buildkite/embedding.yml
+ buildkite-agent pipeline upload .buildkite/linux64.yml
  agents:
  queue: julia

.buildkite/rr_capture.jl

@@ -0,0 +1,125 @@
+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) < 3
+ throw(ErrorException("Usage: rr_capture.jl [buildnumber] [shortcommit] [command...]"))
+end
+
+const TIMEOUT = 2 * 60 * 60 # timeout in seconds
+
+const run_id = popfirst!(ARGS)
+const shortcommit = popfirst!(ARGS)
+
+@info "" run_id shortcommit
+
+# We only use RR on the `tester_linux64` builder.
+@info "" haskey(ENV, "BUILDKITE_LABEL")
+@info "" ENV["BUILDKITE_LABEL"]
+if ENV["BUILDKITE_LABEL"] != "tester_linux64"
+ run(`$ARGS`)
+end
+
+num_cores = min(Sys.CPU_THREADS, 8, parse(Int, get(ENV, "JULIA_TEST_NUM_CORES", "8")) + 1)
+
+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