Merge pull request #3496 from CliMA/ck/repro_fixes

Reproducibility infrastructure fixes

reproducibility_tests/ref_counter.jl

@@ -1,4 +1,4 @@
-191
+192
 
 # **README**
 #
@@ -21,6 +21,9 @@
 
 #=
 
+192
+- Reproducibility infrastructure fixes.
+
 191
 - Reproducibility infrastructure debugging.
 

reproducibility_tests/reproducibility_tools.jl

@@ -27,8 +27,15 @@ function error_if_dissimilar_dicts(dicts, dict)
     end
 end
 
-all_files_in_dir(dir) =
-    map((root, dirs, files) -> joinpath(root, file), walkdir(dir))
+function all_files_in_dir(dir)
+    all_files = String[]
+    for (root, dirs, files) in walkdir(dir)
+        for file in files
+            push!(all_files, joinpath(root, file))
+        end
+    end
+    return all_files
+end
 
 function no_comparison_error(dirs, non_existent_files)
     msg = "\n\n"
@@ -252,11 +259,12 @@ function export_reproducibility_results(
         joinpath(@__DIR__, "ref_counter.jl"),
     ),
     skip::Bool = !haskey(ENV, "BUILDKITE_COMMIT"),
+    repro_folder = "reproducibility_bundle",
 )
-    repro_folder = joinpath(computed_dir, "reproducibility_bundle")
-    data_file_computed = joinpath(repro_folder, computed_filename)
+    repro_dir = joinpath(computed_dir, repro_folder)
+    data_file_computed = joinpath(repro_dir, computed_filename)
 
-    mkpath(repro_folder)
+    mkpath(repro_dir)
     hdfwriter = InputOutput.HDF5Writer(data_file_computed, comms_ctx)
     InputOutput.write!(hdfwriter, field_vec, name)
     Base.close(hdfwriter)
@@ -277,7 +285,7 @@ function export_reproducibility_results(
     for (computed_mse, dir) in zip(computed_mses, dirs)
         commit_hash = basename(dirname(dir))
         computed_mse_file =
-            joinpath(repro_folder, "computed_mse_$commit_hash.json")
+            joinpath(repro_dir, "computed_mse_$commit_hash.json")
 
         open(computed_mse_file, "w") do io
             JSON.print(io, computed_mse)

reproducibility_tests/reproducibility_utils.jl

@@ -401,6 +401,7 @@ function move_data_to_save_dir(;
     ref_counter_PR = read_ref_counter(ref_counter_file_PR),
     skip = get(ENV, "BUILDKITE_PIPELINE_SLUG", nothing) != "climaatmos-ci",
     n_hash_characters = 7,
+    repro_folder = "reproducibility_bundle",
 )
     buildkite_ci || return nothing
 
@@ -419,15 +420,17 @@ function move_data_to_save_dir(;
         mkpath(dest_root)
         dest_dir = joinpath(dest_root, commit_sha)
         mkpath(dest_dir)
+        dest_repro = joinpath(dest_dir, repro_folder)
+        mkpath(dest_repro)
         # Always move reproducibility data, so that we
         # can compare against multiple references
         for src in dirs_src
-            dst = joinpath(dest_dir, basename(src))
+            dst = joinpath(dest_repro, basename(src))
             mv(src, dst; force = true)
             debug_reproducibility() &&
                 @info "Reproducibility: File $src moved to $dst"
         end
-        ref_counter_file_main = joinpath(dest_dir, "ref_counter.jl")
+        ref_counter_file_main = joinpath(dest_repro, "ref_counter.jl")
         mv(ref_counter_file_PR, ref_counter_file_main; force = true)
     else
         if debug_reproducibility()

test/unit_reproducibility_infra.jl

@@ -1199,6 +1199,8 @@ if pkgversion(ClimaCore) ≥ v"0.14.18"
             @test source_checksum(hash1) == source_checksum(computed_dir)
             @test source_checksum(hash2) == source_checksum(computed_dir)
 
+            repro_folder = "repro_bundle"
+            repro_dir = joinpath(save_dir, "hash_new", repro_folder)
             move_data_to_save_dir(;
                 dest_root = save_dir,
                 buildkite_ci = true,
@@ -1212,25 +1214,12 @@ if pkgversion(ClimaCore) ≥ v"0.14.18"
                     "ref_counter.jl",
                 ),
                 ref_counter_PR = 3,
+                repro_folder,
                 skip = false,
             )
-            @test isfile(
-                joinpath(
-                    save_dir,
-                    "hash_new",
-                    "job_id_1",
-                    "ref_prog_state.hdf5",
-                ),
-            )
-            @test isfile(
-                joinpath(
-                    save_dir,
-                    "hash_new",
-                    "job_id_2",
-                    "ref_prog_state.hdf5",
-                ),
-            )
-            @test isfile(joinpath(save_dir, "hash_new", "ref_counter.jl"))
+            @test isfile(joinpath(repro_dir, "job_id_1", "ref_prog_state.hdf5"))
+            @test isfile(joinpath(repro_dir, "job_id_2", "ref_prog_state.hdf5"))
+            @test isfile(joinpath(repro_dir, "ref_counter.jl"))
         end
     end
 
@@ -1283,6 +1272,8 @@ if pkgversion(ClimaCore) ≥ v"0.14.18"
             @test source_checksum(hash1) == source_checksum(computed_dir)
             @test source_checksum(hash2) == source_checksum(computed_dir)
 
+            repro_folder = "repro_bundle"
+            repro_dir = joinpath(save_dir, "hash_new", repro_folder)
             move_data_to_save_dir(;
                 dest_root = save_dir,
                 buildkite_ci = true,
@@ -1295,26 +1286,13 @@ if pkgversion(ClimaCore) ≥ v"0.14.18"
                     ref_counter_file_dir,
                     "ref_counter.jl",
                 ),
+                repro_folder,
                 ref_counter_PR = 3,
                 skip = false,
             )
-            @test isfile(
-                joinpath(
-                    save_dir,
-                    "hash_new",
-                    "job_id_1",
-                    "ref_prog_state.hdf5",
-                ),
-            )
-            @test isfile(
-                joinpath(
-                    save_dir,
-                    "hash_new",
-                    "job_id_2",
-                    "ref_prog_state.hdf5",
-                ),
-            )
-            @test isfile(joinpath(save_dir, "hash_new", "ref_counter.jl"))
+            @test isfile(joinpath(repro_dir, "job_id_1", "ref_prog_state.hdf5"))
+            @test isfile(joinpath(repro_dir, "job_id_2", "ref_prog_state.hdf5"))
+            @test isfile(joinpath(repro_dir, "ref_counter.jl"))
         end
     end
 end