[WIP] Parallel GH actions workflow for Nixpkgs eval

Partly taken from NixOS#352808 and NixOS#269403

.github/workflows/eval.yml

@@ -0,0 +1,53 @@
+name: Eval
+
+on: pull_request_target
+
+permissions:
+  contents: read
+
+jobs:
+  tests:
+    name: eval-check
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        system: [x86_64-linux, aarch64-linux, aarch64-darwin, x86_64-darwin]
+    steps:
+      # Important: Because of `pull_request_target`, this doesn't check out the PR,
+      # but rather the base branch of the PR, which is needed so we don't run untrusted code
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          path: base
+          sparse-checkout: ci
+      - name: Resolving the merge commit
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          if mergedSha=$(base/ci/get-merge-commit.sh ${{ github.repository }} ${{ github.event.number }}); then
+            echo "Checking the merge commit $mergedSha"
+            echo "mergedSha=$mergedSha" >> "$GITHUB_ENV"
+          else
+            # Skipping so that no notifications are sent
+            echo "Skipping the rest..."
+          fi
+          rm -rf base
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        # Add this to _all_ subsequent steps to skip them
+        if: env.mergedSha
+        with:
+          ref: ${{ env.mergedSha }}
+
+      - uses: cachix/install-nix-action@08dcb3a5e62fa31e2da3d490afc4176ef55ecd72 # v30
+        if: env.mergedSha
+
+      - name: Enable swap
+        if: env.mergedSha
+        run: |
+          sudo fallocate -l 10G /swapfile
+          sudo chmod 600 /swapfile
+          sudo mkswap /swapfile
+          sudo swapon /swapfile
+
+      - name: Check eval
+        if: env.mergedSha
+        run: ./ci/eval-nixpkgs.sh --system "${{ matrix.system }}"

ci/eval-nixpkgs.sh

@@ -0,0 +1,61 @@
+#!/usr/bin/env nix-shell
+#!nix-shell -i bash -p moreutils -I nixpkgs=channel:nixpkgs-unstable
+
+set -euxo pipefail
+
+system="x86_64-linux"
+SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
+NIXPKGS_PATH="$(readlink -f "$SCRIPT_DIR"/..)"
+
+parseArgs() {
+  while [[ $# -gt 0 ]]; do
+    case $1 in
+      --system)
+        system=$2
+        shift 2
+        ;;
+      *)
+        echo "Unknown argument: $1"
+        exit 1
+        ;;
+    esac
+  done
+}
+
+main() {
+    parseArgs "$@"
+    tmpdir=$(mktemp -d)
+    trap 'rm -rf "$tmpdir"' EXIT
+
+    nix-instantiate --eval --strict --json --arg enableWarnings false "$NIXPKGS_PATH"/pkgs/top-level/release-attrpaths-superset.nix -A paths > "$tmpdir/paths.json"
+
+    CORES=$(nproc)
+    # Originally @amjoseph: note that the number of processes spawned is four times
+    # the number of cores -- this helps in two ways:
+    # 1. Keeping cores busy while I/O operations are in flight
+    # 2. Since the amount of time needed for the jobs is *not* balanced
+    # this minimizes the "tail latency" for the very last job to finish
+    # (on one core) by making the job size smaller.
+    NUM_CHUNKS=$(( 4 * CORES ))
+
+
+    (
+      set +e
+      parallel -j "$CORES" \
+          nix-env -qaP --no-name --out-path --arg checkMeta true --arg includeBroken true \
+          --arg systems "[\"$system\"]" \
+          -f "$NIXPKGS_PATH"/ci/parallel.nix --arg attrPathFile "$tmpdir"/paths.json \
+          --arg numChunks "$NUM_CHUNKS" --show-trace --arg myChunk \
+          -- $(seq 0 $(( NUM_CHUNKS - 1 ))) > "$tmpdir/paths"
+      echo $? > "$tmpdir/exit-code"
+    ) &
+    pid=$!
+    while kill -0 "$pid"; do
+        free -g >&2
+        sleep 20
+    done
+    jq --raw-input --slurp 'split("\n") | map(select(. != "") | split(" ") | map(select(. != "")) | { key: .[0], value: .[1] }) | from_entries' "$tmpdir/paths"
+    exit "$(cat "$tmpdir/exit-code")"
+}
+
+main "$@"

ci/parallel.nix

@@ -0,0 +1,61 @@
+/*
+Invocation:
+
+
+Invocation; note that the number of processes spawned is four times
+the number of cores -- this helps in two ways:
+
+1. Keeping cores busy while I/O operations are in flight
+
+2. Since the amount of time needed for the jobs is *not* balanced
+   this minimizes the "tail latency" for the very last job to finish
+   (on one core) by making the job size smaller.
+
+*/
+# see pkgs/top-level/nohydra
+{ lib ? import ../lib
+, checkMeta
+, includeBroken ? true
+, path ? ./..
+, systems
+, myChunk
+, numChunks
+, attrPathFile
+}:
+
+let
+  attrPaths = builtins.fromJSON (builtins.readFile attrPathFile);
+  chunkSize = (lib.length attrPaths) / numChunks;
+  myPaths =
+    let
+      dropped = lib.drop (chunkSize*myChunk) attrPaths;
+    in
+      if myChunk == numChunks - 1
+      then dropped
+      else lib.take chunkSize dropped;
+
+  unfiltered = import ../pkgs/top-level/release-outpaths.nix {
+    inherit checkMeta path includeBroken systems;
+  };
+
+  f = i: m: a:
+    lib.mapAttrs (name: values:
+      if a ? ${name} then
+        if lib.any (value: lib.length value <= i + 1) values then
+          a.${name}
+        else
+          f (i + 1) values a.${name}
+      else
+        null
+    ) (lib.groupBy (a: lib.elemAt a i) m);
+
+  filtered = f 0 myPaths unfiltered;
+
+  recurseEverywhere = val:
+    if lib.isDerivation val || !(lib.isAttrs val)
+    then val
+    else (builtins.mapAttrs (_: v: recurseEverywhere v) val)
+         // { recurseForDerivations = true; };
+
+in
+  recurseEverywhere filtered