Run per-host/per-package nix eval and dry-run build concurrently

codex-maintenance.sh evaluated each affected host/package one at a time, even though those calls are independent. Added scripts/lib/nix-parallel.sh (run_nix_parallel) and wired it into the host-eval, package-eval, and dry-run-build loops. Concurrency defaults to core count capped by available memory (~1GB/job) rather than plain nproc: empirically, nproc-many concurrent full-flake evals OOM-killed each other on a 4GB/6-core box, while 3-4 ran clean and were still ~2x faster than serial. Override via NIX_PARALLEL_JOBS. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-22 02:04:21 +00:00
parent a2ce01d6ce
commit f237a6a3d2
2 changed files with 129 additions and 26 deletions
@@ -0,0 +1,95 @@
+#!/usr/bin/env bash
+# Shared parallel-nix-invocation helper for scripts/codex-maintenance.sh.
+# Source alongside nix-eval.sh:
+#   source "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/lib/nix-parallel.sh"
+#
+# The per-host/per-package `nix eval`/`nix build --dry-run` calls in
+# codex-maintenance.sh are independent of each other, so running them one at
+# a time leaves most cores idle for most of the sweep -- run_nix_parallel
+# fans a batch of them out across up to NIX_PARALLEL_JOBS processes instead.
+
+# NIX_PARALLEL_JOBS: how many `nix` invocations run_nix_parallel runs at
+# once. Defaults to core count capped by available memory (~1GB/job,
+# floor 1) rather than plain `nproc` -- each concurrent `nix eval` here
+# evaluates a whole NixOS system closure from scratch, and on a small/
+# memory-constrained CI runner, `nproc` concurrent evals can OOM-kill each
+# other (confirmed empirically: on a 4GB/6-core box, 5-6 concurrent evals
+# started getting killed while 3-4 ran clean and were still ~2x faster than
+# serial). Override via env if a given machine/CI runner has room to spare
+# or needs a tighter cap.
+default_nix_parallel_jobs() {
+  local cores mem_avail_kb mem_cap
+  cores="$(nproc 2>/dev/null || echo 4)"
+  mem_avail_kb="$(awk '/^MemAvailable:/ {print $2}' /proc/meminfo 2>/dev/null)"
+  if [[ -z "$mem_avail_kb" ]]; then
+    echo "$cores"
+    return
+  fi
+  mem_cap=$((mem_avail_kb / 1024 / 1024))
+  ((mem_cap < 1)) && mem_cap=1
+  ((mem_cap < cores)) && echo "$mem_cap" || echo "$cores"
+}
+NIX_PARALLEL_JOBS="${NIX_PARALLEL_JOBS:-$(default_nix_parallel_jobs)}"
+
+# Separator between a job's label and its flake attr in the arrays
+# run_nix_parallel takes -- a control character so it can't collide with
+# anything a label or attr path would plausibly contain.
+NIX_PARALLEL_SEP=$'\x1f'
+
+# run_nix_parallel <jobs_array_name> <nix subcommand + flags...>
+#
+# jobs_array_name: name of an already-populated bash array whose entries are
+# "<label>${NIX_PARALLEL_SEP}<attr>" pairs, e.g.
+#   jobs=("proxmox-docker${NIX_PARALLEL_SEP}.#nixosConfigurations.proxmox-docker...drvPath")
+# Remaining args are passed to `nix` before the attr, e.g.:
+#   run_nix_parallel jobs eval --raw "${NIX_EVAL_FLAGS[@]}"
+#   run_nix_parallel jobs build --dry-run --no-link "${NIX_EVAL_FLAGS[@]}"
+#
+# Prints "==> <label>" followed by that job's stdout+stderr for every job,
+# in submission order (not completion order) so a run stays readable and
+# diffable across invocations even though the work itself doesn't finish in
+# that order. Returns non-zero if any job failed, only after every job has
+# finished and been printed -- same "surface everything, then fail" contract
+# a `set -e` caller gets, just parallelized instead of stopping at the first
+# failure.
+run_nix_parallel() {
+  local -n jobs_ref="$1"
+  shift
+  local -a nix_args=("$@")
+
+  local n=${#jobs_ref[@]}
+  [[ $n -eq 0 ]] && return 0
+
+  local tmp_dir
+  tmp_dir="$(mktemp -d)"
+
+  local i=0 running=0
+  for job in "${jobs_ref[@]}"; do
+    local attr="${job#*"${NIX_PARALLEL_SEP}"}"
+    printf '%s\n' "${job%%"${NIX_PARALLEL_SEP}"*}" >"${tmp_dir}/${i}.label"
+    (
+      if nix "${nix_args[@]}" "$attr" >"${tmp_dir}/${i}.out" 2>&1; then
+        echo 0 >"${tmp_dir}/${i}.status"
+      else
+        echo 1 >"${tmp_dir}/${i}.status"
+      fi
+    ) &
+    i=$((i + 1))
+    running=$((running + 1))
+    if ((running >= NIX_PARALLEL_JOBS)); then
+      wait -n
+      running=$((running - 1))
+    fi
+  done
+  wait
+
+  local failed=0 j
+  for ((j = 0; j < n; j++)); do
+    echo "==> $(cat "${tmp_dir}/${j}.label")"
+    cat "${tmp_dir}/${j}.out"
+    [[ "$(cat "${tmp_dir}/${j}.status")" -ne 0 ]] && failed=1
+  done
+
+  rm -rf "$tmp_dir"
+  return $failed
+}