Add RocksDB native compaction policy task

2.0/README.md

@@ -47,6 +47,14 @@ applies the submitted patch to a clean skeleton, runs a hidden arena against
 multiple baseline bot families, and scores by mean baseline win rate with a
 small faster-win tiebreak. The online generals.io service is not used.
 
+## RocksDB Native Compaction Policy
+
+This systems problem asks agents to patch the leveled compaction picker in a
+pinned RocksDB checkout. Its problem ID is `rocksdb_native_compaction_policy`.
+The judge runs native RocksDB workloads, checks snapshots and full database
+contents, and scores paired improvements in write/read/space amplification and
+compaction debt against unmodified RocksDB.
+
 ## vLLM LLM-Serving Optimization
 
 This systems problem asks agents to patch a clean upstream vLLM checkout to

2.0/problems/rocksdb_native_compaction_policy/config.yaml

@@ -0,0 +1,49 @@
+tag: systems
+runtime:
+  language: cpp
+  timeout_seconds: 10800
+  environment: "Patch a pinned RocksDB v10.10.1 checkout; native correctness and compaction-cost judge"
+  apt_packages:
+    - bash
+    - build-essential
+    - ca-certificates
+    - git
+    - libbz2-dev
+    - libgflags-dev
+    - liblz4-dev
+    - libsnappy-dev
+    - libzstd-dev
+    - zlib1g-dev
+  docker:
+    image: python:3.12-slim-bookworm
+    judge_image: frontiercs/rocksdb-native-compaction-judge:experimental-v10.10.1-task2
+  visible_inputs:
+    - source: /opt/rocksdb-clean
+      destination: /app/rocksdb
+environment:
+  cpus: 8
+  memory_mb: 16384
+  storage_mb: 32768
+  build_timeout_seconds: 7200
+evaluation:
+  schema_version: rocksdb-native-compaction-v2
+  public_suite_id: rocksdb-native-public-v2
+  final_suite_id: rocksdb-native-final-v2
+  rocksdb_commit: "4595a5e95ae8525c42e172a054435782b3479c57"
+  feedback_cases:
+    - {seed: 1101, profile: smoke}
+    - {seed: 1202, profile: l0_pressure}
+    - {seed: 1303, profile: range_snapshot}
+    - {seed: 1404, profile: scanmix}
+    - {seed: 1505, profile: multi_cf}
+    - {seed: 1606, profile: time_series}
+    - {seed: 1707, profile: difficulty}
+    - {seed: 1808, profile: overlap_rewrite}
+  build_timeout_seconds: 7200
+  run_timeout_seconds: 1800
+  build_jobs: 3
+submission:
+  kind: file
+  path: /app/solution.patch
+  allow_empty: true
+  max_queue_size: 2

2.0/problems/rocksdb_native_compaction_policy/docker/build_images.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROCKSDB_COMMIT="${ROCKSDB_COMMIT:-4595a5e95ae8525c42e172a054435782b3479c57}"
+ROCKSDB_ARCHIVE_SHA256="${ROCKSDB_ARCHIVE_SHA256:-9469e7d24644e298134d0464ec2d398a9b91cc401102009ec1ecde0383485d6c}"
+ROCKSDB_BUILD_JOBS="${ROCKSDB_BUILD_JOBS:-3}"
+JUDGE_TAG="${JUDGE_TAG:-frontiercs/rocksdb-native-compaction-judge:experimental-v10.10.1-task2}"
+PUSH_IMAGE="${PUSH_IMAGE:-0}"
+PLATFORMS="${PLATFORMS:-linux/amd64,linux/arm64}"
+PROBLEM_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)
+
+BUILD_ARGS=(
+  --build-arg "ROCKSDB_COMMIT=${ROCKSDB_COMMIT}" \
+  --build-arg "ROCKSDB_ARCHIVE_SHA256=${ROCKSDB_ARCHIVE_SHA256}" \
+  --build-arg "ROCKSDB_BUILD_JOBS=${ROCKSDB_BUILD_JOBS}" \
+  -f "${PROBLEM_DIR}/docker/judge/Dockerfile" \
+  -t "${JUDGE_TAG}" \
+  "${PROBLEM_DIR}"
+)
+
+if [[ "${PUSH_IMAGE}" == 1 ]]; then
+  docker buildx build --platform "${PLATFORMS}" --push "${BUILD_ARGS[@]}"
+  echo "Published ${JUDGE_TAG} for ${PLATFORMS}"
+else
+  DOCKER_BUILDKIT=1 docker build "${BUILD_ARGS[@]}"
+  echo "Built ${JUDGE_TAG} for the local platform"
+fi

2.0/problems/rocksdb_native_compaction_policy/docker/judge/Dockerfile

@@ -0,0 +1,82 @@
+# syntax=docker/dockerfile:1.7
+FROM python:3.12-slim-bookworm
+
+ARG ROCKSDB_COMMIT=4595a5e95ae8525c42e172a054435782b3479c57
+ARG ROCKSDB_ARCHIVE_SHA256=9469e7d24644e298134d0464ec2d398a9b91cc401102009ec1ecde0383485d6c
+ARG ROCKSDB_BUILD_JOBS=3
+ARG DEBIAN_FRONTEND=noninteractive
+
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        bash \
+        build-essential \
+        ca-certificates \
+        cmake \
+        git \
+        ninja-build \
+        pkg-config \
+        libgflags-dev \
+        libsnappy-dev \
+        zlib1g-dev \
+        libbz2-dev \
+        liblz4-dev \
+        libzstd-dev \
+        python3 && \
+    rm -rf /var/lib/apt/lists/*
+
+# Prebuild one tree for incremental candidate builds, then copy it for vanilla.
+RUN python3 - "${ROCKSDB_COMMIT}" "${ROCKSDB_ARCHIVE_SHA256}" <<'PY'
+import hashlib
+import io
+import shutil
+import sys
+import tarfile
+import time
+import urllib.request
+from pathlib import Path
+
+commit, expected_sha256 = sys.argv[1:]
+url = f"https://codeload.github.com/facebook/rocksdb/tar.gz/{commit}"
+for attempt in range(5):
+    try:
+        request = urllib.request.Request(url, headers={"User-Agent": "Frontier-CS"})
+        with urllib.request.urlopen(request, timeout=30) as response:
+            archive = response.read()
+        actual_sha256 = hashlib.sha256(archive).hexdigest()
+        if actual_sha256 != expected_sha256:
+            raise ValueError(f"archive sha256 mismatch: {actual_sha256}")
+        break
+    except Exception:
+        if attempt == 4:
+            raise
+        time.sleep(3 * (attempt + 1))
+
+extract_dir = Path("/tmp/rocksdb-extract")
+shutil.rmtree(extract_dir, ignore_errors=True)
+extract_dir.mkdir(parents=True)
+with tarfile.open(fileobj=io.BytesIO(archive), mode="r:gz") as source:
+    source.extractall(extract_dir, filter="data")
+roots = list(extract_dir.iterdir())
+if len(roots) != 1 or not roots[0].is_dir():
+    raise RuntimeError("unexpected RocksDB archive layout")
+shutil.move(str(roots[0]), "/opt/rocksdb-clean")
+shutil.rmtree(extract_dir)
+PY
+
+RUN printf '%s\n' "${ROCKSDB_COMMIT}" \
+        > /opt/rocksdb-clean/.frontier-upstream-commit && \
+    git -C /opt/rocksdb-clean init -q && \
+    git -C /opt/rocksdb-clean config user.name "Frontier-CS" && \
+    git -C /opt/rocksdb-clean config user.email frontier-cs@example.invalid && \
+    git -C /opt/rocksdb-clean add -A && \
+    GIT_AUTHOR_DATE=2026-01-01T00:00:00Z \
+    GIT_COMMITTER_DATE=2026-01-01T00:00:00Z \
+        git -C /opt/rocksdb-clean commit -q -m "Pinned RocksDB ${ROCKSDB_COMMIT}" && \
+    cd /opt/rocksdb-clean && \
+    PORTABLE=1 DEBUG_LEVEL=0 DISABLE_WARNING_AS_ERROR=1 \
+        make -j"${ROCKSDB_BUILD_JOBS}" static_lib && \
+    cp -a /opt/rocksdb-clean /opt/rocksdb-vanilla
+
+COPY judge/harness/lsm_policy_harness.cc /judge/harness/lsm_policy_harness.cc
+
+WORKDIR /judge

2.0/problems/rocksdb_native_compaction_policy/evaluate.sh

@@ -0,0 +1,19 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+
+if [[ $# -gt 0 ]]; then
+  exec python3 "$SCRIPT_DIR/evaluator.py" "$1"
+fi
+
+for candidate in \
+  /work/execution_env/solution_env/solution.patch \
+  /work/execution_env/solution_env/solution.cpp; do
+  if [[ -f "$candidate" ]]; then
+    exec python3 "$SCRIPT_DIR/evaluator.py" "$candidate"
+  fi
+done
+
+echo "Error: solution artifact is missing" >&2
+exit 1