adityamukho · Oct 30, 2026 · Oct 29, 2026
@@ -0,0 +1,40 @@
+#!/usr/bin/env python3
+"""Compare dist thin-LTO and release full-LTO on a fixed CLI workload.
+
+Usage: cli_compare.py DIST_EXE RELEASE_EXE OUTPUT_JSON
+Includes startup, ingestion and printing; not a general query benchmark.
+"""
+from concurrent.futures import ThreadPoolExecutor
+import json
+from pathlib import Path
+import subprocess
+import sys
+import time
+
+base, candidate, output = sys.argv[1:]
+ingest = "(transact [" + " ".join(f"[:e{i} :v {i % 20}]" for i in range(2000)) + "])\n"
+query = "(query [:find (count ?e) :where [?e :v ?v]])\n"
+commands = (ingest + query * 400).encode()
+
+
+def run(binary, node):
+    start = time.perf_counter()
+    r = subprocess.run(["numactl", f"--cpunodebind={node}", f"--membind={node}", binary],
+                       input=commands, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=True)
+    elapsed = time.perf_counter() - start
+    assert not r.stderr, r.stderr
+    assert r.stdout.splitlines().count(b"2000") == 400
+    return elapsed
+
+
+rows = []
+with ThreadPoolExecutor(max_workers=2) as pool:
+    for i in range(8):
+        a_node, b_node = (1, 3) if i % 2 == 0 else (3, 1)
+        a = pool.submit(run, base, a_node)
+        b = pool.submit(run, candidate, b_node)
+        row = {"round": i, "base_node": a_node, "candidate_node": b_node,
+               "dist_seconds": a.result(), "release_seconds": b.result()}
+        rows.append(row)
+        Path(output).write_text(json.dumps(rows, indent=2) + "\n")
+        print(row, flush=True)
@@ -0,0 +1,44 @@
+#!/usr/bin/env python3
+"""Run the existing envelope query probes on opposite sockets, then swap.
+
+Usage: compare.py BASE_EXE CANDIDATE_EXE FIXTURE OUTPUT_JSON [ROUNDS] [INDICES]
+INDICES is a comma-separated subset of the envelope's query-shape indices.
+The fixture and adjacent .json metadata must already exist. Run only when
+builds/tests have finished. This measures warm read queries, not fsync latency.
+"""
+from concurrent.futures import ThreadPoolExecutor
+import json
+from pathlib import Path
+import subprocess
+import sys
+
+base, candidate, fixture, output = sys.argv[1:5]
+round_count = int(sys.argv[5]) if len(sys.argv) > 5 else 4
+# Representative point, scans, joins, not/or, rules, temporal, grouped/window
+# and long-history reads from the pinned perf_envelope source.
+indices = [0, 1, 2, 3, 5, 7, 8, 9, 10, 12, 13, 14, 19, 20, 21, 22, 23]
+if len(sys.argv) > 6:
+    indices = [int(i) for i in sys.argv[6].split(",")]
+
+
+def run(binary, node, index):
+    cmd = ["numactl", f"--cpunodebind={node}", f"--membind={node}",
+           binary, "__query", fixture, str(index), "2", "30"]
+    return json.loads(subprocess.check_output(cmd, text=True))
+
+
+rows = []
+with ThreadPoolExecutor(max_workers=2) as pool:
+    for round_no in range(round_count):
+        base_node, candidate_node = (1, 3) if round_no % 2 == 0 else (3, 1)
+        for index in indices:
+            a = pool.submit(run, base, base_node, index)
+            b = pool.submit(run, candidate, candidate_node, index)
+            a, b = a.result(), b.result()
+            assert not any(k in r for r in (a, b) for k in ("error", "capped_ms")), (a, b)
+            assert a["name"] == b["name"] and a["rows"] == b["rows"], (a, b)
+            row = {"round": round_no, "base_node": base_node, "candidate_node": candidate_node,
+                   "base": a, "candidate": b}
+            rows.append(row)
+            Path(output).write_text(json.dumps(rows, indent=2) + "\n")
+            print(round_no, a["name"], a["p50_ms"], b["p50_ms"], flush=True)