bench: Cloud benchmark

This commit is contained in:
2026-03-19 12:50:13 +01:00
parent f0ef42fe3f
commit 833c85f4ac
9 changed files with 2036 additions and 929 deletions

View File

@@ -7,100 +7,178 @@ cd "$ROOT_DIR"
usage() {
cat <<'EOF'
usage:
./scripts/run_bench_update.sh [machine_id]
./scripts/run_bench_update.sh [machine_id|all]
./scripts/run_bench_update.sh --machine <machine_id|all> [--run-id <run_id>]
./scripts/run_bench_update.sh --list
Regenerates bench/chart.svg and updates the benchmark table in README.md
from collected data in bench/history.jsonl.
Arguments:
machine_id Optional. Filter to a specific machine's data.
Default: current machine (hostname -s)
Use "all" to include all machines (will use latest entry per tag)
Examples:
# Update chart for current machine
./scripts/run_bench_update.sh
# Update chart for specific machine
./scripts/run_bench_update.sh my-server
# Update chart using all machines (latest entry per tag wins)
./scripts/run_bench_update.sh all
Options:
--machine <id|all> Filter by machine_id (default: hostname -s)
--run-id <id> Filter to an exact run_id
--history-file <path> History JSONL file (default: bench/history.jsonl)
--list List available machines and runs, then exit
-h, --help
EOF
}
if [[ "${1:-}" == "-h" || "${1:-}" == "--help" ]]; then
usage
exit 0
fi
# --- Configuration -----------------------------------------------------------
BENCH_DIR="$ROOT_DIR/bench"
HISTORY_FILE="$BENCH_DIR/history.jsonl"
CHART_FILE="$BENCH_DIR/chart.svg"
GNUPLOT_TEMPLATE="$BENCH_DIR/chart.gnuplot"
README_FILE="$ROOT_DIR/README.md"
MACHINE_ID="${1:-$(hostname -s)}"
MACHINE_ID="$(hostname -s)"
RUN_ID=""
LIST_ONLY=0
POSITIONAL_MACHINE=""
while [[ $# -gt 0 ]]; do
case "$1" in
-h|--help)
usage
exit 0
;;
--machine)
MACHINE_ID="$2"
shift 2
;;
--run-id)
RUN_ID="$2"
shift 2
;;
--history-file)
HISTORY_FILE="$2"
shift 2
;;
--list)
LIST_ONLY=1
shift
;;
*)
if [[ -z "$POSITIONAL_MACHINE" ]]; then
POSITIONAL_MACHINE="$1"
shift
else
echo "Unexpected argument: $1" >&2
usage
exit 1
fi
;;
esac
done
if [[ -n "$POSITIONAL_MACHINE" ]]; then
MACHINE_ID="$POSITIONAL_MACHINE"
fi
if [[ ! -f "$HISTORY_FILE" ]]; then
echo "Error: No history file found at $HISTORY_FILE" >&2
echo "Run ./scripts/run_bench_collect.sh first to collect benchmark data" >&2
echo "Run ./scripts/run_bench_collect.sh or ./scripts/run_bench_cloud.sh first" >&2
exit 1
fi
if [[ "$LIST_ONLY" == "1" ]]; then
node - "$HISTORY_FILE" <<'NODE'
const fs = require("node:fs");
const [, , historyFile] = process.argv;
const entries = fs.readFileSync(historyFile, "utf8")
.split("\n")
.filter((l) => l.trim().length > 0)
.map((l) => JSON.parse(l));
if (entries.length === 0) {
console.log("No entries in history file.");
process.exit(0);
}
entries.sort((a, b) => b.timestamp.localeCompare(a.timestamp));
const machines = new Map();
for (const e of entries) {
const machineId = e.machine_id || "unknown";
const prev = machines.get(machineId);
if (!prev) {
machines.set(machineId, { count: 1, latest: e });
} else {
prev.count += 1;
if ((e.timestamp || "") > (prev.latest.timestamp || "")) prev.latest = e;
}
}
console.log("Machines:");
console.log(" machine_id entries latest_timestamp latest_tag");
for (const [machineId, info] of [...machines.entries()].sort((a, b) => a[0].localeCompare(b[0]))) {
const id = machineId.padEnd(34, " ");
const count = String(info.count).padStart(7, " ");
const ts = (info.latest.timestamp || "").padEnd(24, " ");
const tag = info.latest.git_tag || "";
console.log(` ${id} ${count} ${ts} ${tag}`);
}
console.log("\nRuns (newest first):");
console.log(" timestamp run_id machine_id source git_tag targets");
for (const e of entries) {
const ts = (e.timestamp || "").slice(0, 19).padEnd(24, " ");
const runId = (e.run_id || "").slice(0, 36).padEnd(36, " ");
const machineId = (e.machine_id || "").slice(0, 24).padEnd(24, " ");
const source = (e.source?.kind || "").padEnd(6, " ");
const tag = (e.git_tag || "").slice(0, 16).padEnd(16, " ");
const targets = (e.bench?.targets || Object.keys(e.servers || {})).join(",");
console.log(` ${ts} ${runId} ${machineId} ${source} ${tag} ${targets}`);
}
NODE
exit 0
fi
WORK_DIR="$(mktemp -d)"
trap 'rm -rf "$WORK_DIR"' EXIT
# --- Generate chart ----------------------------------------------------------
echo "Generating chart (machine=$MACHINE_ID${RUN_ID:+, run_id=$RUN_ID})"
echo "Generating chart for machine: $MACHINE_ID"
node - "$HISTORY_FILE" "$MACHINE_ID" "$WORK_DIR" <<'NODE'
if ! node - "$HISTORY_FILE" "$MACHINE_ID" "$RUN_ID" "$WORK_DIR" <<'NODE'
const fs = require("node:fs");
const path = require("node:path");
const [, , historyFile, machineId, workDir] = process.argv;
if (!fs.existsSync(historyFile)) {
console.log(" no history file, skipping chart generation");
process.exit(0);
}
const lines = fs.readFileSync(historyFile, "utf8")
.split("\n")
.filter(l => l.trim().length > 0)
.map(l => JSON.parse(l));
// Filter to selected machine(s)
let entries;
if (machineId === "all") {
entries = lines;
console.log(" using all machines");
} else {
entries = lines.filter(e => e.machine_id === machineId);
console.log(" filtered to machine: " + machineId);
}
if (entries.length === 0) {
console.log(" no history entries for machine '" + machineId + "', skipping chart");
process.exit(0);
}
// Sort chronologically, deduplicate by tag (latest wins),
// then order the resulting series by git tag.
entries.sort((a, b) => a.timestamp.localeCompare(b.timestamp));
const byTag = new Map();
for (const e of entries) {
byTag.set(e.git_tag, e);
}
const deduped = [...byTag.values()];
const [, , historyFile, machineId, runId, workDir] = process.argv;
function parseSemverTag(tag) {
const match = /^v?(\d+)\.(\d+)\.(\d+)$/.exec(tag);
const match = /^v?(\d+)\.(\d+)\.(\d+)$/.exec(tag || "");
return match ? match.slice(1).map(Number) : null;
}
const all = fs.readFileSync(historyFile, "utf8")
.split("\n")
.filter((l) => l.trim().length > 0)
.map((l) => JSON.parse(l));
let selected = all;
if (runId && runId.length > 0) {
selected = all.filter((e) => e.run_id === runId);
console.log(` filtered by run_id: ${runId}`);
} else if (machineId !== "all") {
selected = all.filter((e) => e.machine_id === machineId);
console.log(` filtered to machine: ${machineId}`);
} else {
console.log(" using all machines");
}
if (selected.length === 0) {
console.error(" no matching history entries");
process.exit(1);
}
selected.sort((a, b) => (a.timestamp || "").localeCompare(b.timestamp || ""));
const byTag = new Map();
for (const e of selected) {
byTag.set(e.git_tag || "untagged", e);
}
const deduped = [...byTag.values()];
deduped.sort((a, b) => {
const aTag = parseSemverTag(a.git_tag);
const bTag = parseSemverTag(b.git_tag);
@@ -109,24 +187,19 @@ deduped.sort((a, b) => {
return aTag[0] - bTag[0] || aTag[1] - bTag[1] || aTag[2] - bTag[2];
}
return a.git_tag.localeCompare(b.git_tag, undefined, { numeric: true });
return (a.git_tag || "").localeCompare(b.git_tag || "", undefined, { numeric: true });
});
// Determine which non-parrhesia servers are present
const baselineServerNames = ["strfry", "nostr-rs-relay"];
const presentBaselines = baselineServerNames.filter(srv =>
deduped.some(e => e.servers[srv])
);
const presentBaselines = baselineServerNames.filter((srv) => deduped.some((e) => e.servers?.[srv]));
// Metrics to chart
const chartMetrics = [
{ key: "event_tps", label: "Event Throughput (TPS) — higher is better", file: "event_tps.tsv", ylabel: "TPS" },
{ key: "req_tps", label: "Req Throughput (TPS) — higher is better", file: "req_tps.tsv", ylabel: "TPS" },
{ key: "echo_tps", label: "Echo Throughput (TPS) — higher is better", file: "echo_tps.tsv", ylabel: "TPS" },
{ key: "connect_avg_ms", label: "Connect Avg Latency (ms) — lower is better", file: "connect_avg_ms.tsv", ylabel: "ms" },
{ key: "event_tps", label: "Event Throughput (TPS) — higher is better", file: "event_tps.tsv", ylabel: "TPS" },
{ key: "req_tps", label: "Req Throughput (TPS) — higher is better", file: "req_tps.tsv", ylabel: "TPS" },
{ key: "echo_tps", label: "Echo Throughput (TPS) — higher is better", file: "echo_tps.tsv", ylabel: "TPS" },
{ key: "connect_avg_ms", label: "Connect Avg Latency (ms) — lower is better", file: "connect_avg_ms.tsv", ylabel: "ms" },
];
// Write per-metric TSV files
for (const cm of chartMetrics) {
const header = ["tag", "parrhesia-pg", "parrhesia-memory"];
for (const srv of presentBaselines) header.push(srv);
@@ -134,12 +207,12 @@ for (const cm of chartMetrics) {
const rows = [header.join("\t")];
for (const e of deduped) {
const row = [
e.git_tag,
e.servers["parrhesia-pg"]?.[cm.key] ?? "NaN",
e.servers["parrhesia-memory"]?.[cm.key] ?? "NaN",
e.git_tag || "untagged",
e.servers?.["parrhesia-pg"]?.[cm.key] ?? "NaN",
e.servers?.["parrhesia-memory"]?.[cm.key] ?? "NaN",
];
for (const srv of presentBaselines) {
row.push(e.servers[srv]?.[cm.key] ?? "NaN");
row.push(e.servers?.[srv]?.[cm.key] ?? "NaN");
}
rows.push(row.join("\t"));
}
@@ -147,7 +220,6 @@ for (const cm of chartMetrics) {
fs.writeFileSync(path.join(workDir, cm.file), rows.join("\n") + "\n", "utf8");
}
// Generate gnuplot plot commands (handles variable column counts)
const serverLabels = ["parrhesia-pg", "parrhesia-memory"];
for (const srv of presentBaselines) serverLabels.push(srv);
@@ -158,10 +230,9 @@ for (const cm of chartMetrics) {
plotLines.push(`set ylabel "${cm.ylabel}"`);
const plotParts = [];
// Column 2 = parrhesia-pg, 3 = parrhesia-memory, 4+ = baselines
plotParts.push(`${dataFile} using 0:2:xtic(1) lt 1 title "${serverLabels[0]}"`);
plotParts.push(`'' using 0:3 lt 2 title "${serverLabels[1]}"`);
for (let i = 0; i < presentBaselines.length; i++) {
for (let i = 0; i < presentBaselines.length; i += 1) {
plotParts.push(`'' using 0:${4 + i} lt ${3 + i} title "${serverLabels[2 + i]}"`);
}
@@ -169,14 +240,22 @@ for (const cm of chartMetrics) {
plotLines.push("");
}
fs.writeFileSync(
path.join(workDir, "plot_commands.gnuplot"),
plotLines.join("\n") + "\n",
"utf8"
);
fs.writeFileSync(path.join(workDir, "plot_commands.gnuplot"), plotLines.join("\n") + "\n", "utf8");
console.log(" " + deduped.length + " tag(s), " + presentBaselines.length + " baseline server(s)");
const latestForReadme = [...selected]
.sort((a, b) => (b.timestamp || "").localeCompare(a.timestamp || ""))
.find((e) => e.servers?.["parrhesia-pg"] && e.servers?.["parrhesia-memory"]);
if (latestForReadme) {
fs.writeFileSync(path.join(workDir, "latest_entry.json"), JSON.stringify(latestForReadme), "utf8");
}
console.log(` selected=${selected.length}, series_tags=${deduped.length}, baselines=${presentBaselines.length}`);
NODE
then
echo "No matching data for chart/update" >&2
exit 1
fi
if [[ -f "$WORK_DIR/plot_commands.gnuplot" ]]; then
gnuplot \
@@ -185,52 +264,26 @@ if [[ -f "$WORK_DIR/plot_commands.gnuplot" ]]; then
"$GNUPLOT_TEMPLATE"
echo " chart written to $CHART_FILE"
else
echo " chart generation skipped (no data for this machine)"
exit 0
echo " chart generation skipped"
fi
# --- Update README.md -------------------------------------------------------
echo "Updating README.md with latest benchmark..."
# Find the most recent entry for this machine
LATEST_ENTRY=$(node - "$HISTORY_FILE" "$MACHINE_ID" <<'NODE'
const fs = require("node:fs");
const [, , historyFile, machineId] = process.argv;
const lines = fs.readFileSync(historyFile, "utf8")
.split("\n")
.filter(l => l.trim().length > 0)
.map(l => JSON.parse(l));
let entries;
if (machineId === "all") {
entries = lines;
} else {
entries = lines.filter(e => e.machine_id === machineId);
}
if (entries.length === 0) {
console.error("No entries found for machine: " + machineId);
process.exit(1);
}
// Get latest entry
entries.sort((a, b) => b.timestamp.localeCompare(a.timestamp));
console.log(JSON.stringify(entries[0]));
NODE
)
if [[ -z "$LATEST_ENTRY" ]]; then
echo "Warning: Could not find latest entry, skipping README update" >&2
if [[ ! -f "$WORK_DIR/latest_entry.json" ]]; then
echo "Warning: no selected entry contains both parrhesia-pg and parrhesia-memory; skipping README table update" >&2
echo
echo "Benchmark rendering complete. Files updated:"
echo " $CHART_FILE"
echo
exit 0
fi
node - "$LATEST_ENTRY" "$ROOT_DIR/README.md" <<'NODE'
LATEST_ENTRY="$(cat "$WORK_DIR/latest_entry.json")"
node - "$LATEST_ENTRY" "$README_FILE" <<'NODE'
const fs = require("node:fs");
const [, , entryJson, readmePath] = process.argv;
const entry = JSON.parse(entryJson);
const servers = entry.servers || {};
@@ -240,11 +293,7 @@ const strfry = servers["strfry"];
const nostrRs = servers["nostr-rs-relay"];
if (!pg || !mem) {
const present = Object.keys(servers).sort().join(", ") || "(none)";
console.error(
"Latest benchmark entry must include parrhesia-pg and parrhesia-memory. Present servers: " +
present
);
console.error("Selected entry is missing parrhesia-pg or parrhesia-memory");
process.exit(1);
}
@@ -259,26 +308,26 @@ function ratio(base, other) {
function boldIf(ratioStr, lowerIsBetter) {
if (ratioStr === "n/a") return ratioStr;
const num = parseFloat(ratioStr);
const num = Number.parseFloat(ratioStr);
if (!Number.isFinite(num)) return ratioStr;
const better = lowerIsBetter ? num < 1 : num > 1;
return better ? "**" + ratioStr + "**" : ratioStr;
return better ? `**${ratioStr}**` : ratioStr;
}
const metricRows = [
["connect avg latency (ms) ↓", "connect_avg_ms", true],
["connect max latency (ms) ↓", "connect_max_ms", true],
["echo throughput (TPS) ↑", "echo_tps", false],
["echo throughput (MiB/s) ↑", "echo_mibs", false],
["event throughput (TPS) ↑", "event_tps", false],
["event throughput (MiB/s) ↑", "event_mibs", false],
["req throughput (TPS) ↑", "req_tps", false],
["req throughput (MiB/s) ↑", "req_mibs", false],
["echo throughput (TPS) ↑", "echo_tps", false],
["echo throughput (MiB/s) ↑", "echo_mibs", false],
["event throughput (TPS) ↑", "event_tps", false],
["event throughput (MiB/s) ↑", "event_mibs", false],
["req throughput (TPS) ↑", "req_tps", false],
["req throughput (MiB/s) ↑", "req_mibs", false],
];
const hasStrfry = !!strfry;
const hasNostrRs = !!nostrRs;
// Build header
const header = ["metric", "parrhesia-pg", "parrhesia-mem"];
if (hasStrfry) header.push("strfry");
if (hasNostrRs) header.push("nostr-rs-relay");
@@ -287,7 +336,7 @@ if (hasStrfry) header.push("strfry/pg");
if (hasNostrRs) header.push("nostr-rs/pg");
const alignRow = ["---"];
for (let i = 1; i < header.length; i++) alignRow.push("---:");
for (let i = 1; i < header.length; i += 1) alignRow.push("---:");
const rows = metricRows.map(([label, key, lowerIsBetter]) => {
const row = [label, toFixed(pg[key]), toFixed(mem[key])];
@@ -304,13 +353,12 @@ const rows = metricRows.map(([label, key, lowerIsBetter]) => {
const tableLines = [
"| " + header.join(" | ") + " |",
"| " + alignRow.join(" | ") + " |",
...rows.map(r => "| " + r.join(" | ") + " |"),
...rows.map((r) => "| " + r.join(" | ") + " |"),
];
// Replace the first markdown table in the ## Benchmark section
const readme = fs.readFileSync(readmePath, "utf8");
const readmeLines = readme.split("\n");
const benchIdx = readmeLines.findIndex(l => /^## Benchmark/.test(l));
const lines = readme.split("\n");
const benchIdx = lines.findIndex((l) => /^## Benchmark/.test(l));
if (benchIdx === -1) {
console.error("Could not find '## Benchmark' section in README.md");
process.exit(1);
@@ -318,8 +366,8 @@ if (benchIdx === -1) {
let tableStart = -1;
let tableEnd = -1;
for (let i = benchIdx + 1; i < readmeLines.length; i++) {
if (readmeLines[i].startsWith("|")) {
for (let i = benchIdx + 1; i < lines.length; i += 1) {
if (lines[i].startsWith("|")) {
if (tableStart === -1) tableStart = i;
tableEnd = i;
} else if (tableStart !== -1) {
@@ -332,19 +380,19 @@ if (tableStart === -1) {
process.exit(1);
}
const before = readmeLines.slice(0, tableStart);
const after = readmeLines.slice(tableEnd + 1);
const updated = [...before, ...tableLines, ...after].join("\n");
const updated = [
...lines.slice(0, tableStart),
...tableLines,
...lines.slice(tableEnd + 1),
].join("\n");
fs.writeFileSync(readmePath, updated, "utf8");
console.log(" table updated (" + tableLines.length + " rows)");
console.log(` table updated (${tableLines.length} rows)`);
NODE
# --- Done ---------------------------------------------------------------------
echo
echo "Benchmark rendering complete. Files updated:"
echo " $CHART_FILE"
echo " $ROOT_DIR/README.md"
echo " $README_FILE"
echo
echo "Review with: git diff"