contrib: plot getMemoryLoad()

Sjors · Sjors · commit acd76876f376 · 2025-12-01T11:56:26.000+01:00
Add Python script to make plots.

Assisted-by: GitHub Copilot
Assisted-by: OpenAI GPT-5.1-Codex
diff --git a/.gitignore b/.gitignore
@@ -26,3 +26,5 @@ target/
 /guix-build-*
 
 /ci/scratch/
+
+plots
diff --git a/contrib/devtools/plot_memory_load.py b/contrib/devtools/plot_memory_load.py
@@ -0,0 +1,344 @@
+#!/usr/bin/env python3
+"""Plot GetTemplateMemoryUsage durations and template memory footprints over time."""
+from __future__ import annotations
+
+import argparse
+import datetime as dt
+import pickle
+import re
+from pathlib import Path
+from typing import List, Tuple
+
+import matplotlib.dates as mdates  # type: ignore[import, import-not-found]
+import matplotlib.pyplot as plt  # type: ignore[import, import-not-found]
+from matplotlib import transforms  # type: ignore[import, import-not-found]
+from matplotlib.ticker import NullFormatter  # type: ignore[import, import-not-found]
+
+MEMORY_LOAD_COMPLETED = re.compile(
+    r"^(?P<ts>\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}\.\d+Z).*GetTemplateMemoryUsage:.*completed \((?P<ms>\d+\.\d+)ms\)"
+)
+UPDATE_TIP = re.compile(
+    r"^(?P<ts>\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}\.\d+Z).*UpdateTip:.*height=(?P<height>\d+) "
+)
+TP_LOG_ENTRY = re.compile(
+    r"^(?P<ts>\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(?:\.\d+)?)Z.*Template memory footprint (?P<mib>\d+(?:\.\d+)?) MiB"
+)
+
+TIP_LABEL_MIN_FRACTION = 0.02
+CLIPPED_LABEL_MIN_FRACTION = 0.01
+DEFAULT_DURATION_CAP_MS = 15.0  # Hard cap for GetTemplateMemoryUsage durations
+
+
+def format_with_machine(base: str, machine: str | None) -> str:
+    return f"{base} ({machine})" if machine else base
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "bitcoin_log",
+        type=Path,
+        help="Path to bitcoin debug log containing GetTemplateMemoryUsage entries",
+    )
+    parser.add_argument(
+        "--output",
+        type=Path,
+        default=Path("getmemoryload-scatter.svg"),
+        help="Destination image path (default: ./getmemoryload-scatter.svg)",
+    )
+    parser.add_argument(
+        "--show",
+        action="store_true",
+        help="Display the plot interactively instead of saving it",
+    )
+    parser.add_argument(
+        "tp_log",
+        type=Path,
+        help="SV2 template provider log containing 'Template memory footprint' lines",
+    )
+    parser.add_argument(
+        "--duration-cap",
+        type=float,
+        default=DEFAULT_DURATION_CAP_MS,
+        help=(
+            "Clip GetTemplateMemoryUsage durations above this value (ms) and annotate their "
+            f"true value at the top of the chart. Default: {DEFAULT_DURATION_CAP_MS:.0f} ms."
+        ),
+    )
+    parser.add_argument(
+        "--figure-pickle",
+        type=Path,
+        default=None,
+        help=(
+            "Optional path to write a pickle of the Matplotlib Figure for further "
+            "inspection."
+        ),
+    )
+    parser.add_argument(
+        "--dpi",
+        type=int,
+        default=300,
+        help="Raster output resolution (applies to PNG/JPEG/TIFF). Default: 300 dpi.",
+    )
+    parser.add_argument(
+        "--machine",
+        type=str,
+        default=None,
+        help="Optional machine label to append to dataset names (e.g. 'M4').",
+    )
+    return parser.parse_args()
+
+
+def parse_timestamp(raw: str) -> dt.datetime:
+    return dt.datetime.fromisoformat(raw.replace("Z", "+00:00"))
+
+
+def load_events(log_path: Path):
+    mem_points = []
+    tip_events = []
+    with log_path.open("r", encoding="utf-8") as handle:
+        for line in handle:
+            mem_match = MEMORY_LOAD_COMPLETED.search(line)
+            if mem_match:
+                mem_points.append(
+                    (parse_timestamp(mem_match.group("ts")), float(mem_match.group("ms")))
+                )
+                continue
+            tip_match = UPDATE_TIP.search(line)
+            if tip_match:
+                tip_events.append(
+                    (parse_timestamp(tip_match.group("ts")), int(tip_match.group("height")))
+                )
+    mem_points.sort(key=lambda item: item[0])
+    tip_events.sort(key=lambda item: item[0])
+    return mem_points, tip_events
+
+
+def load_tp_points(log_path: Path) -> List[Tuple[dt.datetime, float]]:
+    tp_points: List[Tuple[dt.datetime, float]] = []
+    if not log_path.exists():
+        raise SystemExit(f"Template provider log '{log_path}' not found")
+
+    with log_path.open("r", encoding="utf-8") as handle:
+        for line in handle:
+            match = TP_LOG_ENTRY.search(line)
+            if match:
+                mib_val = float(match.group("mib"))
+                tp_points.append((parse_timestamp(match.group("ts")), mib_val))
+
+    tp_points.sort(key=lambda item: item[0])
+    if not tp_points:
+        raise SystemExit(
+            f"No template provider entries were found in '{log_path}'. Does it contain "
+            "'Template memory footprint' lines?"
+        )
+    return tp_points
+
+def plot_memory_load(
+    mem_points,
+    tip_events,
+    tp_points,
+    duration_cap,
+    figure_pickle: Path | None,
+    dpi: int,
+    machine_label: str | None,
+    output: Path,
+    show_plot: bool,
+) -> None:
+    if not mem_points:
+        raise SystemExit("No GetTemplateMemoryUsage() completion records found in log")
+
+    times, durations = zip(*mem_points)
+    fig, ax = plt.subplots(figsize=(15, 5))
+    fig.patch.set_facecolor("white")
+    fig.patch.set_alpha(1.0)
+    ax.set_facecolor("white")
+    duration_label = format_with_machine("GetTemplateMemoryUsage() duration", machine_label)
+    ax.scatter(
+        times,
+        durations,
+        s=5,
+        color="tab:blue",
+        label=duration_label,
+        zorder=3,
+    )
+
+    ax.set_ylabel("Duration (ms)")
+    ax.set_xlabel("")
+    ax.set_title("")
+    ax.xaxis.set_major_formatter(NullFormatter())
+    ax.margins(x=0)
+
+    max_duration = max(durations)
+    effective_cap = duration_cap
+    clipped_points = []
+    if effective_cap is not None and effective_cap < max_duration:
+        ax.set_ylim(bottom=0, top=effective_cap)
+        clipped_points = [(timestamp, value) for timestamp, value in mem_points if value > effective_cap]
+    else:
+        upper_pad = max_duration * 1.05 if max_duration > 0 else 1
+        ax.set_ylim(bottom=0, top=upper_pad)
+
+    twin_ax = None
+    if tp_points:
+        tp_times, tp_mib = zip(*tp_points)
+        twin_ax = ax.twinx()
+        twin_ax.margins(x=0)
+        twin_ax.vlines(
+            tp_times,
+            [0] * len(tp_times),
+            tp_mib,
+            color="#4fbf73",
+            alpha=0.5,
+            linewidth=1,
+            zorder=1,
+            label="Template provider memory",
+        )
+        twin_ax.set_ylabel("Memory footprint (MiB)", color="#2b8a46")
+        twin_ax.tick_params(axis="y", colors="#2b8a46")
+        twin_ax.set_ylim(bottom=0)
+
+    if clipped_points:
+        clip_trans = transforms.blended_transform_factory(ax.transData, ax.transAxes)
+        marker_y = 1.0
+        clipped_times = [timestamp for timestamp, _ in clipped_points]
+        clipped_values = [value for _, value in clipped_points]
+        clipped_time_nums = mdates.date2num(clipped_times)
+        ax.scatter(
+            clipped_time_nums,
+            [marker_y] * len(clipped_points),
+            marker="o",
+            s=22,
+            facecolors="none",
+            edgecolors="tab:blue",
+            linewidths=1,
+            label=f"Clipped duration (> {effective_cap:.2f} ms)",
+            transform=clip_trans,
+            clip_on=False,
+            zorder=4,
+        )
+        label_min_delta = (ax.get_xlim()[1] - ax.get_xlim()[0]) * CLIPPED_LABEL_MIN_FRACTION
+        last_clipped_label_num: float | None = None
+        for ts_num, value in zip(clipped_time_nums, clipped_values):
+            if last_clipped_label_num is not None and ts_num - last_clipped_label_num < label_min_delta:
+                continue
+            label = f"{value:.0f} ms"
+            ax.annotate(
+                label,
+                xy=(ts_num, marker_y),
+                xycoords=clip_trans,
+                xytext=(0, -8),
+                textcoords="offset points",
+                ha="center",
+                va="top",
+                fontsize=8,
+                color="tab:blue",
+                rotation=270,
+                clip_on=False,
+            )
+            last_clipped_label_num = ts_num
+
+    if tip_events:
+        tip_color = "tab:red"
+        tip_times, tip_heights = zip(*tip_events)
+        tip_time_nums = mdates.date2num(tip_times)
+        trans = transforms.blended_transform_factory(ax.transData, ax.transAxes)
+        marker_y = -0.01
+        ax.scatter(
+            tip_time_nums,
+            [marker_y] * len(tip_events),
+            marker="^",
+            s=70,
+            color=tip_color,
+            edgecolors="none",
+            label="Tip update",
+            zorder=4,
+            transform=trans,
+            clip_on=False,
+        )
+        label_trans = transforms.blended_transform_factory(ax.transData, ax.transAxes)
+        x_start, x_end = ax.get_xlim()
+        min_delta = (x_end - x_start) * TIP_LABEL_MIN_FRACTION
+        last_label_num: float | None = None
+        for ts_num, height in zip(tip_time_nums, tip_heights):
+            if last_label_num is not None and ts_num - last_label_num < min_delta:
+                continue
+            label = f"{height:,}"
+            ax.annotate(
+                label,
+                xy=(ts_num, 0),
+                xycoords=label_trans,
+                xytext=(1, -8),
+                textcoords="offset points",
+                rotation=45,
+                ha="right",
+                va="top",
+                color=tip_color,
+                fontsize=8,
+                clip_on=False,
+            )
+            last_label_num = ts_num
+
+    handles, labels = ax.get_legend_handles_labels()
+    if twin_ax is not None:
+        twin_handles, twin_labels = twin_ax.get_legend_handles_labels()
+        handles += twin_handles
+        labels += twin_labels
+    try:
+        tip_idx = labels.index("Tip update")
+        tip_handle = handles.pop(tip_idx)
+        tip_label = labels.pop(tip_idx)
+        handles.append(tip_handle)
+        labels.append(tip_label)
+    except ValueError:
+        pass
+    legend = fig.legend(
+        handles,
+        labels,
+        loc="upper left",
+        bbox_to_anchor=(0.12, 0.78),
+    )
+    legend.set_zorder(10)
+    frame = legend.get_frame()
+    frame.set_alpha(0.75)
+    frame.set_zorder(legend.get_zorder())
+    ax.grid(True, axis="y", linestyle="--", alpha=0.3)
+
+    if figure_pickle is not None:
+        figure_pickle.parent.mkdir(parents=True, exist_ok=True)
+        with figure_pickle.open("wb") as handle:
+            pickle.dump(fig, handle)
+
+    if show_plot:
+        plt.show()
+    else:
+        output.parent.mkdir(parents=True, exist_ok=True)
+        raster_exts = {".png", ".jpg", ".jpeg", ".tif", ".tiff", ".bmp", ".gif"}
+        if output.suffix.lower() in raster_exts:
+            fig.savefig(output, bbox_inches="tight", facecolor=fig.get_facecolor(), dpi=dpi)
+        else:
+            fig.savefig(output, bbox_inches="tight", facecolor=fig.get_facecolor())
+        print(f"Wrote {output}")
+
+    plt.close(fig)
+
+
+def main() -> None:
+    args = parse_args()
+    mem_points, tip_events = load_events(args.bitcoin_log)
+    tp_points = load_tp_points(args.tp_log)
+    plot_memory_load(
+        mem_points,
+        tip_events,
+        tp_points,
+        args.duration_cap,
+        args.figure_pickle,
+        args.dpi,
+        args.machine,
+        args.output,
+        args.show,
+    )
+
+
+if __name__ == "__main__":
+    main()
