klipper-dgus/scripts/graphstats.py

#!/usr/bin/env python
# Script to parse a logging file, extract the stats, and graph them
#
# Copyright (C) 2016-2021  Kevin O'Connor <kevin@koconnor.net>
#
# This file may be distributed under the terms of the GNU GPLv3 license.
import optparse, datetime
import matplotlib

MAXBANDWIDTH=25000.
MAXBUFFER=2.
STATS_INTERVAL=5.
TASK_MAX=0.0025

APPLY_PREFIX = [
    'mcu_awake', 'mcu_task_avg', 'mcu_task_stddev', 'bytes_write',
    'bytes_read', 'bytes_retransmit', 'freq', 'adj',
    'target', 'temp', 'pwm'
]

def parse_log(logname, mcu):
    if mcu is None:
        mcu = "mcu"
    mcu_prefix = mcu + ":"
    apply_prefix = { p: 1 for p in APPLY_PREFIX }
    f = open(logname, 'r')
    out = []
    for line in f:
        parts = line.split()
        if not parts or parts[0] not in ('Stats', 'INFO:root:Stats'):
            #if parts and parts[0] == 'INFO:root:shutdown:':
            #    break
            continue
        prefix = ""
        keyparts = {}
        for p in parts[2:]:
            if '=' not in p:
                prefix = p
                if prefix == mcu_prefix:
                    prefix = ''
                continue
            name, val = p.split('=', 1)
            if name in apply_prefix:
                name = prefix + name
            keyparts[name] = val
        if 'print_time' not in keyparts:
            continue
        keyparts['#sampletime'] = float(parts[1][:-1])
        out.append(keyparts)
    f.close()
    return out

def setup_matplotlib(output_to_file):
    global matplotlib
    if output_to_file:
        matplotlib.use('Agg')
    import matplotlib.pyplot, matplotlib.dates, matplotlib.font_manager
    import matplotlib.ticker

def find_print_restarts(data):
    runoff_samples = {}
    last_runoff_start = last_buffer_time = last_sampletime = 0.
    last_print_stall = 0
    for d in reversed(data):
        # Check for buffer runoff
        sampletime = d['#sampletime']
        buffer_time = float(d.get('buffer_time', 0.))
        if (last_runoff_start and last_sampletime - sampletime < 5
            and buffer_time > last_buffer_time):
            runoff_samples[last_runoff_start][1].append(sampletime)
        elif buffer_time < 1.:
            last_runoff_start = sampletime
            runoff_samples[last_runoff_start] = [False, [sampletime]]
        else:
            last_runoff_start = 0.
        last_buffer_time = buffer_time
        last_sampletime = sampletime
        # Check for print stall
        print_stall = int(d['print_stall'])
        if print_stall < last_print_stall:
            if last_runoff_start:
                runoff_samples[last_runoff_start][0] = True
        last_print_stall = print_stall
    sample_resets = {sampletime: 1 for stall, samples in runoff_samples.values()
                     for sampletime in samples if not stall}
    return sample_resets

def plot_mcu(data, maxbw):
    # Generate data for plot
    basetime = lasttime = data[0]['#sampletime']
    lastbw = float(data[0]['bytes_write']) + float(data[0]['bytes_retransmit'])
    sample_resets = find_print_restarts(data)
    times = []
    bwdeltas = []
    loads = []
    awake = []
    hostbuffers = []
    for d in data:
        st = d['#sampletime']
        timedelta = st - lasttime
        if timedelta <= 0.:
            continue
        bw = float(d['bytes_write']) + float(d['bytes_retransmit'])
        if bw < lastbw:
            lastbw = bw
            continue
        load = float(d['mcu_task_avg']) + 3*float(d['mcu_task_stddev'])
        if st - basetime < 15.:
            load = 0.
        pt = float(d['print_time'])
        hb = float(d['buffer_time'])
        if hb >= MAXBUFFER or st in sample_resets:
            hb = 0.
        else:
            hb = 100. * (MAXBUFFER - hb) / MAXBUFFER
        hostbuffers.append(hb)
        times.append(datetime.datetime.utcfromtimestamp(st))
        bwdeltas.append(100. * (bw - lastbw) / (maxbw * timedelta))
        loads.append(100. * load / TASK_MAX)
        awake.append(100. * float(d.get('mcu_awake', 0.)) / STATS_INTERVAL)
        lasttime = st
        lastbw = bw

    # Build plot
    fig, ax1 = matplotlib.pyplot.subplots()
    ax1.set_title("MCU bandwidth and load utilization")
    ax1.set_xlabel('Time')
    ax1.set_ylabel('Usage (%)')
    ax1.plot_date(times, bwdeltas, 'g', label='Bandwidth', alpha=0.8)
    ax1.plot_date(times, loads, 'r', label='MCU load', alpha=0.8)
    ax1.plot_date(times, hostbuffers, 'c', label='Host buffer', alpha=0.8)
    ax1.plot_date(times, awake, 'y', label='Awake time', alpha=0.6)
    fontP = matplotlib.font_manager.FontProperties()
    fontP.set_size('x-small')
    ax1.legend(loc='best', prop=fontP)
    ax1.xaxis.set_major_formatter(matplotlib.dates.DateFormatter('%H:%M'))
    ax1.grid(True)
    return fig

def plot_system(data):
    # Generate data for plot
    lasttime = data[0]['#sampletime']
    lastcputime = float(data[0]['cputime'])
    times = []
    sysloads = []
    cputimes = []
    memavails = []
    for d in data:
        st = d['#sampletime']
        timedelta = st - lasttime
        if timedelta <= 0.:
            continue
        lasttime = st
        times.append(datetime.datetime.utcfromtimestamp(st))
        cputime = float(d['cputime'])
        cpudelta = max(0., min(1.5, (cputime - lastcputime) / timedelta))
        lastcputime = cputime
        cputimes.append(cpudelta * 100.)
        sysloads.append(float(d['sysload']) * 100.)
        memavails.append(float(d['memavail']))

    # Build plot
    fig, ax1 = matplotlib.pyplot.subplots()
    ax1.set_title("System load utilization")
    ax1.set_xlabel('Time')
    ax1.set_ylabel('Load (% of a core)')
    ax1.plot_date(times, sysloads, '-', label='system load',
                  color='cyan', alpha=0.8)
    ax1.plot_date(times, cputimes, '-', label='process time',
                  color='red', alpha=0.8)
    ax2 = ax1.twinx()
    ax2.set_ylabel('Available memory (KB)')
    ax2.plot_date(times, memavails, '-', label='system memory',
                  color='yellow', alpha=0.3)
    fontP = matplotlib.font_manager.FontProperties()
    fontP.set_size('x-small')
    ax1li, ax1la = ax1.get_legend_handles_labels()
    ax2li, ax2la = ax2.get_legend_handles_labels()
    ax1.legend(ax1li + ax2li, ax1la + ax2la, loc='best', prop=fontP)
    ax1.xaxis.set_major_formatter(matplotlib.dates.DateFormatter('%H:%M'))
    ax1.grid(True)
    return fig

def plot_mcu_frequencies(data):
    all_keys = {}
    for d in data:
        all_keys.update(d)
    graph_keys = { key: ([], []) for key in all_keys
                   if (key in ("freq", "adj")
                       or (key.endswith(":freq") or key.endswith(":adj"))) }
    for d in data:
        st = datetime.datetime.utcfromtimestamp(d['#sampletime'])
        for key, (times, values) in graph_keys.items():
            val = d.get(key)
            if val not in (None, '0', '1'):
                times.append(st)
                values.append(float(val))
    est_mhz = { key: round((sum(values)/len(values)) / 1000000.)
                for key, (times, values) in graph_keys.items() }

    # Build plot
    fig, ax1 = matplotlib.pyplot.subplots()
    ax1.set_title("MCU frequencies")
    ax1.set_xlabel('Time')
    ax1.set_ylabel('Microsecond deviation')
    for key in sorted(graph_keys):
        times, values = graph_keys[key]
        mhz = est_mhz[key]
        label = "%s(%dMhz)" % (key, mhz)
        hz = mhz * 1000000.
        ax1.plot_date(times, [(v - hz)/mhz for v in values], '.', label=label)
    fontP = matplotlib.font_manager.FontProperties()
    fontP.set_size('x-small')
    ax1.legend(loc='best', prop=fontP)
    ax1.xaxis.set_major_formatter(matplotlib.dates.DateFormatter('%H:%M'))
    ax1.yaxis.set_major_formatter(matplotlib.ticker.FormatStrFormatter('%d'))
    ax1.grid(True)
    return fig

def plot_mcu_frequency(data, mcu):
    all_keys = {}
    for d in data:
        all_keys.update(d)
    graph_keys = { key: ([], []) for key in all_keys
                   if key in ("freq", "adj") }
    for d in data:
        st = datetime.datetime.utcfromtimestamp(d['#sampletime'])
        for key, (times, values) in graph_keys.items():
            val = d.get(key)
            if val not in (None, '0', '1'):
                times.append(st)
                values.append(float(val))

    # Build plot
    fig, ax1 = matplotlib.pyplot.subplots()
    ax1.set_title("MCU '%s' frequency" % (mcu,))
    ax1.set_xlabel('Time')
    ax1.set_ylabel('Frequency')
    for key in sorted(graph_keys):
        times, values = graph_keys[key]
        ax1.plot_date(times, values, '.', label=key)
    fontP = matplotlib.font_manager.FontProperties()
    fontP.set_size('x-small')
    ax1.legend(loc='best', prop=fontP)
    ax1.xaxis.set_major_formatter(matplotlib.dates.DateFormatter('%H:%M'))
    ax1.yaxis.set_major_formatter(matplotlib.ticker.FormatStrFormatter('%d'))
    ax1.grid(True)
    return fig

def plot_temperature(data, heaters):
    fig, ax1 = matplotlib.pyplot.subplots()
    ax2 = ax1.twinx()
    for heater in heaters.split(','):
        heater = heater.strip()
        temp_key = heater + ':' + 'temp'
        target_key = heater + ':' + 'target'
        pwm_key = heater + ':' + 'pwm'
        times = []
        temps = []
        targets = []
        pwm = []
        for d in data:
            temp = d.get(temp_key)
            if temp is None:
                continue
            times.append(datetime.datetime.utcfromtimestamp(d['#sampletime']))
            temps.append(float(temp))
            pwm.append(float(d.get(pwm_key, 0.)))
            targets.append(float(d.get(target_key, 0.)))
        ax1.plot_date(times, temps, '-', label='%s temp' % (heater,), alpha=0.8)
        if any(targets):
            label = '%s target' % (heater,)
            ax1.plot_date(times, targets, '-', label=label, alpha=0.3)
        if any(pwm):
            label = '%s pwm' % (heater,)
            ax2.plot_date(times, pwm, '-', label=label, alpha=0.2)
    # Build plot
    ax1.set_title("Temperature of %s" % (heaters,))
    ax1.set_xlabel('Time')
    ax1.set_ylabel('Temperature')
    ax2.set_ylabel('pwm')
    fontP = matplotlib.font_manager.FontProperties()
    fontP.set_size('x-small')
    ax1li, ax1la = ax1.get_legend_handles_labels()
    ax2li, ax2la = ax2.get_legend_handles_labels()
    ax1.legend(ax1li + ax2li, ax1la + ax2la, loc='best', prop=fontP)
    ax1.xaxis.set_major_formatter(matplotlib.dates.DateFormatter('%H:%M'))
    ax1.grid(True)
    return fig

def main():
    # Parse command-line arguments
    usage = "%prog [options] <logfile>"
    opts = optparse.OptionParser(usage)
    opts.add_option("-f", "--frequency", action="store_true",
                    help="graph mcu frequency")
    opts.add_option("-s", "--system", action="store_true",
                    help="graph system load")
    opts.add_option("-o", "--output", type="string", dest="output",
                    default=None, help="filename of output graph")
    opts.add_option("-t", "--temperature", type="string", dest="heater",
                    default=None, help="graph heater temperature")
    opts.add_option("-m", "--mcu", type="string", dest="mcu", default=None,
                    help="limit stats to the given mcu")
    options, args = opts.parse_args()
    if len(args) != 1:
        opts.error("Incorrect number of arguments")
    logname = args[0]

    # Parse data
    data = parse_log(logname, options.mcu)
    if not data:
        return

    # Draw graph
    setup_matplotlib(options.output is not None)
    if options.heater is not None:
        fig = plot_temperature(data, options.heater)
    elif options.frequency:
        if options.mcu is not None:
            fig = plot_mcu_frequency(data, options.mcu)
        else:
            fig = plot_mcu_frequencies(data)
    elif options.system:
        fig = plot_system(data)
    else:
        fig = plot_mcu(data, MAXBANDWIDTH)

    # Show graph
    if options.output is None:
        matplotlib.pyplot.show()
    else:
        fig.set_size_inches(8, 6)
        fig.savefig(options.output)

if __name__ == '__main__':
    main()