#!/usr/bin/env python # # Copyright 2010 the V8 project authors. All rights reserved. # Redistribution and use in source and binary forms, with or without # modification, are permitted provided that the following conditions are # met: # # * Redistributions of source code must retain the above copyright # notice, this list of conditions and the following disclaimer. # * Redistributions in binary form must reproduce the above # copyright notice, this list of conditions and the following # disclaimer in the documentation and/or other materials provided # with the distribution. # * Neither the name of Google Inc. nor the names of its # contributors may be used to endorse or promote products derived # from this software without specific prior written permission. # # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. # # # This is an utility for plotting charts based on GC traces produced by V8 when # run with flags --trace-gc --trace-gc-nvp. Relies on gnuplot for actual # plotting. # # Usage: gc-nvp-trace-processor.py <GC-trace-filename> # from __future__ import with_statement import sys, types, subprocess, math import gc_nvp_common def flatten(l): flat = [] for i in l: flat.extend(i) return flat def gnuplot(script): gnuplot = subprocess.Popen(["gnuplot"], stdin=subprocess.PIPE) gnuplot.stdin.write(script) gnuplot.stdin.close() gnuplot.wait() x1y1 = 'x1y1' x1y2 = 'x1y2' x2y1 = 'x2y1' x2y2 = 'x2y2' class Item(object): def __init__(self, title, field, axis = x1y1, **keywords): self.title = title self.axis = axis self.props = keywords if type(field) is types.ListType: self.field = field else: self.field = [field] def fieldrefs(self): return self.field def to_gnuplot(self, context): args = ['"%s"' % context.datafile, 'using %s' % context.format_fieldref(self.field), 'title "%s"' % self.title, 'axis %s' % self.axis] if 'style' in self.props: args.append('with %s' % self.props['style']) if 'lc' in self.props: args.append('lc rgb "%s"' % self.props['lc']) if 'fs' in self.props: args.append('fs %s' % self.props['fs']) return ' '.join(args) class Plot(object): def __init__(self, *items): self.items = items def fieldrefs(self): return flatten([item.fieldrefs() for item in self.items]) def to_gnuplot(self, ctx): return 'plot ' + ', '.join([item.to_gnuplot(ctx) for item in self.items]) class Set(object): def __init__(self, value): self.value = value def to_gnuplot(self, ctx): return 'set ' + self.value def fieldrefs(self): return [] class Context(object): def __init__(self, datafile, field_to_index): self.datafile = datafile self.field_to_index = field_to_index def format_fieldref(self, fieldref): return ':'.join([str(self.field_to_index[field]) for field in fieldref]) def collect_fields(plot): field_to_index = {} fields = [] def add_field(field): if field not in field_to_index: fields.append(field) field_to_index[field] = len(fields) for field in flatten([item.fieldrefs() for item in plot]): add_field(field) return (fields, field_to_index) def is_y2_used(plot): for subplot in plot: if isinstance(subplot, Plot): for item in subplot.items: if item.axis == x1y2 or item.axis == x2y2: return True return False def get_field(trace_line, field): t = type(field) if t is types.StringType: return trace_line[field] elif t is types.FunctionType: return field(trace_line) def generate_datafile(datafile_name, trace, fields): with open(datafile_name, 'w') as datafile: for line in trace: data_line = [str(get_field(line, field)) for field in fields] datafile.write('\t'.join(data_line)) datafile.write('\n') def generate_script_and_datafile(plot, trace, datafile, output): (fields, field_to_index) = collect_fields(plot) generate_datafile(datafile, trace, fields) script = [ 'set terminal png', 'set output "%s"' % output, 'set autoscale', 'set ytics nomirror', 'set xtics nomirror', 'set key below' ] if is_y2_used(plot): script.append('set autoscale y2') script.append('set y2tics') context = Context(datafile, field_to_index) for item in plot: script.append(item.to_gnuplot(context)) return '\n'.join(script) def plot_all(plots, trace, prefix): charts = [] for plot in plots: outfilename = "%s_%d.png" % (prefix, len(charts)) charts.append(outfilename) script = generate_script_and_datafile(plot, trace, '~datafile', outfilename) print 'Plotting %s...' % outfilename gnuplot(script) return charts def reclaimed_bytes(row): return row['total_size_before'] - row['total_size_after'] def other_scope(r): if r['gc'] == 's': # there is no 'other' scope for scavenging collections. return 0 return r['pause'] - r['mark'] - r['sweep'] - r['external'] def scavenge_scope(r): if r['gc'] == 's': return r['pause'] - r['external'] return 0 def real_mutator(r): return r['mutator'] - r['steps_took'] plots = [ [ Set('style fill solid 0.5 noborder'), Set('style histogram rowstacked'), Set('style data histograms'), Plot(Item('Scavenge', scavenge_scope, lc = 'green'), Item('Marking', 'mark', lc = 'purple'), Item('Sweep', 'sweep', lc = 'blue'), Item('External', 'external', lc = '#489D43'), Item('Other', other_scope, lc = 'grey'), Item('IGC Steps', 'steps_took', lc = '#FF6347')) ], [ Set('style fill solid 0.5 noborder'), Set('style histogram rowstacked'), Set('style data histograms'), Plot(Item('Scavenge', scavenge_scope, lc = 'green'), Item('Marking', 'mark', lc = 'purple'), Item('Sweep', 'sweep', lc = 'blue'), Item('External', 'external', lc = '#489D43'), Item('Other', other_scope, lc = '#ADD8E6'), Item('External', 'external', lc = '#D3D3D3')) ], [ Plot(Item('Mutator', real_mutator, lc = 'black', style = 'lines')) ], [ Set('style histogram rowstacked'), Set('style data histograms'), Plot(Item('Heap Size (before GC)', 'total_size_before', x1y2, fs = 'solid 0.4 noborder', lc = 'green'), Item('Total holes (after GC)', 'holes_size_before', x1y2, fs = 'solid 0.4 noborder', lc = 'red'), Item('GC Time', ['i', 'pause'], style = 'lines', lc = 'red')) ], [ Set('style histogram rowstacked'), Set('style data histograms'), Plot(Item('Heap Size (after GC)', 'total_size_after', x1y2, fs = 'solid 0.4 noborder', lc = 'green'), Item('Total holes (after GC)', 'holes_size_after', x1y2, fs = 'solid 0.4 noborder', lc = 'red'), Item('GC Time', ['i', 'pause'], style = 'lines', lc = 'red')) ], [ Set('style fill solid 0.5 noborder'), Set('style data histograms'), Plot(Item('Allocated', 'allocated'), Item('Reclaimed', reclaimed_bytes), Item('Promoted', 'promoted', style = 'lines', lc = 'black')) ], ] def freduce(f, field, trace, init): return reduce(lambda t,r: f(t, r[field]), trace, init) def calc_total(trace, field): return freduce(lambda t,v: t + long(v), field, trace, long(0)) def calc_max(trace, field): return freduce(lambda t,r: max(t, r), field, trace, 0) def count_nonzero(trace, field): return freduce(lambda t,r: t if r == 0 else t + 1, field, trace, 0) def process_trace(filename): trace = gc_nvp_common.parse_gc_trace(filename) marksweeps = filter(lambda r: r['gc'] == 'ms', trace) scavenges = filter(lambda r: r['gc'] == 's', trace) globalgcs = filter(lambda r: r['gc'] != 's', trace) charts = plot_all(plots, trace, filename) def stats(out, prefix, trace, field): n = len(trace) total = calc_total(trace, field) max = calc_max(trace, field) if n > 0: avg = total / n else: avg = 0 if n > 1: dev = math.sqrt(freduce(lambda t,r: t + (r - avg) ** 2, field, trace, 0) / (n - 1)) else: dev = 0 out.write('<tr><td>%s</td><td>%d</td><td>%d</td>' '<td>%d</td><td>%d [dev %f]</td></tr>' % (prefix, n, total, max, avg, dev)) def HumanReadable(size): suffixes = ['bytes', 'kB', 'MB', 'GB'] power = 1 for i in range(len(suffixes)): if size < power*1024: return "%.1f" % (float(size) / power) + " " + suffixes[i] power *= 1024 def throughput(name, trace): total_live_after = calc_total(trace, 'total_size_after') total_live_before = calc_total(trace, 'total_size_before') total_gc = calc_total(trace, 'pause') if total_gc == 0: return out.write('GC %s Throughput (after): %s / %s ms = %s/ms<br/>' % (name, HumanReadable(total_live_after), total_gc, HumanReadable(total_live_after / total_gc))) out.write('GC %s Throughput (before): %s / %s ms = %s/ms<br/>' % (name, HumanReadable(total_live_before), total_gc, HumanReadable(total_live_before / total_gc))) with open(filename + '.html', 'w') as out: out.write('<html><body>') out.write('<table>') out.write('<tr><td>Phase</td><td>Count</td><td>Time (ms)</td>') out.write('<td>Max</td><td>Avg</td></tr>') stats(out, 'Total in GC', trace, 'pause') stats(out, 'Scavenge', scavenges, 'pause') stats(out, 'MarkSweep', marksweeps, 'pause') stats(out, 'Mark', filter(lambda r: r['mark'] != 0, trace), 'mark') stats(out, 'Sweep', filter(lambda r: r['sweep'] != 0, trace), 'sweep') stats(out, 'External', filter(lambda r: r['external'] != 0, trace), 'external') out.write('</table>') throughput('TOTAL', trace) throughput('MS', marksweeps) throughput('OLDSPACE', globalgcs) out.write('<br/>') for chart in charts: out.write('<img src="%s">' % chart) out.write('</body></html>') print "%s generated." % (filename + '.html') if len(sys.argv) != 2: print "Usage: %s <GC-trace-filename>" % sys.argv[0] sys.exit(1) process_trace(sys.argv[1])