| 1 |
#!/usr/bin/env python3 |
| 2 |
|
| 3 |
# Generate a heap visualization for SGen from the heap dump written by |
| 4 |
# mono if the MONO_GC_DEBUG is set to something like |
| 5 |
# "heap-dump=/path/to/file". This script accepts the file as stdin |
| 6 |
# and generates HTML and PNG files. |
| 7 |
|
| 8 |
from __future__ import print_function |
| 9 |
import sys, os |
| 10 |
import Image, ImageDraw |
| 11 |
from xml.sax import ContentHandler, make_parser |
| 12 |
from xml.sax.handler import feature_namespaces |
| 13 |
from optparse import OptionParser |
| 14 |
|
| 15 |
chunk_size = 1024 # number of bytes in a chunk |
| 16 |
chunk_pixel_size = 2 # a chunk is a square with this side length |
| 17 |
large_sections = False |
| 18 |
|
| 19 |
def mark_chunk (img_draw, i, color, section_width): |
| 20 |
row = i / section_width |
| 21 |
col = i % section_width |
| 22 |
pixel_col = col * chunk_pixel_size |
| 23 |
pixel_row = row * chunk_pixel_size |
| 24 |
img_draw.rectangle ([(pixel_col, pixel_row), (pixel_col + chunk_pixel_size - 1, pixel_row + chunk_pixel_size - 1)], fill = color) |
| 25 |
|
| 26 |
class Range: |
| 27 |
pass |
| 28 |
|
| 29 |
class OccupiedRange (Range): |
| 30 |
def __init__ (self, offset, size): |
| 31 |
self.offset = offset |
| 32 |
self.size = size |
| 33 |
|
| 34 |
def mark (self, img_draw, color, section_width): |
| 35 |
start = self.offset / chunk_size |
| 36 |
end = (self.offset + self.size - 1) / chunk_size |
| 37 |
for i in range (start, end + 1): |
| 38 |
mark_chunk (img_draw, i, color, section_width) |
| 39 |
|
| 40 |
class ObjectRange (OccupiedRange): |
| 41 |
def __init__ (self, klass, offset, size): |
| 42 |
OccupiedRange.__init__ (self, offset, size) |
| 43 |
self.klass = klass; |
| 44 |
|
| 45 |
class SectionHandler: |
| 46 |
def __init__ (self, width): |
| 47 |
self.width = width |
| 48 |
self.ranges = [] |
| 49 |
self.size = 0 |
| 50 |
self.used = 0 |
| 51 |
|
| 52 |
def add_object (self, klass, offset, size): |
| 53 |
self.ranges.append (ObjectRange (klass, offset, size)) |
| 54 |
self.used += size |
| 55 |
|
| 56 |
def add_occupied (self, offset, size): |
| 57 |
self.ranges.append (OccupiedRange (offset, size)) |
| 58 |
self.used += size |
| 59 |
|
| 60 |
def draw (self): |
| 61 |
height = (((self.size / chunk_size) + (self.width - 1)) / self.width) * chunk_pixel_size |
| 62 |
|
| 63 |
color_background = (255, 255, 255) |
| 64 |
color_free = (0, 0, 0) |
| 65 |
color_occupied = (0, 255, 0) |
| 66 |
|
| 67 |
img = Image.new ('RGB', (self.width * chunk_pixel_size, height), color_free) |
| 68 |
img_draw = ImageDraw.Draw (img) |
| 69 |
#FIXME: remove filling after end of heap |
| 70 |
|
| 71 |
for r in self.ranges: |
| 72 |
r.mark (img_draw, color_occupied, self.width) |
| 73 |
|
| 74 |
return img |
| 75 |
|
| 76 |
def emit (self, collection_file, collection_kind, collection_num, section_num): |
| 77 |
print ('<h2>%s</h2>' % self.header (), file = collection_file) |
| 78 |
print ('<p>Size %d kB - ' % (self.size / 1024), file = collection_file) |
| 79 |
print ('used %d kB</p>' % (self.used / 1024), file = collection_file) |
| 80 |
|
| 81 |
filename = '%s_%d_%d.png' % (collection_kind, collection_num, section_num) |
| 82 |
print ('<p><img src="%s"></img></p>' % filename, file = collection_file) |
| 83 |
img = self.draw () |
| 84 |
img.save (filename) |
| 85 |
|
| 86 |
class SmallSectionHandler (SectionHandler): |
| 87 |
def __init__ (self): |
| 88 |
SectionHandler.__init__ (self, -1) |
| 89 |
self.offset = 0 |
| 90 |
|
| 91 |
def start_section (self, kind, size): |
| 92 |
assert kind == 'old' |
| 93 |
if self.width <= 0: |
| 94 |
self.width = (size + chunk_size - 1) / chunk_size |
| 95 |
if self.width < 128: |
| 96 |
self.width = 512 |
| 97 |
self.current_section_size = size |
| 98 |
else: |
| 99 |
self.current_section_size = self.width * chunk_size |
| 100 |
self.size += size |
| 101 |
|
| 102 |
def add_object (self, klass, offset, size): |
| 103 |
SectionHandler.add_object (self, klass, self.offset + offset, size) |
| 104 |
|
| 105 |
def add_occupied (self, offset, size): |
| 106 |
SectionHandler.add_occupied (self, self.offset + offset, size) |
| 107 |
|
| 108 |
def end_section (self): |
| 109 |
self.offset += self.current_section_size |
| 110 |
|
| 111 |
def header (self): |
| 112 |
return 'old sections' |
| 113 |
|
| 114 |
class LargeSectionHandler (SectionHandler): |
| 115 |
def __init__ (self): |
| 116 |
SectionHandler.__init__ (self, 512) |
| 117 |
|
| 118 |
def start_section (self, kind, size): |
| 119 |
self.kind = kind |
| 120 |
self.ranges = [] |
| 121 |
self.size = size |
| 122 |
self.used = 0 |
| 123 |
|
| 124 |
def end_section (self): |
| 125 |
pass |
| 126 |
|
| 127 |
def header (self): |
| 128 |
return self.kind + ' section' |
| 129 |
|
| 130 |
class DocHandler (ContentHandler): |
| 131 |
def start (self): |
| 132 |
self.collection_index = 0 |
| 133 |
self.index_file = open ('index.html', 'w') |
| 134 |
print ('<html><body>', file = self.index_file) |
| 135 |
|
| 136 |
def end (self): |
| 137 |
print ('</body></html>', file = self.index_file) |
| 138 |
self.index_file.close () |
| 139 |
|
| 140 |
def startElement (self, name, attrs): |
| 141 |
if name == 'collection': |
| 142 |
self.collection_kind = attrs.get('type', None) |
| 143 |
self.collection_num = int(attrs.get('num', None)) |
| 144 |
reason = attrs.get('reason', None) |
| 145 |
if reason: |
| 146 |
reason = ' (%s)' % reason |
| 147 |
else: |
| 148 |
reason = '' |
| 149 |
self.section_num = 0 |
| 150 |
filename = 'collection_%d.html' % self.collection_index |
| 151 |
print ('<a href="%s">%s%s collection %d</a>' % (filename, self.collection_kind, reason, self.collection_num), file = self.index_file) |
| 152 |
self.collection_file = open (filename, 'w') |
| 153 |
print ('<html><body>', file = self.collection_file) |
| 154 |
print ('<p><a href="collection_%d.html">Prev</a> <a href="collection_%d.html">Next</a> <a href="index.html">Index</a></p>' % (self.collection_index - 1, self.collection_index + 1), file = self.collection_file) |
| 155 |
print ('<h1>%s collection %d</h1>' % (self.collection_kind, self.collection_num), file = self.collection_file) |
| 156 |
self.usage = {} |
| 157 |
self.los_usage = {} |
| 158 |
self.pinned_usage = {} |
| 159 |
self.occupancies = {} |
| 160 |
self.in_los = False |
| 161 |
self.in_pinned = False |
| 162 |
self.heap_used = 0 |
| 163 |
self.heap_size = 0 |
| 164 |
self.los_size = 0 |
| 165 |
if large_sections: |
| 166 |
self.section_handler = LargeSectionHandler () |
| 167 |
else: |
| 168 |
self.section_handler = self.small_section_handler = SmallSectionHandler () |
| 169 |
elif name == 'pinned': |
| 170 |
kind = attrs.get('type', None) |
| 171 |
bytes = int(attrs.get('bytes', None)) |
| 172 |
print ('Pinned from %s: %d kB<br>' % (kind, bytes / 1024), file = self.collection_file) |
| 173 |
elif name == 'occupancy': |
| 174 |
size = int (attrs.get ('size', None)) |
| 175 |
available = int (attrs.get ('available', None)) |
| 176 |
used = int (attrs.get ('used', None)) |
| 177 |
unused = available - used |
| 178 |
|
| 179 |
print ('Occupancy of %d byte slots: %d / %d (%d kB / %d%% wasted)<br>' % (size, used, available, unused * size / 1024, unused * 100 / available), file = self.collection_file) |
| 180 |
elif name == 'section': |
| 181 |
kind = attrs.get('type', None) |
| 182 |
size = int(attrs.get('size', None)) |
| 183 |
|
| 184 |
self.heap_size += size |
| 185 |
|
| 186 |
if not large_sections: |
| 187 |
if kind == 'nursery': |
| 188 |
self.section_handler = LargeSectionHandler () |
| 189 |
else: |
| 190 |
self.section_handler = self.small_section_handler |
| 191 |
|
| 192 |
self.section_handler.start_section (kind, size) |
| 193 |
elif name == 'object': |
| 194 |
klass = attrs.get('class', None) |
| 195 |
size = int(attrs.get('size', None)) |
| 196 |
|
| 197 |
if self.in_los: |
| 198 |
usage_dict = self.los_usage |
| 199 |
self.los_size += size |
| 200 |
elif self.in_pinned: |
| 201 |
location = attrs.get('location', None) |
| 202 |
if location not in self.pinned_usage: |
| 203 |
self.pinned_usage[location] = {} |
| 204 |
usage_dict = self.pinned_usage[location] |
| 205 |
else: |
| 206 |
usage_dict = self.usage |
| 207 |
offset = int(attrs.get('offset', None)) |
| 208 |
|
| 209 |
self.section_handler.add_object (klass, offset, size) |
| 210 |
self.heap_used += size |
| 211 |
if not (klass in usage_dict): |
| 212 |
usage_dict [klass] = (0, 0) |
| 213 |
usage = usage_dict [klass] |
| 214 |
usage_dict [klass] = (usage [0] + 1, usage [1] + size) |
| 215 |
elif name == 'occupied': |
| 216 |
offset = int(attrs.get('offset', None)) |
| 217 |
size = int(attrs.get('size', None)) |
| 218 |
|
| 219 |
self.section_handler.add_occupied (offset, size) |
| 220 |
self.heap_used += size |
| 221 |
elif name == 'los': |
| 222 |
self.in_los = True |
| 223 |
elif name == 'pinned-objects': |
| 224 |
self.in_pinned = True |
| 225 |
|
| 226 |
def dump_usage (self, usage_dict, limit): |
| 227 |
klasses = sorted (usage_dict.keys (), lambda x, y: usage_dict [y][1] - usage_dict [x][1]) |
| 228 |
if limit: |
| 229 |
klasses = klasses [0:limit] |
| 230 |
for klass in klasses: |
| 231 |
usage = usage_dict [klass] |
| 232 |
if usage [1] < 100000: |
| 233 |
print ('%s %d bytes' % (klass, usage [1]), file = self.collection_file) |
| 234 |
else: |
| 235 |
print ('%s %d kB' % (klass, usage [1] / 1024), file = self.collection_file) |
| 236 |
print (' (%d)<br>' % usage [0], file = self.collection_file) |
| 237 |
|
| 238 |
def endElement (self, name): |
| 239 |
if name == 'section': |
| 240 |
self.section_handler.end_section () |
| 241 |
|
| 242 |
if large_sections or self.section_handler != self.small_section_handler: |
| 243 |
self.section_handler.emit (self.collection_file, self.collection_kind, self.collection_num, self.section_num) |
| 244 |
self.section_num += 1 |
| 245 |
elif name == 'collection': |
| 246 |
if not large_sections: |
| 247 |
self.small_section_handler.emit (self.collection_file, self.collection_kind, self.collection_num, self.section_num) |
| 248 |
|
| 249 |
self.dump_usage (self.usage, 10) |
| 250 |
print ('<h3>LOS</h3>', file = self.collection_file) |
| 251 |
self.dump_usage (self.los_usage, None) |
| 252 |
print ('<h3>Pinned</h3>', file = self.collection_file) |
| 253 |
for location in sorted (self.pinned_usage.keys ()): |
| 254 |
print ('<h4>%s</h4>' % location, file = self.collection_file) |
| 255 |
self.dump_usage (self.pinned_usage[location], None) |
| 256 |
print ('</body></html>', file = self.collection_file) |
| 257 |
print (' - %d kB / %d kB (%d%%) - %d kB LOS</a><br>' % (self.heap_used / 1024, self.heap_size / 1024, int(100.0 * self.heap_used / self.heap_size), self.los_size / 1024), file = self.index_file) |
| 258 |
self.collection_file.close () |
| 259 |
self.collection_index += 1 |
| 260 |
elif name == 'los': |
| 261 |
self.in_los = False |
| 262 |
elif name == 'pinned-objects': |
| 263 |
self.in_pinned = False |
| 264 |
|
| 265 |
def main (): |
| 266 |
usage = "usage: %prog [options]" |
| 267 |
parser = OptionParser (usage) |
| 268 |
parser.add_option ("-l", "--large-sections", action = "store_true", dest = "large_sections") |
| 269 |
parser.add_option ("-s", "--small-sections", action = "store_false", dest = "large_sections") |
| 270 |
(options, args) = parser.parse_args () |
| 271 |
if options.large_sections: |
| 272 |
large_sections = True |
| 273 |
|
| 274 |
dh = DocHandler () |
| 275 |
parser = make_parser () |
| 276 |
parser.setFeature (feature_namespaces, 0) |
| 277 |
parser.setContentHandler (dh) |
| 278 |
dh.start () |
| 279 |
parser.parse (sys.stdin) |
| 280 |
dh.end () |
| 281 |
|
| 282 |
if __name__ == "__main__": |
| 283 |
main () |