3pp-mirror/Meshroom
1
0
Fork 0
mirror of https://github.com/alicevision/Meshroom.git synced 2025-05-23 22:16:30 +02:00
Code Issues Projects Releases Packages Wiki Activity

[core] stats: minor simplification

Browse source 
add timeout on nvidia-smi
add try/except
remove duplicated code
no error is no nvidia-smi available
This commit is contained in:
Fabien Castan 2020-10-13 16:07:42 +02:00
parent 88099dd28c
commit cdee25e26e
1 changed files with 26 additions and 36 deletions
Download patch file Download diff file Expand all files Collapse all files

48
meshroom/core/stats.py
View file

@ -44,7 +44,7 @@ class ComputerStatistics:
self.gpuMemoryTotal = 0 self.gpuMemoryTotal = 0
self.gpuName = '' self.gpuName = ''
self.curves = defaultdict(list) self.curves = defaultdict(list)
self.nvidia_smi = None
self._isInit = False self._isInit = False
def initOnFirstTime(self): def initOnFirstTime(self):
@ -53,40 +53,21 @@ class ComputerStatistics:
self._isInit = True self._isInit = True
self.cpuFreq = psutil.cpu_freq().max self.cpuFreq = psutil.cpu_freq().max
self.ramTotal = psutil.virtual_memory().total / 1024/1024/1024 self.ramTotal = psutil.virtual_memory().total / (1024*1024*1024)
if platform.system() == "Windows": if platform.system() == "Windows":
from distutils import spawn from distutils import spawn
# If the platform is Windows and nvidia-smi # If the platform is Windows and nvidia-smi
# could not be found from the environment path,
# try to find it from system drive with default installation path
self.nvidia_smi = spawn.find_executable('nvidia-smi') self.nvidia_smi = spawn.find_executable('nvidia-smi')
if self.nvidia_smi is None: if self.nvidia_smi is None:
self.nvidia_smi = "%s\\Program Files\\NVIDIA Corporation\\NVSMI\\nvidia-smi.exe" % os.environ['systemdrive'] # could not be found from the environment path,
# try to find it from system drive with default installation path
default_nvidia_smi = "%s\\Program Files\\NVIDIA Corporation\\NVSMI\\nvidia-smi.exe" % os.environ['systemdrive']
if os.path.isfile(default_nvidia_smi):
self.nvidia_smi = default_nvidia_smi
else: else:
self.nvidia_smi = "nvidia-smi" self.nvidia_smi = "nvidia-smi"
try:
p = subprocess.Popen([self.nvidia_smi, "-q", "-x"], stdout=subprocess.PIPE)
xmlGpu, stdError = p.communicate()
smiTree = ET.fromstring(xmlGpu)
gpuTree = smiTree.find('gpu')
try:
self.gpuMemoryTotal = gpuTree.find('fb_memory_usage').find('total').text.split(" ")[0]
except Exception as e:
logging.debug('Failed to get gpuMemoryTotal: "{}".'.format(str(e)))
pass
try:
self.gpuName = gpuTree.find('product_name').text
except Exception as e:
logging.debug('Failed to get gpuName: "{}".'.format(str(e)))
pass
except Exception as e:
logging.debug('Failed to get information from nvidia_smi at init: "{}".'.format(str(e)))
def _addKV(self, k, v): def _addKV(self, k, v):
if isinstance(v, tuple): if isinstance(v, tuple):
for ki, vi in v._asdict().items(): for ki, vi in v._asdict().items():
@ -98,6 +79,7 @@ class ComputerStatistics:
self.curves[k].append(v) self.curves[k].append(v)
def update(self): def update(self):
try:
self.initOnFirstTime() self.initOnFirstTime()
self._addKV('cpuUsage', psutil.cpu_percent(percpu=True)) # interval=None => non-blocking (percentage since last call) self._addKV('cpuUsage', psutil.cpu_percent(percpu=True)) # interval=None => non-blocking (percentage since last call)
self._addKV('ramUsage', psutil.virtual_memory().percent) self._addKV('ramUsage', psutil.virtual_memory().percent)
@ -105,11 +87,15 @@ class ComputerStatistics:
self._addKV('vramUsage', 0) self._addKV('vramUsage', 0)
self._addKV('ioCounters', psutil.disk_io_counters()) self._addKV('ioCounters', psutil.disk_io_counters())
self.updateGpu() self.updateGpu()
except Exception as e:
logging.debug('Failed to get statistics: "{}".'.format(str(e)))
def updateGpu(self): def updateGpu(self):
if not self.nvidia_smi:
return
try: try:
p = subprocess.Popen([self.nvidia_smi, "-q", "-x"], stdout=subprocess.PIPE) p = subprocess.Popen([self.nvidia_smi, "-q", "-x"], stdout=subprocess.PIPE, stderr=subprocess.PIPE)
xmlGpu, stdError = p.communicate() xmlGpu, stdError = p.communicate(timeout=10) # 10 seconds
smiTree = ET.fromstring(xmlGpu) smiTree = ET.fromstring(xmlGpu)
gpuTree = smiTree.find('gpu') gpuTree = smiTree.find('gpu')
@ -129,7 +115,11 @@ class ComputerStatistics:
except Exception as e: except Exception as e:
logging.debug('Failed to get gpuTemperature: "{}".'.format(str(e))) logging.debug('Failed to get gpuTemperature: "{}".'.format(str(e)))
pass pass
except subprocess.TimeoutExpired as e:
logging.debug('Timeout when retrieving information from nvidia_smi: "{}".'.format(str(e)))
p.kill()
outs, errs = p.communicate()
return
except Exception as e: except Exception as e:
logging.debug('Failed to get information from nvidia_smi: "{}".'.format(str(e))) logging.debug('Failed to get information from nvidia_smi: "{}".'.format(str(e)))
return return