LotusGuild/hwmonDaemon
2
0
Fork 0
Code Issues 3 Pull Requests Actions Packages Projects Releases Wiki Activity

Did not have memory check within class

Browse Source
This commit is contained in:
Jared Vititoe
2024-12-13 18:43:10 -05:00
parent 45179d89a5
commit 586f7df0c7
1 changed files with 43 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

78
hwmonDaemon.py
View File

@ -431,50 +431,50 @@ class SystemHealthMonitor:
bytes_value /= 1024.0
return f"{bytes_value:.1f}Y{suffix}"
def _check_memory_usage(self) -> Dict[str, Any]:
"""
Check for ECC memory errors if ECC memory is present.
"""
memory_health = {
'has_ecc': False,
'ecc_errors': [],
'status': 'OK',
'total_memory': self._convert_bytes(psutil.virtual_memory().total),
'used_memory': self._convert_bytes(psutil.virtual_memory().used),
'memory_percent': psutil.virtual_memory().percent
}
def _check_memory_usage(self) -> Dict[str, Any]:
"""
Check for ECC memory errors if ECC memory is present.
"""
memory_health = {
'has_ecc': False,
'ecc_errors': [],
'status': 'OK',
'total_memory': self._convert_bytes(psutil.virtual_memory().total),
'used_memory': self._convert_bytes(psutil.virtual_memory().used),
'memory_percent': psutil.virtual_memory().percent
}
try:
# More robust ECC detection - check both directory existence and content
edac_path = '/sys/devices/system/edac/mc'
if os.path.exists(edac_path) and os.listdir(edac_path):
# Verify actual ECC controller presence
for mc_dir in glob.glob('/sys/devices/system/edac/mc/mc[0-9]*'):
if os.path.exists(f"{mc_dir}/csrow0"):
memory_health['has_ecc'] = True
try:
# More robust ECC detection - check both directory existence and content
edac_path = '/sys/devices/system/edac/mc'
if os.path.exists(edac_path) and os.listdir(edac_path):
# Verify actual ECC controller presence
for mc_dir in glob.glob('/sys/devices/system/edac/mc/mc[0-9]*'):
if os.path.exists(f"{mc_dir}/csrow0"):
memory_health['has_ecc'] = True
# Check uncorrectable errors
ue_count = self._read_ecc_count(f"{mc_dir}/csrow0/ue_count")
if ue_count > 0:
memory_health['status'] = 'CRITICAL'
memory_health['ecc_errors'].append(
f"Uncorrectable ECC errors detected in {os.path.basename(mc_dir)}: {ue_count}"
)
# Check uncorrectable errors
ue_count = self._read_ecc_count(f"{mc_dir}/csrow0/ue_count")
if ue_count > 0:
memory_health['status'] = 'CRITICAL'
memory_health['ecc_errors'].append(
f"Uncorrectable ECC errors detected in {os.path.basename(mc_dir)}: {ue_count}"
)
# Check correctable errors
ce_count = self._read_ecc_count(f"{mc_dir}/csrow0/ce_count")
if ce_count > 0:
if memory_health['status'] != 'CRITICAL':
memory_health['status'] = 'WARNING'
memory_health['ecc_errors'].append(
f"Correctable ECC errors detected in {os.path.basename(mc_dir)}: {ce_count}"
)
# Check correctable errors
ce_count = self._read_ecc_count(f"{mc_dir}/csrow0/ce_count")
if ce_count > 0:
if memory_health['status'] != 'CRITICAL':
memory_health['status'] = 'WARNING'
memory_health['ecc_errors'].append(
f"Correctable ECC errors detected in {os.path.basename(mc_dir)}: {ce_count}"
)
except Exception as e:
memory_health['status'] = 'ERROR'
memory_health['ecc_errors'].append(f"Error checking ECC status: {str(e)}")
except Exception as e:
memory_health['status'] = 'ERROR'
memory_health['ecc_errors'].append(f"Error checking ECC status: {str(e)}")
return memory_health
return memory_health
def _read_ecc_count(self, filepath: str) -> int:
"""