RepoMirrors/ceph
1
0
Fork 0
mirror of https://github.com/ceph/ceph synced 2025-01-01 08:32:24 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge PR #25806 into master

Browse Source 
* refs/pull/25806/head:
	mgr: diskprediction_cloud Add nvme devcie support

Reviewed-by: Sage Weil <sage@redhat.com>
This commit is contained in:
Sage Weil 2019-01-12 05:13:32 -06:00
commit ed56c86b1c
9 changed files with 102 additions and 96 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/pybind/mgr/diskprediction_cloud/agent/metrics/ceph_cluster.py
View File

@ -54,8 +54,7 @@ class CephClusterAgent(MetricsAgent):
c_data.tags['cluster_id'] = cluster_id
c_data.fields['cluster_health'] = str(cluster_state)
c_data.fields['agenthost'] = socket.gethostname()
c_data.tags['agenthost_domain_id'] = \
'%s_%s' % (cluster_id, c_data.fields['agenthost'])
c_data.tags['agenthost_domain_id'] = cluster_id
c_data.fields['osd_epoch'] = obj_api.get_osd_epoch()
c_data.fields['num_mon'] = len(obj_api.get_mons())
c_data.fields['num_mon_quorum'] = \
@ -86,7 +85,7 @@ class CephClusterAgent(MetricsAgent):
c_data.fields['osd_bytes_avail'] = total_avail_bytes
if total_bytes and total_avail_bytes:
c_data.fields['osd_bytes_used_percentage'] = \
round(float(total_used_bytes) / float(total_bytes) * 100, 4)
round((float(total_used_bytes) / float(total_bytes)) * 100, 4)
else:
c_data.fields['osd_bytes_used_percentage'] = 0.0000

12
src/pybind/mgr/diskprediction_cloud/agent/metrics/ceph_mon_osd.py
View File

@ -117,7 +117,7 @@ class CephMonOsdAgent(MetricsAgent):
e_osd.fields['name'] = n
e_osd.tags['cluster_id'] = cluster_id
e_osd.fields['agenthost'] = socket.gethostname()
e_osd.tags['agenthost_domain_id'] = '%s_%s' % (cluster_id, socket.gethostname())
e_osd.tags['agenthost_domain_id'] = cluster_id
e_osd.tags['host_domain_id'] = '%s_%s' % (cluster_id, socket.gethostname())
for k in n_value.keys():
e_osd.fields[k] = str(n_value[k])
@ -131,7 +131,7 @@ class CephMonOsdAgent(MetricsAgent):
n_node = CephOsdTree()
n_node.tags['cluster_id'] = cluster_id
n_node.fields['agenthost'] = socket.gethostname()
n_node.tags['agenthost_domain_id'] = '%s_%s' % (cluster_id, socket.gethostname())
n_node.tags['agenthost_domain_id'] = cluster_id
n_node.tags['host_domain_id'] = '%s_%s' % (cluster_id, socket.gethostname())
n_node.fields['children'] = ','.join(str(x) for x in node.get('children', []))
n_node.fields['type_id'] = str(node.get('type_id', ''))
@ -153,8 +153,7 @@ class CephMonOsdAgent(MetricsAgent):
d_osd.tags['cluster_id'] = cluster_id
d_osd.tags['osd_id'] = service_name[4:]
d_osd.fields['agenthost'] = socket.gethostname()
d_osd.tags['agenthost_domain_id'] = \
'%s_%s' % (cluster_id, d_osd.fields['agenthost'])
d_osd.tags['agenthost_domain_id'] = cluster_id
d_osd.tags['host_domain_id'] = \
'%s_%s' % (cluster_id,
obj_api.get_osd_hostname(d_osd.tags['osd_id']))
@ -177,7 +176,7 @@ class CephMonOsdAgent(MetricsAgent):
if stat_bytes and stat_bytes_used:
d_osd.fields['stat_bytes_used_percentage'] = \
round(float(stat_bytes_used) / float(stat_bytes) * 100, 4)
round((float(stat_bytes_used) / float(stat_bytes)) * 100, 4)
else:
d_osd.fields['stat_bytes_used_percentage'] = 0.0000
self.data.append(d_osd)
@ -187,8 +186,7 @@ class CephMonOsdAgent(MetricsAgent):
d_mon.tags['cluster_id'] = cluster_id
d_mon.tags['mon_id'] = service_name[4:]
d_mon.fields['agenthost'] = socket.gethostname()
d_mon.tags['agenthost_domain_id'] = \
'%s_%s' % (cluster_id, d_mon.fields['agenthost'])
d_mon.tags['agenthost_domain_id'] = cluster_id
d_mon.fields['num_sessions'] = \
perf_counts.get('mon.num_sessions', {}).get('value', 0)
d_mon.fields['session_add'] = \

3
src/pybind/mgr/diskprediction_cloud/agent/metrics/ceph_pool.py
View File

@ -39,8 +39,7 @@ class CephPoolAgent(MetricsAgent):
d_pool.tags['cluster_id'] = cluster_id
d_pool.tags['pool_id'] = p_id
d_pool.fields['agenthost'] = socket.gethostname()
d_pool.tags['agenthost_domain_id'] = \
'%s_%s' % (cluster_id, d_pool.fields['agenthost'])
d_pool.tags['agenthost_domain_id'] = cluster_id
d_pool.fields['bytes_used'] = \
pool.get('stats', {}).get('bytes_used', 0)
d_pool.fields['max_avail'] = \

30
src/pybind/mgr/diskprediction_cloud/agent/metrics/db_relay.py
View File

@ -501,8 +501,7 @@ class DBRelayAgent(MetricsAgent):
)
cluster_host = socket.gethostname()
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)
@ -521,8 +520,7 @@ class DBRelayAgent(MetricsAgent):
)
cluster_host = socket.gethostname()
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (self._cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = self._cluster_id
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)
@ -542,8 +540,7 @@ class DBRelayAgent(MetricsAgent):
)
cluster_host = socket.gethostname()
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id, data
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)
@ -563,8 +560,7 @@ class DBRelayAgent(MetricsAgent):
)
cluster_host = socket.gethostname()
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)
@ -591,8 +587,7 @@ class DBRelayAgent(MetricsAgent):
d_node,
v)
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)
@ -611,8 +606,7 @@ class DBRelayAgent(MetricsAgent):
'VmHostContainsVmDisk'
)
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)
@ -632,8 +626,7 @@ class DBRelayAgent(MetricsAgent):
'OsdContainsPool'
)
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)
@ -652,8 +645,7 @@ class DBRelayAgent(MetricsAgent):
'PoolContainsRBD'
)
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)
@ -675,8 +667,7 @@ class DBRelayAgent(MetricsAgent):
'MetadataPoolContainsFS' if p_attr == 'metadata_pool' else 'DataPoolContainsFS'
)
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)
@ -691,8 +682,7 @@ class DBRelayAgent(MetricsAgent):
'MDSContainsFS'
)
data.fields['agenthost'] = cluster_host
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['host'] = cluster_host
data.fields['cmd'] = str(cypher_cmd)
self.data.append(data)

3
src/pybind/mgr/diskprediction_cloud/agent/metrics/sai_agent.py
View File

@ -50,8 +50,7 @@ class SAIAgent(MetricsAgent):
cluster_id = obj_api.get_cluster_id()
c_data.fields['cluster_domain_id'] = str(cluster_id)
c_data.fields['agenthost'] = str(socket.gethostname())
c_data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, c_data.fields['agenthost']))
c_data.tags['agenthost_domain_id'] = cluster_id
c_data.fields['heartbeat_interval'] = \
int(obj_api.get_configuration('diskprediction_upload_metrics_interval'))
c_data.fields['host_ip'] = str(mgr_meta.get('addr', '127.0.0.1'))

3
src/pybind/mgr/diskprediction_cloud/agent/metrics/sai_cluster.py
View File

@ -30,7 +30,6 @@ class SAICluserAgent(MetricsAgent):
c_data.tags['domain_id'] = str(cluster_id)
c_data.tags['host_domain_id'] = '%s_%s' % (str(cluster_id), str(socket.gethostname()))
c_data.fields['agenthost'] = str(socket.gethostname())
c_data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, c_data.fields['agenthost']))
c_data.tags['agenthost_domain_id'] = cluster_id
c_data.fields['name'] = 'Ceph mgr plugin'
self.data.append(c_data)

3
src/pybind/mgr/diskprediction_cloud/agent/metrics/sai_disk.py
View File

@ -92,8 +92,7 @@ class SAIDiskAgent(MetricsAgent):
str('%s_%s'
% (cluster_id, osds_meta.get('hostname', 'None')))
d_data.fields['agenthost'] = str(socket.gethostname())
d_data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, d_data.fields['agenthost']))
d_data.tags['agenthost_domain_id'] = cluster_id
serial_number = s_val.get('serial_number')
wwn = s_val.get('wwn', {})
wwpn = ''

130
src/pybind/mgr/diskprediction_cloud/agent/metrics/sai_disk_smart.py
View File

@ -61,62 +61,88 @@ class SAIDiskSmartAgent(MetricsAgent):
str('%s_%s'
% (cluster_id, osds_meta.get('hostname', 'None')))
smart_data.fields['agenthost'] = str(socket.gethostname())
smart_data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, smart_data.fields['agenthost']))
smart_data.tags['agenthost_domain_id'] = cluster_id
# parse attributes
ata_smart = s_val.get('ata_smart_attributes', {})
for attr in ata_smart.get('table', []):
if attr.get('raw', {}).get('string'):
if str(attr.get('raw', {}).get('string', '0')).isdigit():
smart_data.fields['%s_raw' % attr.get('id')] = \
int(attr.get('raw', {}).get('string', '0'))
protocol = s_val.get('device', {}).get('protocol', '')
if str(protocol).lower() == 'nvme':
nvme_info = s_val.get('nvme_smart_health_information_log', {})
smart_data['CriticalWarniing_raw'] = int(nvme_info.get('critical_warning', 0))
smart_data['CurrentDriveTemperature_raw'] = int(nvme_info.get('temperature', 0))
smart_data['AvailableSpare_raw'] = int(nvme_info.get('available_spare', 0))
smart_data['AvailableSpareThreshold_raw'] = int(nvme_info.get('available_spare_threshold', 0))
smart_data['PercentageUsed_raw'] = int(nvme_info.get('percentage_used', 0))
smart_data['DataUnitsRead_raw'] = int(nvme_info.get('data_units_read', 0))
smart_data['DataUnitsRead'] = int(nvme_info.get('data_units_written', 0))
smart_data['HostReadCommands_raw'] = int(nvme_info.get('host_reads', 0))
smart_data['HostWriteCommands_raw'] = int(nvme_info.get('host_writes', 0))
smart_data['ControllerBusyTime_raw'] = int(nvme_info.get('controller_busy_time', 0))
smart_data['PowerCycles_raw'] = int(nvme_info.get('power_cycles', 0))
smart_data['PowerOnHours_raw'] = int(nvme_info.get('power_on_hours', 0))
smart_data['UnsafeShutdowns_raw'] = int(nvme_info.get('unsafe_shutdowns', 0))
smart_data['MediaandDataIntegrityErrors_raw'] = int(nvme_info.get('media_errors', 0))
smart_data['ErrorInformationLogEntries'] = int(nvme_info.get('num_err_log_entries'))
nvme_addition = s_val.get('nvme_smart_health_information_add_log', {})
for k, v in nvme_addition.get("Device stats", {}).items():
if v.get('raw') is None:
continue
if isinstance(v.get('raw'), int):
smart_data[k] = int(v['raw'])
else:
if str(attr.get('raw', {}).get('string', '0')).split(' ')[0].isdigit():
smart_data[k] = str(v.get('raw'))
else:
ata_smart = s_val.get('ata_smart_attributes', {})
for attr in ata_smart.get('table', []):
if attr.get('raw', {}).get('string'):
if str(attr.get('raw', {}).get('string', '0')).isdigit():
smart_data.fields['%s_raw' % attr.get('id')] = \
int(attr.get('raw', {}).get('string', '0').split(' ')[0])
int(attr.get('raw', {}).get('string', '0'))
else:
smart_data.fields['%s_raw' % attr.get('id')] = \
attr.get('raw', {}).get('value', 0)
smart_data.fields['raw_data'] = str(json.dumps(osds_smart[s_date]).replace("\"", "\'"))
if s_val.get('temperature', {}).get('current') is not None:
smart_data.fields['CurrentDriveTemperature_raw'] = \
int(s_val['temperature']['current'])
if s_val.get('temperature', {}).get('drive_trip') is not None:
smart_data.fields['DriveTripTemperature_raw'] = \
int(s_val['temperature']['drive_trip'])
if s_val.get('elements_grown_list') is not None:
smart_data.fields['ElementsInGrownDefectList_raw'] = int(s_val['elements_grown_list'])
if s_val.get('power_on_time', {}).get('hours') is not None:
smart_data.fields['9_raw'] = int(s_val['power_on_time']['hours'])
if s_val.get('scsi_percentage_used_endurance_indicator') is not None:
smart_data.fields['PercentageUsedEnduranceIndicator_raw'] = \
int(s_val['scsi_percentage_used_endurance_indicator'])
if s_val.get('scsi_error_counter_log') is not None:
s_err_counter = s_val['scsi_error_counter_log']
for s_key in s_err_counter.keys():
if s_key.lower() in ['read', 'write']:
for s1_key in s_err_counter[s_key].keys():
if s1_key.lower() == 'errors_corrected_by_eccfast':
smart_data.fields['ErrorsCorrectedbyECCFast%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['errors_corrected_by_eccfast'])
elif s1_key.lower() == 'errors_corrected_by_eccdelayed':
smart_data.fields['ErrorsCorrectedbyECCDelayed%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['errors_corrected_by_eccdelayed'])
elif s1_key.lower() == 'errors_corrected_by_rereads_rewrites':
smart_data.fields['ErrorCorrectedByRereadsRewrites%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['errors_corrected_by_rereads_rewrites'])
elif s1_key.lower() == 'total_errors_corrected':
smart_data.fields['TotalErrorsCorrected%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['total_errors_corrected'])
elif s1_key.lower() == 'correction_algorithm_invocations':
smart_data.fields['CorrectionAlgorithmInvocations%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['correction_algorithm_invocations'])
elif s1_key.lower() == 'gigabytes_processed':
smart_data.fields['GigaBytesProcessed%s_raw' % s_key.capitalize()] = \
float(s_err_counter[s_key]['gigabytes_processed'])
elif s1_key.lower() == 'total_uncorrected_errors':
smart_data.fields['TotalUncorrectedErrors%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['total_uncorrected_errors'])
if str(attr.get('raw', {}).get('string', '0')).split(' ')[0].isdigit():
smart_data.fields['%s_raw' % attr.get('id')] = \
int(attr.get('raw', {}).get('string', '0').split(' ')[0])
else:
smart_data.fields['%s_raw' % attr.get('id')] = \
attr.get('raw', {}).get('value', 0)
smart_data.fields['raw_data'] = str(json.dumps(osds_smart[s_date]).replace("\"", "\'"))
if s_val.get('temperature', {}).get('current') is not None:
smart_data.fields['CurrentDriveTemperature_raw'] = \
int(s_val['temperature']['current'])
if s_val.get('temperature', {}).get('drive_trip') is not None:
smart_data.fields['DriveTripTemperature_raw'] = \
int(s_val['temperature']['drive_trip'])
if s_val.get('elements_grown_list') is not None:
smart_data.fields['ElementsInGrownDefectList_raw'] = int(s_val['elements_grown_list'])
if s_val.get('power_on_time', {}).get('hours') is not None:
smart_data.fields['9_raw'] = int(s_val['power_on_time']['hours'])
if s_val.get('scsi_percentage_used_endurance_indicator') is not None:
smart_data.fields['PercentageUsedEnduranceIndicator_raw'] = \
int(s_val['scsi_percentage_used_endurance_indicator'])
if s_val.get('scsi_error_counter_log') is not None:
s_err_counter = s_val['scsi_error_counter_log']
for s_key in s_err_counter.keys():
if s_key.lower() in ['read', 'write']:
for s1_key in s_err_counter[s_key].keys():
if s1_key.lower() == 'errors_corrected_by_eccfast':
smart_data.fields['ErrorsCorrectedbyECCFast%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['errors_corrected_by_eccfast'])
elif s1_key.lower() == 'errors_corrected_by_eccdelayed':
smart_data.fields['ErrorsCorrectedbyECCDelayed%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['errors_corrected_by_eccdelayed'])
elif s1_key.lower() == 'errors_corrected_by_rereads_rewrites':
smart_data.fields['ErrorCorrectedByRereadsRewrites%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['errors_corrected_by_rereads_rewrites'])
elif s1_key.lower() == 'total_errors_corrected':
smart_data.fields['TotalErrorsCorrected%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['total_errors_corrected'])
elif s1_key.lower() == 'correction_algorithm_invocations':
smart_data.fields['CorrectionAlgorithmInvocations%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['correction_algorithm_invocations'])
elif s1_key.lower() == 'gigabytes_processed':
smart_data.fields['GigaBytesProcessed%s_raw' % s_key.capitalize()] = \
float(s_err_counter[s_key]['gigabytes_processed'])
elif s1_key.lower() == 'total_uncorrected_errors':
smart_data.fields['TotalUncorrectedErrors%s_raw' % s_key.capitalize()] = \
int(s_err_counter[s_key]['total_uncorrected_errors'])
serial_number = s_val.get('serial_number')
wwn = s_val.get('wwn', {})

9
src/pybind/mgr/diskprediction_cloud/agent/metrics/sai_host.py
View File

@ -48,8 +48,7 @@ class SAIHostAgent(MetricsAgent):
if osd_host not in hosts:
data = SAIHostFields()
data.fields['agenthost'] = str(socket.gethostname())
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['domain_id'] = \
str('%s_%s' % (cluster_id, osd_host))
data.fields['cluster_domain_id'] = str(cluster_id)
@ -72,8 +71,7 @@ class SAIHostAgent(MetricsAgent):
if mon_host not in hosts:
data = SAIHostFields()
data.fields['agenthost'] = str(socket.gethostname())
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['domain_id'] = \
str('%s_%s' % (cluster_id, mon_host))
data.fields['cluster_domain_id'] = str(cluster_id)
@ -95,8 +93,7 @@ class SAIHostAgent(MetricsAgent):
if mds_host not in hosts:
data = SAIHostFields()
data.fields['agenthost'] = str(socket.gethostname())
data.tags['agenthost_domain_id'] = \
str('%s_%s' % (cluster_id, data.fields['agenthost']))
data.tags['agenthost_domain_id'] = cluster_id
data.tags['domain_id'] = \
str('%s_%s' % (cluster_id, mds_host))
data.fields['cluster_domain_id'] = str(cluster_id)