mirror of
https://github.com/ClusterCockpit/cc-specifications.git
synced 2024-12-26 05:19:05 +01:00
Rework schema
* Add partition and topology to cluster * Move accelerator spec from job resources to partitions * Change to consistent camelCase * Fix ref URLS
This commit is contained in:
Jan Eitzinger
parent
be2b977e7a
commit
fc69a29b34
@ -4,43 +4,15 @@
|
|||||||
"description": "Meta data information of a HPC cluster",
|
"description": "Meta data information of a HPC cluster",
|
||||||
"type": "object",
|
"type": "object",
|
||||||
"properties":{
|
"properties":{
|
||||||
"ClusterId": {
|
"name": {
|
||||||
"description": "The unique identifier of a cluster",
|
"description": "The unique identifier of a cluster",
|
||||||
"type": "string"
|
"type": "string"
|
||||||
},
|
},
|
||||||
"ProcessorType": {
|
"metricDataRepository": {
|
||||||
"description": "Processor type",
|
|
||||||
"type": "string"
|
|
||||||
},
|
|
||||||
"SocketsPerNode": {
|
|
||||||
"description": "Number of sockets per node",
|
|
||||||
"type": "integer"
|
|
||||||
},
|
|
||||||
"CoresPerSocket": {
|
|
||||||
"description": "Number of cores per socket",
|
|
||||||
"type": "integer"
|
|
||||||
},
|
|
||||||
"ThreadsPerCore": {
|
|
||||||
"description": "Number of SMT threads per core",
|
|
||||||
"type": "integer"
|
|
||||||
},
|
|
||||||
"FlopRateScalar": {
|
|
||||||
"description": "Theoretical node peak flop rate for scalar code in GFlops/s",
|
|
||||||
"type": "integer"
|
|
||||||
},
|
|
||||||
"FlopRateSimd": {
|
|
||||||
"description": "Theoretical node peak flop rate for SIMD code in GFlops/s",
|
|
||||||
"type": "integer"
|
|
||||||
},
|
|
||||||
"MemoryBandwidth": {
|
|
||||||
"description": "Theoretical node peak memory bandwidth in GB/s",
|
|
||||||
"type": "integer"
|
|
||||||
},
|
|
||||||
"MetricDataRepository": {
|
|
||||||
"description": "Type of the metric data repository for this cluster",
|
"description": "Type of the metric data repository for this cluster",
|
||||||
"type": "object",
|
"type": "object",
|
||||||
"properties": {
|
"properties": {
|
||||||
"Kind": {
|
"kind": {
|
||||||
"type": "string",
|
"type": "string",
|
||||||
"enum": [
|
"enum": [
|
||||||
"influxdb-v1",
|
"influxdb-v1",
|
||||||
@ -49,71 +21,193 @@
|
|||||||
"cc-metric-store"
|
"cc-metric-store"
|
||||||
]
|
]
|
||||||
},
|
},
|
||||||
"Url": {
|
"url": {
|
||||||
"type": "string"
|
"type": "string"
|
||||||
},
|
},
|
||||||
"Token": {
|
"token": {
|
||||||
"type": "string"
|
"type": "string"
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"required": [
|
"required": [
|
||||||
"Kind",
|
"kind",
|
||||||
"Url"
|
"url"
|
||||||
]
|
]
|
||||||
},
|
},
|
||||||
"MetricConfig": {
|
"metricConfig": {
|
||||||
"description": "Metric specifications",
|
"description": "Metric specifications",
|
||||||
"type": "array",
|
"type": "array",
|
||||||
"items": {
|
"items": {
|
||||||
"type": "object",
|
"type": "object",
|
||||||
"properties":{
|
"properties":{
|
||||||
"Name": {
|
"name": {
|
||||||
"description": "Metric name",
|
"description": "Metric name",
|
||||||
"type": "string"
|
"type": "string"
|
||||||
},
|
},
|
||||||
"Unit": {
|
"unit": {
|
||||||
"description": "Metric unit",
|
"description": "Metric unit",
|
||||||
"type": "string"
|
"type": "string"
|
||||||
},
|
},
|
||||||
"Timestep": {
|
"timestep": {
|
||||||
"description": "Frequency of timeseries points",
|
"description": "Frequency of timeseries points",
|
||||||
"type": "integer"
|
"type": "integer"
|
||||||
},
|
},
|
||||||
"Peak": {
|
"peak": {
|
||||||
"type": "number"
|
"type": "number"
|
||||||
},
|
},
|
||||||
"Normal": {
|
"normal": {
|
||||||
"type": "number"
|
"type": "number"
|
||||||
},
|
},
|
||||||
"Caution": {
|
"caution": {
|
||||||
"type": "number"
|
"type": "number"
|
||||||
},
|
},
|
||||||
"Alert": {
|
"alert": {
|
||||||
"type": "number"
|
"type": "number"
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"required":[
|
"partitions": {
|
||||||
"Name",
|
"description": "Array of cluster partitions",
|
||||||
"Unit",
|
"type": "array",
|
||||||
"Timestep",
|
"items": {
|
||||||
"Peak",
|
"type": "object",
|
||||||
"Normal",
|
"properties":{
|
||||||
"Caution",
|
"name": {
|
||||||
"Alert"
|
"description": "Metric name",
|
||||||
]
|
"type": "string"
|
||||||
}
|
},
|
||||||
},
|
"processorType": {
|
||||||
"required":[
|
"description": "Processor type",
|
||||||
"ClusterId",
|
"type": "string"
|
||||||
"ProcessorType",
|
},
|
||||||
"SocketsPerNode",
|
"socketsPerNode": {
|
||||||
"CoresPerSocket",
|
"description": "Number of sockets per node",
|
||||||
"ThreadsPerCore",
|
"type": "integer"
|
||||||
"FlopRateScalar",
|
},
|
||||||
"FlopRateSimd",
|
"coresPerSocket": {
|
||||||
"MemoryBandwidth",
|
"description": "Number of cores per socket",
|
||||||
"MetricDataRepository",
|
"type": "integer"
|
||||||
"MetricConfig"
|
},
|
||||||
]
|
"threadsPerCore": {
|
||||||
}
|
"description": "Number of SMT threads per core",
|
||||||
|
"type": "integer"
|
||||||
|
},
|
||||||
|
"flopRateScalar": {
|
||||||
|
"description": "Theoretical node peak flop rate for scalar code in GFlops/s",
|
||||||
|
"type": "integer"
|
||||||
|
},
|
||||||
|
"flopRateSimd": {
|
||||||
|
"description": "Theoretical node peak flop rate for SIMD code in GFlops/s",
|
||||||
|
"type": "integer"
|
||||||
|
},
|
||||||
|
"memoryBandwidth": {
|
||||||
|
"description": "Theoretical node peak memory bandwidth in GB/s",
|
||||||
|
"type": "integer"
|
||||||
|
},
|
||||||
|
"topology": {
|
||||||
|
"description": "Node topology",
|
||||||
|
"type": "object",
|
||||||
|
"properties":{
|
||||||
|
"node": {
|
||||||
|
"description": "HwTread lists of node",
|
||||||
|
"type": "array",
|
||||||
|
"items": {
|
||||||
|
"type": "integer"
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"socket": {
|
||||||
|
"description": "HwTread lists of sockets",
|
||||||
|
"type": "array",
|
||||||
|
"items": {
|
||||||
|
"type": "array",
|
||||||
|
"items": {
|
||||||
|
"type": "integer"
|
||||||
|
}
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"memoryDomain": {
|
||||||
|
"description": "HwTread lists of memory domains",
|
||||||
|
"type": "array",
|
||||||
|
"items": {
|
||||||
|
"type": "array",
|
||||||
|
"items": {
|
||||||
|
"type": "integer"
|
||||||
|
}
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"die": {
|
||||||
|
"description": "HwTread lists of dies",
|
||||||
|
"type": "array",
|
||||||
|
"items": {
|
||||||
|
"type": "array",
|
||||||
|
"items": {
|
||||||
|
"type": "integer"
|
||||||
|
}
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"core": {
|
||||||
|
"description": "HwTread lists of cores",
|
||||||
|
"type": "array",
|
||||||
|
"items": {
|
||||||
|
"type": "array",
|
||||||
|
"items": {
|
||||||
|
"type": "integer"
|
||||||
|
}
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"accelerators": {
|
||||||
|
"type": "array",
|
||||||
|
"description": "List of of accelerator devices",
|
||||||
|
"items": {
|
||||||
|
"type": "object",
|
||||||
|
"properties": {
|
||||||
|
"id": {
|
||||||
|
"type": "string",
|
||||||
|
"description": "The unique device id"
|
||||||
|
},
|
||||||
|
"type": {
|
||||||
|
"type": "string",
|
||||||
|
"description": "The accelerator type",
|
||||||
|
"enum": [
|
||||||
|
"Nvidia GPU",
|
||||||
|
"AMD GPU",
|
||||||
|
"Intel GPU"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"model": {
|
||||||
|
"type": "string",
|
||||||
|
"description": "The accelerator model"
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"required": [
|
||||||
|
"id",
|
||||||
|
"type",
|
||||||
|
"model"
|
||||||
|
]
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"required":[
|
||||||
|
"node",
|
||||||
|
"socket",
|
||||||
|
"memoryDomain"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"required":[
|
||||||
|
"name",
|
||||||
|
"topology",
|
||||||
|
"processorType",
|
||||||
|
"socketsPerNode",
|
||||||
|
"coresPerSocket",
|
||||||
|
"threadsPerCore",
|
||||||
|
"flopRateScalar",
|
||||||
|
"flopRateSimd",
|
||||||
|
"memoryBandwidth"
|
||||||
|
]
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"required":[
|
||||||
|
"name",
|
||||||
|
"metricDataRepository",
|
||||||
|
"metricConfig",
|
||||||
|
"partitions"
|
||||||
|
]
|
||||||
|
}
|
||||||
|
|||||||
@ -4,7 +4,7 @@
|
|||||||
"description": "Meta data information of a HPC job",
|
"description": "Meta data information of a HPC job",
|
||||||
"type": "object",
|
"type": "object",
|
||||||
"properties": {
|
"properties": {
|
||||||
"job_id": {
|
"jobId": {
|
||||||
"description": "The unique identifier of a job",
|
"description": "The unique identifier of a job",
|
||||||
"type": "integer"
|
"type": "integer"
|
||||||
},
|
},
|
||||||
@ -21,24 +21,24 @@
|
|||||||
"type": "string"
|
"type": "string"
|
||||||
},
|
},
|
||||||
"partition": {
|
"partition": {
|
||||||
"description": "The queue to which the job was submitted",
|
"description": "The cluster partition to which the job was submitted",
|
||||||
"type": "string"
|
"type": "string"
|
||||||
},
|
},
|
||||||
"array_job_id": {
|
"arrayJobId": {
|
||||||
"description": "The unique identifier of an array job",
|
"description": "The unique identifier of an array job",
|
||||||
"type": "integer"
|
"type": "integer"
|
||||||
},
|
},
|
||||||
"num_nodes": {
|
"numNodes": {
|
||||||
"description": "Number of nodes used",
|
"description": "Number of nodes used",
|
||||||
"type": "integer",
|
"type": "integer",
|
||||||
"exclusiveMinimum": 0
|
"exclusiveMinimum": 0
|
||||||
},
|
},
|
||||||
"num_hwthreads": {
|
"numHwthreads": {
|
||||||
"description": "Number of HWThreads used",
|
"description": "Number of HWThreads used",
|
||||||
"type": "integer",
|
"type": "integer",
|
||||||
"exclusiveMinimum": 0
|
"exclusiveMinimum": 0
|
||||||
},
|
},
|
||||||
"num_acc": {
|
"numAcc": {
|
||||||
"description": "Number of accelerators used",
|
"description": "Number of accelerators used",
|
||||||
"type": "integer",
|
"type": "integer",
|
||||||
"exclusiveMinimum": 0
|
"exclusiveMinimum": 0
|
||||||
@ -47,7 +47,7 @@
|
|||||||
"description": "Job uses only exclusive nodes",
|
"description": "Job uses only exclusive nodes",
|
||||||
"type": "integer"
|
"type": "integer"
|
||||||
},
|
},
|
||||||
"monitoring_status": {
|
"monitoringStatus": {
|
||||||
"description": "State of monitoring system during job run",
|
"description": "State of monitoring system during job run",
|
||||||
"type": "integer"
|
"type": "integer"
|
||||||
},
|
},
|
||||||
@ -60,7 +60,7 @@
|
|||||||
"type": "integer",
|
"type": "integer",
|
||||||
"exclusiveMinimum": 0
|
"exclusiveMinimum": 0
|
||||||
},
|
},
|
||||||
"job_state": {
|
"jobState": {
|
||||||
"description": "Final state of job",
|
"description": "Final state of job",
|
||||||
"type": "string",
|
"type": "string",
|
||||||
"enum": [
|
"enum": [
|
||||||
@ -71,12 +71,12 @@
|
|||||||
"timeout"
|
"timeout"
|
||||||
]
|
]
|
||||||
},
|
},
|
||||||
"start_time": {
|
"startTime": {
|
||||||
"description": "Start epoch time stamp in seconds",
|
"description": "Start epoch time stamp in seconds",
|
||||||
"type": "integer",
|
"type": "integer",
|
||||||
"exclusiveMinimum": 0
|
"exclusiveMinimum": 0
|
||||||
},
|
},
|
||||||
"stop_time": {
|
"stopTime": {
|
||||||
"description": "Stop epoch time stamp in seconds",
|
"description": "Stop epoch time stamp in seconds",
|
||||||
"type": "integer",
|
"type": "integer",
|
||||||
"exclusiveMinimum": 0
|
"exclusiveMinimum": 0
|
||||||
@ -104,33 +104,9 @@
|
|||||||
},
|
},
|
||||||
"accelerators": {
|
"accelerators": {
|
||||||
"type": "array",
|
"type": "array",
|
||||||
"description": "List of of accelerator devices",
|
"description": "List of of accelerator ids",
|
||||||
"items": {
|
"items": {
|
||||||
"type": "object",
|
"type": "integer"
|
||||||
"properties": {
|
|
||||||
"id": {
|
|
||||||
"type": "string",
|
|
||||||
"description": "The unique device id"
|
|
||||||
},
|
|
||||||
"type": {
|
|
||||||
"type": "string",
|
|
||||||
"description": "The accelerator type",
|
|
||||||
"enum": [
|
|
||||||
"Nvidia GPU",
|
|
||||||
"AMD GPU",
|
|
||||||
"Intel GPU"
|
|
||||||
]
|
|
||||||
},
|
|
||||||
"model": {
|
|
||||||
"type": "string",
|
|
||||||
"description": "The accelerator model"
|
|
||||||
}
|
|
||||||
},
|
|
||||||
"required": [
|
|
||||||
"id",
|
|
||||||
"type",
|
|
||||||
"model"
|
|
||||||
]
|
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"configuration": {
|
"configuration": {
|
||||||
@ -144,11 +120,11 @@
|
|||||||
"minItems": 1
|
"minItems": 1
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"meta_data": {
|
"metaData": {
|
||||||
"description": "Additional information about the job",
|
"description": "Additional information about the job",
|
||||||
"type": "object",
|
"type": "object",
|
||||||
"properties": {
|
"properties": {
|
||||||
"job_script": {
|
"jobScript": {
|
||||||
"type": "string",
|
"type": "string",
|
||||||
"description": "The batch script of the job"
|
"description": "The batch script of the job"
|
||||||
},
|
},
|
||||||
@ -184,147 +160,147 @@
|
|||||||
"properties": {
|
"properties": {
|
||||||
"mem_used": {
|
"mem_used": {
|
||||||
"description": "Memory capacity used (required)",
|
"description": "Memory capacity used (required)",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"cpu_load": {
|
"cpu_load": {
|
||||||
"description": "CPU requested core utilization (load 1m) (required)",
|
"description": "CPU requested core utilization (load 1m) (required)",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"flops_any": {
|
"flops_any": {
|
||||||
"description": "Total flop rate with DP flops scaled up (required)",
|
"description": "Total flop rate with DP flops scaled up (required)",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"mem_bw": {
|
"mem_bw": {
|
||||||
"description": "Main memory bandwidth (required)",
|
"description": "Main memory bandwidth (required)",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"net_bw": {
|
"net_bw": {
|
||||||
"description": "Total fast interconnect network bandwidth (required)",
|
"description": "Total fast interconnect network bandwidth (required)",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"file_bw": {
|
"file_bw": {
|
||||||
"description": "Total file IO bandwidth (required)",
|
"description": "Total file IO bandwidth (required)",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"ipc": {
|
"ipc": {
|
||||||
"description": "Instructions executed per cycle",
|
"description": "Instructions executed per cycle",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"cpu_used": {
|
"cpu_used": {
|
||||||
"description": "CPU active core utilization",
|
"description": "CPU active core utilization",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"flops_dp": {
|
"flops_dp": {
|
||||||
"description": "Double precision flop rate",
|
"description": "Double precision flop rate",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"flops_sp": {
|
"flops_sp": {
|
||||||
"description": "Single precision flops rate",
|
"description": "Single precision flops rate",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"rapl_power": {
|
"rapl_power": {
|
||||||
"description": "CPU power consumption",
|
"description": "CPU power consumption",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"gpu_used": {
|
"gpu_used": {
|
||||||
"description": "GPU utilization",
|
"description": "GPU utilization",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"gpu_mem_used": {
|
"gpu_mem_used": {
|
||||||
"description": "GPU memory capacity used",
|
"description": "GPU memory capacity used",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"gpu_power": {
|
"gpu_power": {
|
||||||
"description": "GPU power consumption",
|
"description": "GPU power consumption",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"clock": {
|
"clock": {
|
||||||
"description": "Average core frequency",
|
"description": "Average core frequency",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"eth_read_bw": {
|
"eth_read_bw": {
|
||||||
"description": "Ethernet read bandwidth",
|
"description": "Ethernet read bandwidth",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"eth_write_bw": {
|
"eth_write_bw": {
|
||||||
"description": "Ethernet write bandwidth",
|
"description": "Ethernet write bandwidth",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_read_bw": {
|
"pfs_read_bw": {
|
||||||
"description": "Parallel file system read bandwidth",
|
"description": "Parallel file system read bandwidth",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_write_bw": {
|
"pfs_write_bw": {
|
||||||
"description": "Parallel file system write bandwidth",
|
"description": "Parallel file system write bandwidth",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_read_req": {
|
"pfs_read_req": {
|
||||||
"description": "Parallel file system read requests",
|
"description": "Parallel file system read requests",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_write_req": {
|
"pfs_write_req": {
|
||||||
"description": "Parallel file system write requests",
|
"description": "Parallel file system write requests",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_inodes": {
|
"pfs_inodes": {
|
||||||
"description": "Parallel file system inodes used",
|
"description": "Parallel file system inodes used",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_accesses": {
|
"pfs_accesses": {
|
||||||
"description": "Parallel file system open and close",
|
"description": "Parallel file system open and close",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_fsync": {
|
"pfs_fsync": {
|
||||||
"description": "Parallel file system fsync",
|
"description": "Parallel file system fsync",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_create": {
|
"pfs_create": {
|
||||||
"description": "Parallel file system create",
|
"description": "Parallel file system create",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_open": {
|
"pfs_open": {
|
||||||
"description": "Parallel file system open",
|
"description": "Parallel file system open",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_close": {
|
"pfs_close": {
|
||||||
"description": "Parallel file system close",
|
"description": "Parallel file system close",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"pfs_seek": {
|
"pfs_seek": {
|
||||||
"description": "Parallel file system seek",
|
"description": "Parallel file system seek",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"fs_read_bw": {
|
"fs_read_bw": {
|
||||||
"description": "Local file system read bandwidth",
|
"description": "Local file system read bandwidth",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"fs_write_bw": {
|
"fs_write_bw": {
|
||||||
"description": "Local file system write bandwidth",
|
"description": "Local file system write bandwidth",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"fs_inodes": {
|
"fs_inodes": {
|
||||||
"description": "Local file system inodes used",
|
"description": "Local file system inodes used",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"fs_accesses": {
|
"fs_accesses": {
|
||||||
"description": "Local file system open and close",
|
"description": "Local file system open and close",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"ic_rcv_packets": {
|
"ic_rcv_packets": {
|
||||||
"description": "Network interconnect read packets",
|
"description": "Network interconnect read packets",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"ic_send_packets": {
|
"ic_send_packets": {
|
||||||
"description": "Network interconnect send packet",
|
"description": "Network interconnect send packet",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"ic_read_bw": {
|
"ic_read_bw": {
|
||||||
"description": "Network interconnect read bandwidth",
|
"description": "Network interconnect read bandwidth",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
},
|
},
|
||||||
"ic_write_bw": {
|
"ic_write_bw": {
|
||||||
"description": "Network interconnect write bandwidth",
|
"description": "Network interconnect write bandwidth",
|
||||||
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-statistics.schema.json"
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"required": [
|
"required": [
|
||||||
@ -338,13 +314,13 @@
|
|||||||
}
|
}
|
||||||
},
|
},
|
||||||
"required": [
|
"required": [
|
||||||
"job_id",
|
"jobId",
|
||||||
"user_id",
|
"user",
|
||||||
"project_id",
|
"project",
|
||||||
"cluster_id",
|
"cluster",
|
||||||
"num_nodes",
|
"numNodes",
|
||||||
"start_time",
|
"startTime",
|
||||||
"stop_time",
|
"stopTime",
|
||||||
"duration",
|
"duration",
|
||||||
"resources",
|
"resources",
|
||||||
"tags",
|
"tags",
|
||||||
|
|||||||
@ -4,10 +4,10 @@
|
|||||||
"description": "Metric data of a HPC job",
|
"description": "Metric data of a HPC job",
|
||||||
"type": "object",
|
"type": "object",
|
||||||
"properties": {
|
"properties": {
|
||||||
"Unit": {
|
"unit": {
|
||||||
"type": "string"
|
"#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/unit.schema.json"
|
||||||
},
|
},
|
||||||
"Scope": {
|
"scope": {
|
||||||
"type": "string",
|
"type": "string",
|
||||||
"enum": [
|
"enum": [
|
||||||
"node",
|
"node",
|
||||||
@ -19,62 +19,62 @@
|
|||||||
"accelerator"
|
"accelerator"
|
||||||
]
|
]
|
||||||
},
|
},
|
||||||
"Timestep": {
|
"timestep": {
|
||||||
"description": "Measurement interval in seconds",
|
"description": "Measurement interval in seconds",
|
||||||
"type": "integer"
|
"type": "integer"
|
||||||
},
|
},
|
||||||
"Thresholds": {
|
"thresholds": {
|
||||||
"description": "Metric thresholds for specific system",
|
"description": "Metric thresholds for specific system",
|
||||||
"type": "object",
|
"type": "object",
|
||||||
"properties": {
|
"properties": {
|
||||||
"Peak": {
|
"peak": {
|
||||||
"type": "number"
|
"type": "number"
|
||||||
},
|
},
|
||||||
"Normal": {
|
"normal": {
|
||||||
"type": "number"
|
"type": "number"
|
||||||
},
|
},
|
||||||
"Caution": {
|
"caution": {
|
||||||
"type": "number"
|
"type": "number"
|
||||||
},
|
},
|
||||||
"Alert": {
|
"alert": {
|
||||||
"type": "number"
|
"type": "number"
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"Series": {
|
"series": {
|
||||||
"type": "array",
|
"type": "array",
|
||||||
"items": {
|
"items": {
|
||||||
"type": "object",
|
"type": "object",
|
||||||
"properties": {
|
"properties": {
|
||||||
"Hostname": {
|
"hostname": {
|
||||||
"type": "string"
|
"type": "string"
|
||||||
},
|
},
|
||||||
"Id": {
|
"id": {
|
||||||
"type": "integer"
|
"type": "integer"
|
||||||
},
|
},
|
||||||
"Statistics": {
|
"statistics": {
|
||||||
"type": "object",
|
"type": "object",
|
||||||
"properties": {
|
"properties": {
|
||||||
"Avg": {
|
"avg": {
|
||||||
"description": "Series average",
|
"description": "Series average",
|
||||||
"type": "number",
|
"type": "number",
|
||||||
"minimum": 0
|
"minimum": 0
|
||||||
},
|
},
|
||||||
"Min": {
|
"min": {
|
||||||
"description": "Series minimum",
|
"description": "Series minimum",
|
||||||
"type": "number",
|
"type": "number",
|
||||||
"minimum": 0
|
"minimum": 0
|
||||||
},
|
},
|
||||||
"Max": {
|
"max": {
|
||||||
"description": "Series maximum",
|
"description": "Series maximum",
|
||||||
"type": "number",
|
"type": "number",
|
||||||
"minimum": 0
|
"minimum": 0
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"required": [
|
"required": [
|
||||||
"Avg",
|
"avg",
|
||||||
"Min",
|
"min",
|
||||||
"Max"
|
"max"
|
||||||
]
|
]
|
||||||
},
|
},
|
||||||
"data": {
|
"data": {
|
||||||
@ -87,16 +87,16 @@
|
|||||||
}
|
}
|
||||||
},
|
},
|
||||||
"required": [
|
"required": [
|
||||||
"Hostname",
|
"hostname",
|
||||||
"Data"
|
"data"
|
||||||
]
|
]
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"required": [
|
"required": [
|
||||||
"Unit",
|
"unit",
|
||||||
"Scope",
|
"scope",
|
||||||
"Timestep",
|
"timestep",
|
||||||
"Series"
|
"series"
|
||||||
]
|
]
|
||||||
}
|
}
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user