{ "$schema": "http://json-schema.org/draft-07/schema#", "title": "Job meta data", "description": "Meta data information of a HPC job", "type": "object", "properties": { "job_id": { "description": "The unique identifier of a job", "type": "integer" }, "user": { "description": "The unique identifier of a user", "type": "string" }, "project": { "description": "The unique identifier of a project", "type": "string" }, "cluster": { "description": "The unique identifier of a cluster", "type": "string" }, "partition": { "description": "The queue to which the job was submitted", "type": "string" }, "array_job_id": { "description": "The unique identifier of an array job", "type": "integer" }, "num_nodes": { "description": "Number of nodes used", "type": "integer", "exclusiveMinimum": 0 }, "num_hwthreads": { "description": "Number of HWThreads used", "type": "integer", "exclusiveMinimum": 0 }, "num_acc": { "description": "Number of accelerators used", "type": "integer", "exclusiveMinimum": 0 }, "exclusive": { "description": "Job uses only exclusive nodes", "type": "integer" }, "monitoring_status": { "description": "State of monitoring system during job run", "type": "integer" }, "smt": { "description": "SMT threads used by job", "type": "integer" }, "walltime": { "description": "Requested walltime of job in seconds", "type": "integer", "exclusiveMinimum": 0 }, "job_state": { "description": "Final state of job", "type": "string", "enum": [ "completed", "failed", "canceled", "stopped", "timeout" ] }, "start_time": { "description": "Start epoch time stamp in seconds", "type": "integer", "exclusiveMinimum": 0 }, "stop_time": { "description": "Stop epoch time stamp in seconds", "type": "integer", "exclusiveMinimum": 0 }, "duration": { "description": "Duration of job in seconds", "type": "integer", "exclusiveMinimum": 0 }, "resources": { "description": "Resources used by job", "type": "array", "items": { "type": "object", "properties": { "hostname": { "type": "string" }, "hwthreads": { "type": "array", "description": "List of OS processor ids", "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" ] } }, "configuration": { "type": "string", "description": "The configuration options of the node" }, "required": [ "hostname" ] }, "minItems": 1 } }, "meta_data": { "description": "Additional information about the job", "type": "object", "properties": { "job_script": { "type": "string", "description": "The batch script of the job" }, "slurmdata": { "type": "string", "description": "Additional Slurm information" } } }, "tags": { "description": "List of tags", "type": "array", "items": { "type": "object", "properties": { "name": { "type": "string" }, "type": { "type": "string" } }, "required": [ "name", "type" ] }, "uniqueItems": true }, "statistics": { "description": "Job statistic data", "type": "object", "properties": { "mem_used": { "description": "Memory capacity used (required)", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "cpu_load": { "description": "CPU requested core utilization (load 1m) (required)", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "flops_any": { "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" }, "mem_bw": { "description": "Main memory bandwidth (required)", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "net_bw": { "description": "Total fast interconnect network bandwidth (required)", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "file_bw": { "description": "Total file IO bandwidth (required)", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "ipc": { "description": "Instructions executed per cycle", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "cpu_used": { "description": "CPU active core utilization", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "flops_dp": { "description": "Double precision flop rate", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "flops_sp": { "description": "Single precision flops rate", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "rapl_power": { "description": "CPU power consumption", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "gpu_used": { "description": "GPU utilization", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "gpu_mem_used": { "description": "GPU memory capacity used", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "gpu_power": { "description": "GPU power consumption", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "clock": { "description": "Average core frequency", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "eth_read_bw": { "description": "Ethernet read bandwidth", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "eth_write_bw": { "description": "Ethernet write bandwidth", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_read_bw": { "description": "Parallel file system read bandwidth", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_write_bw": { "description": "Parallel file system write bandwidth", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_read_req": { "description": "Parallel file system read requests", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_write_req": { "description": "Parallel file system write requests", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_inodes": { "description": "Parallel file system inodes used", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_accesses": { "description": "Parallel file system open and close", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_fsync": { "description": "Parallel file system fsync", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_create": { "description": "Parallel file system create", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_open": { "description": "Parallel file system open", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_close": { "description": "Parallel file system close", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "pfs_seek": { "description": "Parallel file system seek", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "fs_read_bw": { "description": "Local file system read bandwidth", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "fs_write_bw": { "description": "Local file system write bandwidth", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "fs_inodes": { "description": "Local file system inodes used", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "fs_accesses": { "description": "Local file system open and close", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "ic_rcv_packets": { "description": "Network interconnect read packets", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "ic_send_packets": { "description": "Network interconnect send packet", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "ic_read_bw": { "description": "Network interconnect read bandwidth", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" }, "ic_write_bw": { "description": "Network interconnect write bandwidth", "#ref": "https://github.com/ClusterCockpit/cc-specifications/blob/master/schema/json/job-metric-data.schema.json" } }, "required": [ "mem_used", "cpu_load", "flops_any", "mem_bw", "net_bw", "file_bw" ] } }, "required": [ "job_id", "user_id", "project_id", "cluster_id", "num_nodes", "start_time", "stop_time", "duration", "resources", "tags", "statistics" ] }