cc-backend/pkg/schema/schemas/cluster.schema.json

{
  "$schema": "http://json-schema.org/draft/2020-12/schema",
  "$id": "embedfs://cluster.schema.json",
  "title": "HPC cluster description",
  "description": "Meta data information of a HPC cluster",
  "type": "object",
  "properties": {
    "name": {
      "description": "The unique identifier of a cluster",
      "type": "string"
    },
    "metricConfig": {
      "description": "Metric specifications",
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "name": {
            "description": "Metric name",
            "type": "string"
          },
          "unit": {
            "description": "Metric unit",
            "$ref": "embedfs://unit.schema.json"
          },
          "scope": {
            "description": "Native measurement resolution",
            "type": "string"
          },
          "timestep": {
            "description": "Frequency of timeseries points",
            "type": "integer"
          },
          "aggregation": {
            "description": "How the metric is aggregated",
            "type": "string",
            "enum": [
              "sum",
              "avg"
            ]
          },
          "footprint": {
            "description": "Is it a footprint metric and what type",
            "type": "string",
            "enum": [
              "avg",
              "max",
              "min"
            ]
          },
          "energy": {
            "description": "Is it used to calculate job energy",
            "type": "string",
            "enum": [
              "power",
              "energy"
            ]
          },
          "lowerIsBetter": {
            "description": "Is lower better.",
            "type": "boolean"
          },
          "peak": {
            "description": "Metric peak threshold (Upper metric limit)",
            "type": "number"
          },
          "normal": {
            "description": "Metric normal threshold",
            "type": "number"
          },
          "caution": {
            "description": "Metric caution threshold (Suspicious but does not require immediate action)",
            "type": "number"
          },
          "alert": {
            "description": "Metric alert threshold (Requires immediate action)",
            "type": "number"
          },
          "subClusters": {
            "description": "Array of cluster hardware partition metric thresholds",
            "type": "array",
            "items": {
              "type": "object",
              "properties": {
                "name": {
                  "description": "Hardware partition name",
                  "type": "string"
                },
                "footprint": {
                  "description": "Is it a footprint metric and what type. Overwrite global setting",
                  "type": "string",
                  "enum": [
                    "avg",
                    "max",
                    "min"
                  ]
                },
                "energy": {
                  "description": "Is it used to calculate job energy. Overwrite global",
                  "type": "string",
                  "enum": [
                    "power",
                    "energy"
                  ]
                },
                "lowerIsBetter": {
                  "description": "Is lower better. Overwrite global",
                  "type": "boolean"
                },
                "peak": {
                  "type": "number"
                },
                "normal": {
                  "type": "number"
                },
                "caution": {
                  "type": "number"
                },
                "alert": {
                  "type": "number"
                },
                "remove": {
                  "description": "Remove this metric for this subcluster",
                  "type": "boolean"
                }
              },
              "required": [
                "name"
              ]
            }
          }
        },
        "required": [
          "name",
          "unit",
          "scope",
          "timestep",
          "aggregation",
          "peak",
          "normal",
          "caution",
          "alert"
        ]
      },
      "minItems": 1
    },
    "subClusters": {
      "description": "Array of cluster hardware partitions",
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "name": {
            "description": "Hardware partition name",
            "type": "string"
          },
          "processorType": {
            "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": "object",
            "properties": {
              "unit": {
                "description": "Metric unit",
                "$ref": "embedfs://unit.schema.json"
              },
              "value": {
                "type": "number"
              }
            }
          },
          "flopRateSimd": {
            "description": "Theoretical node peak flop rate for SIMD code in GFlops/s",
            "type": "object",
            "properties": {
              "unit": {
                "description": "Metric unit",
                "$ref": "embedfs://unit.schema.json"
              },
              "value": {
                "type": "number"
              }
            }
          },
          "memoryBandwidth": {
            "description": "Theoretical node peak memory bandwidth in GB/s",
            "type": "object",
            "properties": {
              "unit": {
                "description": "Metric unit",
                "$ref": "embedfs://unit.schema.json"
              },
              "value": {
                "type": "number"
              }
            }
          },
          "nodes": {
            "description": "Node list expression",
            "type": "string"
          },
          "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",
          "nodes",
          "topology",
          "processorType",
          "socketsPerNode",
          "coresPerSocket",
          "threadsPerCore",
          "flopRateScalar",
          "flopRateSimd",
          "memoryBandwidth"
        ]
      },
      "minItems": 1
    }
  },
  "required": [
    "name",
    "metricConfig",
    "subClusters"
  ]
}