titan-iac/services/monitoring/dashboards/atlas-jobs.json

{
  "uid": "atlas-jobs",
  "title": "Atlas Jobs",
  "folderUid": "atlas-internal",
  "editable": true,
  "panels": [
    {
      "id": 1,
      "type": "bargauge",
      "title": "Ariadne Task Errors (24h)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 7,
        "w": 6,
        "x": 0,
        "y": 0
      },
      "targets": [
        {
          "expr": "sum by (task) (increase(ariadne_task_runs_total{status=\"error\"}[24h]))",
          "refId": "A",
          "legendFormat": "{{task}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "none",
          "min": 0,
          "max": null,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": null
              },
              {
                "color": "yellow",
                "value": 1
              },
              {
                "color": "orange",
                "value": 3
              },
              {
                "color": "red",
                "value": 5
              }
            ]
          }
        },
        "overrides": []
      },
      "options": {
        "displayMode": "gradient",
        "orientation": "horizontal",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        }
      },
      "transformations": [
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        }
      ]
    },
    {
      "id": 2,
      "type": "timeseries",
      "title": "Ariadne Attempts vs Failures (1h)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 7,
        "w": 12,
        "x": 6,
        "y": 0
      },
      "targets": [
        {
          "expr": "sum(increase(ariadne_task_runs_total[1h]))",
          "refId": "A",
          "legendFormat": "Attempts"
        },
        {
          "expr": "sum(increase(ariadne_task_runs_total{status=\"error\"}[1h]))",
          "refId": "B",
          "legendFormat": "Failures"
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "none"
        },
        "overrides": []
      },
      "options": {
        "legend": {
          "displayMode": "table",
          "placement": "right"
        },
        "tooltip": {
          "mode": "multi"
        }
      }
    },
    {
      "id": 3,
      "type": "bargauge",
      "title": "One-off Job Pods (age hours)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 7,
        "w": 6,
        "x": 18,
        "y": 0
      },
      "targets": [
        {
          "expr": "((time() - kube_pod_start_time{pod!=\"\"}) / 3600) * on(namespace,pod) group_left(owner_name) (kube_pod_owner{owner_kind=\"Job\"} unless on(namespace, owner_name) label_replace(kube_job_owner{owner_kind=\"CronJob\"}, \"owner_name\", \"$1\", \"job_name\", \"(.*)\")) * on(namespace,pod) group_left(phase) max by (namespace,pod,phase) (kube_pod_status_phase{phase=~\"Running|Succeeded\"})",
          "refId": "A",
          "legendFormat": "{{namespace}}/{{pod}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "h",
          "min": 0,
          "max": null,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": null
              },
              {
                "color": "yellow",
                "value": 6
              },
              {
                "color": "orange",
                "value": 24
              },
              {
                "color": "red",
                "value": 48
              }
            ]
          }
        },
        "overrides": []
      },
      "options": {
        "displayMode": "gradient",
        "orientation": "horizontal",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        }
      },
      "transformations": [
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        },
        {
          "id": "limit",
          "options": {
            "limit": 12
          }
        }
      ]
    },
    {
      "id": 4,
      "type": "stat",
      "title": "Glue Jobs Stale (>36h)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 4,
        "w": 4,
        "x": 0,
        "y": 7
      },
      "targets": [
        {
          "expr": "(sum((((time() - (kube_cronjob_status_last_successful_time and on(namespace,cronjob) kube_cronjob_labels{label_atlas_bstein_dev_glue=\"true\"})) > bool 129600) unless on(namespace,cronjob) (kube_cronjob_spec_suspend and on(namespace,cronjob) kube_cronjob_labels{label_atlas_bstein_dev_glue=\"true\"}) == 1)) + count(((kube_cronjob_labels{label_atlas_bstein_dev_glue=\"true\"} unless on(namespace,cronjob) kube_cronjob_status_last_successful_time) unless on(namespace,cronjob) (kube_cronjob_spec_suspend and on(namespace,cronjob) kube_cronjob_labels{label_atlas_bstein_dev_glue=\"true\"}) == 1)))",
          "refId": "A"
        }
      ],
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": null
              },
              {
                "color": "yellow",
                "value": 1
              },
              {
                "color": "orange",
                "value": 2
              },
              {
                "color": "red",
                "value": 3
              }
            ]
          },
          "unit": "none",
          "custom": {
            "displayMode": "auto"
          }
        },
        "overrides": []
      },
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "center",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "textMode": "value"
      }
    },
    {
      "id": 5,
      "type": "stat",
      "title": "Glue Jobs Missing Success",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 4,
        "w": 4,
        "x": 4,
        "y": 7
      },
      "targets": [
        {
          "expr": "count(((kube_cronjob_labels{label_atlas_bstein_dev_glue=\"true\"} unless on(namespace,cronjob) kube_cronjob_status_last_successful_time) unless on(namespace,cronjob) (kube_cronjob_spec_suspend and on(namespace,cronjob) kube_cronjob_labels{label_atlas_bstein_dev_glue=\"true\"}) == 1))",
          "refId": "A"
        }
      ],
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "rgba(115, 115, 115, 1)",
                "value": null
              },
              {
                "color": "green",
                "value": 1
              }
            ]
          },
          "unit": "none",
          "custom": {
            "displayMode": "auto"
          }
        },
        "overrides": []
      },
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "center",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "textMode": "value"
      }
    },
    {
      "id": 6,
      "type": "stat",
      "title": "Glue Jobs Suspended",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 4,
        "w": 4,
        "x": 8,
        "y": 7
      },
      "targets": [
        {
          "expr": "sum((kube_cronjob_spec_suspend and on(namespace,cronjob) kube_cronjob_labels{label_atlas_bstein_dev_glue=\"true\"}) == 1)",
          "refId": "A"
        }
      ],
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "rgba(115, 115, 115, 1)",
                "value": null
              },
              {
                "color": "green",
                "value": 1
              }
            ]
          },
          "unit": "none",
          "custom": {
            "displayMode": "auto"
          }
        },
        "overrides": []
      },
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "center",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "textMode": "value"
      }
    },
    {
      "id": 7,
      "type": "stat",
      "title": "Ariadne Task Errors (1h)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 4,
        "w": 4,
        "x": 12,
        "y": 7
      },
      "targets": [
        {
          "expr": "sum(increase(ariadne_task_runs_total{status=\"error\"}[1h]))",
          "refId": "A"
        }
      ],
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "rgba(115, 115, 115, 1)",
                "value": null
              },
              {
                "color": "green",
                "value": 1
              }
            ]
          },
          "unit": "none",
          "custom": {
            "displayMode": "auto"
          }
        },
        "overrides": []
      },
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "center",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "textMode": "value"
      }
    },
    {
      "id": 8,
      "type": "stat",
      "title": "Ariadne Task Errors (24h)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 4,
        "w": 4,
        "x": 16,
        "y": 7
      },
      "targets": [
        {
          "expr": "sum(increase(ariadne_task_runs_total{status=\"error\"}[24h]))",
          "refId": "A"
        }
      ],
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "rgba(115, 115, 115, 1)",
                "value": null
              },
              {
                "color": "green",
                "value": 1
              }
            ]
          },
          "unit": "none",
          "custom": {
            "displayMode": "auto"
          }
        },
        "overrides": []
      },
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "center",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "textMode": "value"
      }
    },
    {
      "id": 9,
      "type": "stat",
      "title": "Ariadne Task Runs (1h)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 4,
        "w": 4,
        "x": 20,
        "y": 7
      },
      "targets": [
        {
          "expr": "sum(increase(ariadne_task_runs_total[1h]))",
          "refId": "A"
        }
      ],
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "rgba(115, 115, 115, 1)",
                "value": null
              },
              {
                "color": "green",
                "value": 1
              }
            ]
          },
          "unit": "none",
          "custom": {
            "displayMode": "auto"
          }
        },
        "overrides": []
      },
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "center",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "textMode": "value"
      }
    },
    {
      "id": 10,
      "type": "bargauge",
      "title": "Ariadne Schedule Last Error (hours ago)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 11
      },
      "targets": [
        {
          "expr": "(time() - ariadne_schedule_last_error_timestamp_seconds) / 3600",
          "refId": "A",
          "legendFormat": "{{task}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "h",
          "min": 0,
          "max": null,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "red",
                "value": null
              },
              {
                "color": "orange",
                "value": 1
              },
              {
                "color": "yellow",
                "value": 6
              },
              {
                "color": "green",
                "value": 24
              }
            ]
          }
        },
        "overrides": []
      },
      "options": {
        "displayMode": "gradient",
        "orientation": "horizontal",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        }
      },
      "transformations": [
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        }
      ]
    },
    {
      "id": 11,
      "type": "bargauge",
      "title": "Ariadne Schedule Last Success (hours ago)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 11
      },
      "targets": [
        {
          "expr": "(time() - ariadne_schedule_last_success_timestamp_seconds) / 3600",
          "refId": "A",
          "legendFormat": "{{task}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "h",
          "min": 0,
          "max": null,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": null
              },
              {
                "color": "yellow",
                "value": 6
              },
              {
                "color": "orange",
                "value": 24
              },
              {
                "color": "red",
                "value": 48
              }
            ]
          }
        },
        "overrides": []
      },
      "options": {
        "displayMode": "gradient",
        "orientation": "horizontal",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        }
      },
      "transformations": [
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        }
      ]
    },
    {
      "id": 12,
      "type": "bargauge",
      "title": "Glue Jobs Last Success (hours ago)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 19
      },
      "targets": [
        {
          "expr": "((time() - (kube_cronjob_status_last_successful_time and on(namespace,cronjob) kube_cronjob_labels{label_atlas_bstein_dev_glue=\"true\"}))) / 3600",
          "refId": "A",
          "legendFormat": "{{namespace}}/{{cronjob}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "h",
          "min": 0,
          "max": null,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": null
              },
              {
                "color": "yellow",
                "value": 6
              },
              {
                "color": "orange",
                "value": 24
              },
              {
                "color": "red",
                "value": 48
              }
            ]
          }
        },
        "overrides": []
      },
      "options": {
        "displayMode": "gradient",
        "orientation": "horizontal",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        }
      },
      "transformations": [
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        }
      ]
    },
    {
      "id": 13,
      "type": "bargauge",
      "title": "Glue Jobs Last Schedule (hours ago)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 19
      },
      "targets": [
        {
          "expr": "((time() - (kube_cronjob_status_last_schedule_time and on(namespace,cronjob) kube_cronjob_labels{label_atlas_bstein_dev_glue=\"true\"}))) / 3600",
          "refId": "A",
          "legendFormat": "{{namespace}}/{{cronjob}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "h",
          "min": 0,
          "max": null,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": null
              },
              {
                "color": "yellow",
                "value": 6
              },
              {
                "color": "orange",
                "value": 24
              },
              {
                "color": "red",
                "value": 48
              }
            ]
          }
        },
        "overrides": []
      },
      "options": {
        "displayMode": "gradient",
        "orientation": "horizontal",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        }
      },
      "transformations": [
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        }
      ]
    },
    {
      "id": 14,
      "type": "bargauge",
      "title": "Ariadne Task Errors (1h)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 27
      },
      "targets": [
        {
          "expr": "sum by (task) (increase(ariadne_task_runs_total{status=\"error\"}[1h]))",
          "refId": "A",
          "legendFormat": "{{task}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "none",
          "min": 0,
          "max": null,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": null
              },
              {
                "color": "yellow",
                "value": 1
              },
              {
                "color": "orange",
                "value": 3
              },
              {
                "color": "red",
                "value": 5
              }
            ]
          }
        },
        "overrides": []
      },
      "options": {
        "displayMode": "gradient",
        "orientation": "horizontal",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        }
      },
      "transformations": [
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        }
      ]
    },
    {
      "id": 15,
      "type": "bargauge",
      "title": "Ariadne Task Errors (30d)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 27
      },
      "targets": [
        {
          "expr": "sum by (task) (increase(ariadne_task_runs_total{status=\"error\"}[30d]))",
          "refId": "A",
          "legendFormat": "{{task}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "none",
          "min": 0,
          "max": null,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": null
              },
              {
                "color": "yellow",
                "value": 1
              },
              {
                "color": "orange",
                "value": 3
              },
              {
                "color": "red",
                "value": 5
              }
            ]
          }
        },
        "overrides": []
      },
      "options": {
        "displayMode": "gradient",
        "orientation": "horizontal",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        }
      },
      "transformations": [
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        }
      ]
    },
    {
      "id": 16,
      "type": "bargauge",
      "title": "Ariadne Access Requests",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 6,
        "w": 8,
        "x": 0,
        "y": 35
      },
      "targets": [
        {
          "expr": "ariadne_access_requests_total",
          "refId": "A",
          "legendFormat": "{{status}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "none",
          "min": 0,
          "max": null,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": null
              },
              {
                "color": "yellow",
                "value": 50
              },
              {
                "color": "orange",
                "value": 70
              },
              {
                "color": "red",
                "value": 85
              }
            ]
          }
        },
        "overrides": []
      },
      "options": {
        "displayMode": "gradient",
        "orientation": "horizontal",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        }
      },
      "transformations": [
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        }
      ]
    },
    {
      "id": 17,
      "type": "stat",
      "title": "Ariadne CI Coverage (%)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 6,
        "w": 4,
        "x": 8,
        "y": 35
      },
      "targets": [
        {
          "expr": "ariadne_ci_coverage_percent{repo=\"ariadne\"}",
          "refId": "A",
          "legendFormat": "{{branch}}",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "rgba(115, 115, 115, 1)",
                "value": null
              },
              {
                "color": "green",
                "value": 1
              }
            ]
          },
          "unit": "percent",
          "custom": {
            "displayMode": "auto"
          },
          "decimals": 1
        },
        "overrides": []
      },
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "center",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "textMode": "value"
      }
    },
    {
      "id": 18,
      "type": "table",
      "title": "Ariadne CI Tests (latest)",
      "datasource": {
        "type": "prometheus",
        "uid": "atlas-vm"
      },
      "gridPos": {
        "h": 6,
        "w": 12,
        "x": 12,
        "y": 35
      },
      "targets": [
        {
          "expr": "ariadne_ci_tests_total{repo=\"ariadne\"}",
          "refId": "A",
          "instant": true
        }
      ],
      "fieldConfig": {
        "defaults": {
          "unit": "none",
          "custom": {
            "filterable": true
          }
        },
        "overrides": []
      },
      "options": {
        "showHeader": true,
        "columnFilters": false
      },
      "transformations": [
        {
          "id": "labelsToFields",
          "options": {}
        },
        {
          "id": "sortBy",
          "options": {
            "fields": [
              "Value"
            ],
            "order": "desc"
          }
        }
      ]
    }
  ],
  "time": {
    "from": "now-7d",
    "to": "now"
  },
  "annotations": {
    "list": []
  },
  "schemaVersion": 39,
  "style": "dark",
  "tags": [
    "atlas",
    "jobs",
    "glue"
  ]
}