← Back to reference

Event Schema Reference

Envelope

OTLP-style envelope:

{
  "resourceLogs": [
    {
      "resource": {
        "attributes": [
          { "key": "service.name", "value": "denpex-agent" },
          { "key": "host.name", "value": "node-3" },
          { "key": "denpex.tenant", "value": "acme" }
        ]
      },
      "scopeLogs": [
        {
          "scope": { "name": "denpex.gpu.collector", "version": "0.7.0" },
          "logRecords": [
            {
              "timeUnixNano": 1700000000000000000,
              "severityNumber": 9,
              "severityText": "INFO",
              "attributes": [
                { "key": "gpu.uuid", "value": "GPU-abc" },
                { "key": "metric.kind", "value": "ecc.sbe.delta" },
                { "key": "metric.value", "value": 1 }
              ]
            }
          ]
        }
      ]
    }
  ]
}

Resource Attributes

keyMeaning
host.nameHostname
denpex.tenantTenant ID
denpex.cluster_idCluster CR name
denpex.fabric.kindib, roce, tcp
denpex.gpu.uuidGPU UUID
denpex.job_idTraining job id (when known)

Span / Log Kinds

kindexample
gpu.ecc.dbevalue = <int>
gpu.thermal.throttlevalue = <count>
gpu.xidvalue = <int>
pcie.replayvalue = <int>
nvlink.replayvalue = <int>
ib.link_downedvalue = <int>
ib.link_error_recoveryvalue = <int>
roce.cnpvalue = <count>
nccl.timeoutrank = <int>, op = <string>
gray.stragglerjob_id = <string>, rank = <int>, z = <float>
layer.weight_explosionlayer = <string>, sigma = <float>
layer.gradient_explosionlayer = <string>, sigma = <float>
layer.nan_inflayer = <string>, kind = <string>
layer.checksum_driftlayer = <string>, prev = <hash>, cur = <hash>
optimizer.asymmetrystep = <int>, median = <float>, max_dev = <float>
checkpoint.ingestedobject = <string>, sha256 = <hex>, size = <int>
host.psi.pressurekind = <string>, value = <float>
host.capability_missingcapability = <string>

Metric Kinds

Standard OTLP Sum / Gauge metrics with denpex. prefixes. metric_kind discriminates within an aggregate.

See Also

  • docs/telemetry-schema-reference.md — broader metrics schema.
  • docs/incident-schema-reference.md — incident record schema.