feat: extract and ingest more metrics context (aggregation temporality, unit and monotonicity) (#136)

.changeset/grumpy-paws-do.md

@@ -0,0 +1,7 @@
+---
+'@hyperdx/api': minor
+'@hyperdx/app': minor
+---
+
+feat: extract and ingest more metrics context (aggregation temporality, unit and
+monotonicity)

docker/ingestor/core.toml

@@ -709,8 +709,12 @@ source = '''
   if err == null && structured.event == "metric" {
     # TODO: do this at extract_token
     .hdx_token = del(structured.fields.__HDX_API_KEY)
-    .dt = structured.fields.metric_type
-    filtered_keys = ["metric_type"]
+    .at = to_int(del(structured.fields.metric_aggregation_temporality)) ?? 0
+    .dt = del(structured.fields.metric_type)
+    .im = to_bool(del(structured.fields.metric_is_monotonic)) ?? null
+    .u = del(structured.fields.metric_unit)
+
+    filtered_keys = []
     for_each(object(structured.fields) ?? {})-> |key, value| {
       if is_integer(value) || is_float(value) {
         filtered_keys = push(filtered_keys, key)

docker/otel-collector/config.yaml

@@ -30,6 +30,22 @@ processors:
       - key: __HDX_API_KEY
         from_context: authorization
         action: upsert
+  # TODO: use transform to attach __HDX_API_KEY attribute to spans/metrics/logs
+  transform:
+    error_mode: ignore
+    metric_statements:
+      - context: resource
+        statements:
+          # map metrics context to resource context (so splunk_hec will capture it)
+          - set(attributes["metric_aggregation_temporality"], "0")
+          - set(attributes["metric_unit"], "")
+          - set(attributes["metric_is_monotonic"], false)
+      - context: metric
+        statements:
+          - set(resource.attributes["metric_aggregation_temporality"],
+            aggregation_temporality)
+          - set(resource.attributes["metric_unit"], unit)
+          - set(resource.attributes["metric_is_monotonic"], is_monotonic)
   batch:
   memory_limiter:
     # 80% of maximum memory up to 2G
@@ -78,7 +94,7 @@ service:
       exporters: [logzio/traces, logging]
     metrics:
       receivers: [otlp]
-      processors: [attributes/attachHdxKey, memory_limiter, batch]
+      processors: [attributes/attachHdxKey, transform, memory_limiter, batch]
       exporters: [splunk_hec, logging]
     logs:
       receivers: [otlp, fluentforward]

packages/api/src/clickhouse/index.ts

@@ -633,8 +633,11 @@ const getMetricsTagsUncached = async (teamId: string) => {
   const query = SqlString.format(
     `
         SELECT 
-          format('{} - {}', name, data_type) as name,
+          any(is_delta) as is_delta,
+          any(is_monotonic) as is_monotonic,
+          any(unit) as unit,
           data_type,
+          format('{} - {}', name, data_type) as name,
           groupUniqArray(_string_attributes) AS tags
         FROM ??
         GROUP BY name, data_type

packages/api/src/utils/logParser.ts

@@ -6,6 +6,11 @@ export type JSONBlob = Record<string, any>;
 
 export type KeyPath = string[];
 
+export enum AggregationTemporality {
+  Delta = 1,
+  Cumulative = 2,
+}
+
 export enum LogType {
   Log = 'log',
   Metric = 'metric',
@@ -67,8 +72,11 @@ export type LogStreamModel = KeyValuePairs &
 export type MetricModel = {
   _string_attributes: Record<string, string>;
   data_type: string;
+  is_delta: boolean;
+  is_monotonic: boolean;
   name: string;
   timestamp: number;
+  unit: string;
   value: number;
 };
 
@@ -207,14 +215,17 @@ export type VectorSpan = {
 };
 
 export type VectorMetric = {
+  at: number; // aggregation temporality
   authorization?: string;
   b: JSONBlob; // tags
   dt: string; // data type
   hdx_platform: string;
   hdx_token: string;
+  im: boolean; // is monotonic
   n: string; // name
   ts: number; // timestamp
   tso: number; // observed timestamp
+  u: string; // unit
   v: number; // value
 };
 
@@ -276,8 +287,11 @@ class VectorMetricParser extends ParsingInterface<VectorMetric> {
     return {
       _string_attributes: metric.b,
       data_type: metric.dt,
+      is_delta: metric.at === AggregationTemporality.Delta,
+      is_monotonic: metric.im,
       name: metric.n,
       timestamp: metric.ts,
+      unit: metric.u,
       value: metric.v,
     };
   }