Summary

This change implements slot-based aggregation and semantic gap filling for OpenTelemetry (OTLP) metrics in the otel-plugin, addressing the mismatch between OTLP’s event-based metric model and Netdata’s fixed-interval, slot-based storage model.

Netdata requires exactly one value per dimension per update interval, while OTLP metrics may arrive multiple times within an interval or not at all. This implementation introduces a deterministic aggregation layer that groups incoming OTLP datapoints into fixed time slots, aggregates them according to metric semantics (temporality and monotonicity), and fills gaps when data is missing.

Key design decisions:

• Introduced an explicit, configurable collection_interval instead of relying on auto-detection.
• Implemented slot finalization with a configurable grace period to handle late-arriving data.
• Applied temporality-aware aggregation and gap filling:
  • Gauges repeat the last known value.
  • Delta counters fill gaps with zero.
  • Cumulative monotonic counters repeat the last cumulative value.
• Optimized memory usage by pre-aggregating values per slot instead of storing all datapoints.
• Improved histogram handling by separating bucket/count sums from summary statistics (sum/min/max).

Fixes #21522

Test Plan

Automated tests:

• Added unit tests in netdata_chart.rs:
  • test_delta_counter_gap_fill: verifies that delta counters emit 0 for empty slots.
  • test_cumulative_counter_gap_fill: verifies that cumulative monotonic counters repeat the last value for empty slots.
• All tests pass with cargo test -p otel-plugin.

Manual testing:

• Built the plugin using cargo build --release.
• Ran the plugin in foreground mode with a fixed collection_interval.
• Sent OTLP metrics via the OTLP HTTP /v1/metrics endpoint with irregular timing.
• Verified that:
  • Metrics are emitted only on fixed slot boundaries.
  • Multiple datapoints within a slot result in a single aggregated value.
  • Missing slots are filled according to metric semantics.
  • Late data arriving after slot finalization is dropped (no backfilling).

Additional Information

Previously, OTLP metrics without data in a given interval caused dimensions to be archived, resulting in gaps in charts and unreliable alerting. This change ensures stable visualization and alerting by explicitly filling gaps, while still allowing inactive dimensions to be archived after a configurable timeout.

Backfilling finalized slots is intentionally not supported to avoid reordering complexity and performance overhead, and this behavior is enforced consistently.

Summary by cubic

Adds slot-based aggregation and semantic gap filling to OTLP metrics in the otel-plugin so Netdata emits one value per fixed interval. This stabilizes charts and alerts, and handles late/missing data predictably.

• New Features

  • Fixed-interval slotting with a configurable collection interval and per-metric overrides.
  • Grace period for late data; finalized slots drop late arrivals (no backfill).
  • Gap fill rules: gauges repeat last value; delta counters fill with 0; cumulative monotonic counters repeat the last value.
  • Improved histogram handling: bucket/count outputs plus sum/min/max as gauges with correct units.
  • Chart algorithms chosen by semantics (incremental for monotonic sums; absolute otherwise).
  • New config flags: --otel-metrics-collection-interval, --otel-metrics-grace-period, --otel-metrics-dimension-archive-timeout; optional per-metric collection_interval.

• Migration

  • Set --otel-metrics-collection-interval to your desired frequency; use per-metric overrides if needed.
  • Expect emissions only on slot boundaries; late data after the grace period is dropped.
  • Inactive dimensions auto-archive after --otel-metrics-dimension-archive-timeout.

^{Written for commit a3a4f2a. Summary will update on new commits.}