Package com.linkedin.venice.jobs
Class StageMetricsSnapshot
java.lang.Object
com.linkedin.venice.jobs.StageMetricsSnapshot
Immutable, engine-agnostic snapshot of stage metrics captured after a VPJ job completes.
Use
StageMetricsRegistry.snapshot() to create
an instance after dataFrame.count() triggers execution.-
Nested Class Summary
Nested ClassesModifier and TypeClassDescriptionstatic classImmutable metrics for a single pipeline stage. -
Constructor Summary
ConstructorsConstructorDescription -
Method Summary
Modifier and TypeMethodDescriptionGet the formatted diagnostic report.Get the summary for a specific stage, or null if not registered.toString()
-
Constructor Details
-
StageMetricsSnapshot
-
-
Method Details
-
getStage
Get the summary for a specific stage, or null if not registered. -
getFormattedReport
Get the formatted diagnostic report. Generated lazily on first call.Example output:
Stage Records In Bytes In Records Out Bytes Out Task Time ───────────────────── ────────── ──────── ─────────── ───────── ──────── ttl_filter 1,234,567 456.0 MB 1,222,222 451.0 MB 5.1s compaction 1,222,222 451.0 MB 1,100,000 410.0 MB —
Task Time semantics: For mapPartitions stages (ttl_filter, compression_reencode, kafka_write), this is wall-clock time per partition summed across partitions. For flatMapGroups stages (compaction, chunk_assembly), this is per-group elapsed time summed across all groups. Both represent total task time, not wall-clock duration.
-
toString
-