andygrove commented on code in PR #6429:
URL: https://github.com/apache/datafusion-comet/pull/6429#discussion_r4283977226
##########
spark/src/test/scala/org/apache/spark/sql/benchmark/CometReadBenchmark.scala:
##########
@@ -49,12 +49,13 @@ import org.apache.comet.{CometConf, WithHdfsCluster}
class CometReadBaseBenchmark extends CometBenchmarkBase {
/**
- * Measure the nine scan I/O accumulators separately from storage work. Run
this benchmark with
- * `--scan-metric-overhead`; add `--reverse-cases` to reverse the real-job
comparison. The first
- * two cases isolate driver creation and task-copy/merge costs. The last
runs 10,000 Spark tasks
- * with zero or nine extra SQL accumulators, including task serialization
and scheduler updates.
- * This is not an end-to-end scan benchmark: it excludes native counters,
JNI traversal, the SQL
- * UI's per-node rendering, and storage I/O.
+ * Measure scan I/O accumulators separately from storage work. Run this
benchmark with
+ * `--scan-metric-overhead`; add `--reverse-cases` to reverse the
comparisons. Compare the
+ * original nine metrics with the current eleven (including observed GETs
and retries) and a
+ * hypothetical twelfth counter for total HTTP attempts. The first two
benchmarks isolate driver
+ * creation and task-copy/merge costs. The last runs 10,000 Spark tasks,
including task
+ * serialization and scheduler updates, and adds a zero-accumulator
baseline. This excludes
+ * native counters, JNI traversal, the SQL UI's per-node rendering, and
storage I/O.
*/
def scanMetricAccumulatorBenchmark(reverseCases: Boolean): Unit = {
val names = Seq(
Review Comment:
Nit, optional: this list still has to be kept in step with
`CometMetricNode.nativeScanMetrics` by hand, which is how it fell behind last
time. Could the current names come from
`CometMetricNode.nativeScanMetrics(spark.sparkContext).keys.filter(_.startsWith("scan_io_"))`,
with the original nine and the hypothetical attempts counter derived from
that? Then the next scan I/O counter shows up here without anyone remembering
to add it.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
