Public performance evidence
cdc-orchestrator benchmark reports
Static benchmark evidence for cdc-orchestrator-pro with
Redis workloads. The dashboard is generated from the CSV files in this
directory by bundle exec rake benchmark:analyze.
Benchmark matrix
Throughput, latency, CPU, RSS, Redis ops, mailbox depth, and failures across synthetic, Pub/Sub, Streams, CPU+Redis, and pgoutput+Redis workloads.
Soak stability
Repeated batch runs with RSS, GC, throughput degradation, and stability classification for longer-running workload windows.
Chaos recovery
Redis and PostgreSQL restart/outage scenarios with recovery time, attempts, faulted-batch loss, and post-recovery delivery checks.
CLI verification
Operational verification for the patched pipe command:
batch-aware Streams acknowledgement, direct
NestedRuntime execution, and orchestrator metrics.
Result interpretation
The generated analytics show the downstream orchestrator layer
reaching a 62.3k events/s synthetic runtime peak, a 17.0k events/s
Redis Streams peak, and a 16.8k events/s pgoutput+Redis peak. These
are NestedRuntime processing results with worker-local
LocalResourcePool Redis clients after source acquisition;
complete-pipeline discussions should use the
end-to-end fields in the CSV tables.
The soak and chaos artifacts provide the stability side of the same claim: completed benchmark matrices report zero processing failures, the published chaos suites recovered all 31 outage/restart scenarios with zero post-recovery delivery loss, and the June 28 long soak suite completed 15 runs with zero processing failures. Nine soak runs met all stability thresholds; three overnight phases exceeded the RSS growth threshold despite no event-processing failures.
Published scope
This public snapshot includes generated analytics, CSV data, Markdown reports, suite manifests, and SVG charts. Runtime log files are not published because they contain local filesystem paths from the benchmark host.