You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 3e78ebb
Browse filesBrowse the repository at this point in the historyBrowse files
Fix InfluxDB readback duration and superseded benchmark runs
Read the full InfluxDB corpus through one ordered native JSONL query instead of thousands of repeated sorted pages. Keep bounded row buffering, the frozen record ceiling, cancellation, full-body/digest verification and joined cleanup. Use the vendor-recommended 70% query pool within the existing physical memory budget so deduplication can execute.
Add a real million-record InfluxDB control with update, mid-stream cancellation and healthy continuation. Preserve ClickHouse paging through the common reader contract and handle the last partial native write batch. Newer Benchmarks runs cancel the previous run of their branch, including across full/contracts modes.
Validation: complete Debug solution build with warnings as errors; full formatter verification; repository governance; actionlint; real InfluxDB million-record and ClickHouse million-record controls with physical container/volume cleanup. Full session run40502233284 was cancelled at the owner request and is not complete qualification.
0 commit comments