Repository navigation
feat: experimental setup for paper submission - #907
Open
milindsrivastava1997 wants to merge 20 commits into
Open
milindsrivastava1997 wants to merge 20 commits into
milindsrivastava1997 wants to merge 20 commits into
Conversation
random_workload_batch.py runs every arm of the given workload sets on a cluster, one job per two-node slot, longest first; cleans a slot's nodes before each job, retries a failed job once under a new name, writes a compare_arms summary per workload (ClickHouse+MV as the reference), and keeps its state on disk so a restarted driver resumes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On resume, a job whose run is still alive is left alone (orphaned) instead of rerun, and --skip-slots keeps the driver off busy nodes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
collate_batch.py writes one row per (set, workload, arm): CPU, memory, latency p50/p95/sum, failed queries, the feeder's ingest lag, and whether the run is valid (no failed queries, answers agreeing with Prometheus). It reports each baseline's ratio to ASAPQuery over valid pairs. plot_cost_latency.py draws p95 latency against total CPU per set, invalid runs hollow. compare_arms now reports failed queries instead of crashing. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… each scatter Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…chosen arms The raw table's label-first order made every query read the whole table. The batch driver can rerun some arms under a name tag and from a config variant directory. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It cleaned and reused them, wrecking the runs still going there. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…cost-latency plot Also a figure title, thicker axes, and "total compute cost" on the y axis. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
random_workload_artifacts.py --dashboards K writes <workload>-dashK/ run configs with every query group repeated K times (e2e, VictoriaMetrics, ClickHouse and ClickHouse+MV), on the same plan, feeder and data. Each copy is its own client group and thread, so copies are issued concurrently and their latencies recorded apart. The query client's connection pool now fits every query in flight; the default 10 would open a new connection per excess request at K >= 2. collate_batch.py collates batches of an arm subset, against ClickHouse+MV when Prometheus didn't run. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A workload table's gauge can carry recorded values (values.replay), which the feeder loops one step per sample in place of generated ones. make_replay.py fits Alibaba v2022 pod CPU into the random workloads' label shape and writes such a table. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Panel subtitles drop the workload set name. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…etheus and ClickHouse Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… first Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Frozen for seed 862 with saved plans for all ten workloads, for a cardinality point between base and x5. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.