docs: troubleshooting guide for interpreting anomalous results - #760
Open
Bslabe123 wants to merge 1 commit into
Open
docs: troubleshooting guide for interpreting anomalous results#760Bslabe123 wants to merge 1 commit into
Bslabe123 wants to merge 1 commit into
Conversation
kubernetes-prow
Bot
added
do-not-merge/work-in-progress
github-actions
Bot
added
the
do-not-merge
kubernetes-prow
Bot
added
cncf-cla: yes
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: Bslabe123 The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:
Approvers can indicate their approval by writing |
kubernetes-prow
Bot
added
the
approved
Bslabe123
changed the title
Bslabe123
marked this pull request as ready for review
August 25, 2026 18:40
kubernetes-prow
Bot
removed
the
do-not-merge/work-in-progress
github-actions
Bot
removed
the
do-not-merge
Bslabe123
force-pushed
the
docs/troubleshooting-682
branch
2 times, most recently
from
August 26, 2026 16:02
ca1a763 to
2e4ce2c
Compare
kubernetes-prow
Bot
removed
the
do-not-merge/invalid-commit-message
github-actions
Bot
added
do-not-merge
Bslabe123
force-pushed
the
docs/troubleshooting-682
branch
from
August 26, 2026 17:48
2e4ce2c to
de07987
Compare
github-actions
Bot
added
do-not-merge
Bslabe123
added
the
priority/important-soon
Bslabe123
changed the title
kubernetes-prow
Bot
added
the
needs-rebase
Bslabe123
force-pushed
the
docs/troubleshooting-682
branch
from
August 27, 2026 20:34
de07987 to
88364d4
Compare
kubernetes-prow
Bot
removed
the
needs-rebase
github-actions
Bot
added
do-not-merge
Bslabe123
added
priority/important-longterm
Adds docs/troubleshooting.md, a symptom-indexed guide for the case where a run completes successfully but the output looks wrong. Each symptom routes through a fixed diagnosis procedure that rules out the three root-cause buckets in order of falsifiability cost: the measurement is wrong, the workload is not what was asked for, the system really behaves that way. A fourth outcome (valid but not admissible to a comparison set) is documented separately since its remedy differs. Addresses kubernetes-sigs#682
Bslabe123
force-pushed
the
docs/troubleshooting-682
branch
from
September 3, 2026 20:30
88364d4 to
f0ae293
Compare
github-actions
Bot
added
do-not-merge
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.
Fixes #682. Part of the v0.7.0 release (#606), Documentation pillar.
Adds
docs/troubleshooting.md: a symptom-indexed guide for the reverse direction the existing docs do not cover. The run completed successfully, but the output looks wrong: implausible ITL, throughput above the hardware ceiling, a non-monotonic latency-throughput curve, flat TTFT with climbing e2e, an empty server-metrics field.Structure. Every entry routes through one fixed diagnosis procedure that rules out the issue's three root-cause buckets in order of falsifiability cost:
All confirmation tests use fields already in the reports (
load_summary.requested_ratevsachieved_rate, clientoutput_tokensvsserver_usage.completion_tokens,chunk_times) plus two back-of-envelope physical bounds, so a user needs no rerun and no code reading to reach a verdict. A closing section covers a fourth outcome the issue discussion surfaced: a run that passes every check but is inadmissible to a specific comparison set (sweep shape, stage duration, dataset/seed,ignore_eos). Its remedy differs (rerun under the comparison's constraints), so it is deliberately not a bucket.Relationship to #705. Step 0 of the procedure points at the
validation.jsonreport that #705 adds: errors there are bucket 1 by definition and terminal, warnings are entry points into the manual checks, clean means Step 1 is already done. The guide is written against that interface (filename and severity semantics only, no individual check names) and degrades gracefully while #705 is in review ("run the Step 1 checks by hand"). The two are the human-executed and machine-executed halves of the same invariant set; sequencing-wise this PR should land after #705, or with the Step 0 wording softened if it overtakes it.Citation shift vs the issue body. The issue pointed at #630 for the CI-side invariants. #630 has since been closed as subsumed: the invariants live on main as
e2e/utils/accuracy.py(assert_output_token_accounting,assert_streaming_bookkeeping, via #631/#697), so the guide cites those helpers instead.Also adds a Troubleshooting row to the README doc index.
Rel: #564, #580, #619, #481, #705.