39819962875
85%

Ran 22 Oct 2026 05:16PM UTC

Jobs 1

Files 136

Run time 1min

Badge

Embed ▾

Committed 22 Oct 2026 05:07PM UTC coverage: 84.139% (+0.006%) from 84.133%

Build # 39819962875

Build Type

push

github

Committed by

web-flow

Commit Message

Run the Hub ShareGPT e2e cases one at a time (#916)

Related: #913

The sim suite runs `data_sharegpt_default` under three load shapes. Each
case streams `ShareGPT_V3_unfiltered_cleaned_split.json` from the Hub.
The file is one 670 MB JSON array, so the `datasets` json builder reads
it whole. One case holds about 5.5 GB resident with a 6.2 GB peak. CI
runs the suite with `-n auto --dist loadgroup` on a 16 GB runner, so two
cases can overlap.

On Oct 19-20 the sim suite ran 70 times. The 36 runs without a Hub
overlap had no runner deaths and no short reports. In 31 fork-based runs
with two Hub cases in flight, one runner died with exit 143 (#542) and
two reports came back short (#913, and 389 of 500 on the fast Hub case
in #908). #526 starts workers with forkserver and died in all four of
its attempts, each with two Hub cases in flight. That needs its own fix
in #526.

The Hub cases now share one xdist worker, the same way the vLLM CPU
tests share `vllm-cpu-server`. The whole-file read is tracked in #923.

Changes:
* Pin the `data_sharegpt_default` param to
`xdist_group(name="sharegpt-hub")`.

Coverage Stats

12440 of 14785 relevant lines covered (84.14%)

0.84 hits per line

Jobs

ID	Job ID	Ran	Files	Coverage
1	39819962875.1	22 Oct 2026 05:16PM UTC	136	84.14	GitHub Action Run

kubernetes-sigs / inference-perf / 39819962875
85%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 39819962875

kubernetes-sigs / inference-perf / 39819962875 85%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 39819962875

kubernetes-sigs / inference-perf / 39819962875
85%

README BADGES
x