inference/research
verifiedhigh · 0.95updated 2026-10-04

obs-sglang-scheduler-retraction

repo

Fields

id
sources/obs-sglang-scheduler-retraction
Publisher
sgl-project/sglang
Kind
repo
Accessed
2026-10-04
notes
_update_after_schedule: 'KV cache pool is full. Retract requests.' retract_decode() returns retracted_reqs plus reqs_to_abort; aborted requests get HTTPStatus.SERVICE_UNAVAILABLE ('Retracted under decode memory pressure without a KV backup. Retry later.') or INTERNAL_SERVER_ERROR. The warning line logs #retracted_reqs and #new_tokens_gained.

Notes

_update_after_schedule: 'KV cache pool is full. Retract requests.' retract_decode() returns retracted_reqs plus reqs_to_abort; aborted requests get HTTPStatus.SERVICE_UNAVAILABLE ('Retracted under decode memory pressure without a KV backup. Retry later.') or INTERNAL_SERVER_ERROR. The warning line logs #retracted_reqs and #new_tokens_gained.

Cross-references

This record references no other records.