fix(pd): avoid first-token cache stats race - #1450

Open

sufubao wants to merge 14 commits into

ModelTC:mainfrom

sufubao:fix/pd-first-token-cache-stats-race

Commits on Aug 10, 2026

fix(pd): avoid first-token cache stats race
sufubao
committed
26bfc89
fix(pd): preserve prefill metadata and unblock decode
sufubao
committed
edb070b
refactor(pd): simplify first-token fallback
sufubao
committed
ec68ee2
chore: keep PD fix PR focused
sufubao
committed
9126667
refactor(pd): consume node mode once
sufubao
committed
5c6e012
refactor(pd): clarify first-token buffering
sufubao
committed
1097792
refactor(pd): make prefill lookup explicit
sufubao
committed
f069a42
refactor(pd): avoid rescanning buffered tokens
sufubao
committed
f421a9e
refactor(pd): clarify first-token stream handling
sufubao
committed
f46cd17
refactor(pd): streamline first-token buffering
sufubao
committed
8278cde
refactor(pd): reuse async queue for output tokens
sufubao
committed
729ed9b

Commits on Aug 11, 2026

style(pd): apply project formatting
sufubao
committed
267dd44
fix(pd): bound missing prefill token fallback
sufubao
committed
a637af0
Delete unit_tests/server/httpserver/test_pd_master_cached_tokens.py
sufubao
authored
ced1923