[2/N] elastic-ep: Enable EPLB after scale-up (#30553)

This commit is contained in:
Yoray Zack
2026-07-30 01:06:56 +08:00
committed by GitHub
parent d19999b755
commit 62d0f81f16
6 changed files with 113 additions and 8 deletions
+2
View File
@@ -4453,6 +4453,8 @@ class Scheduler(
pending_ep_size=ElasticEPStateManager.get_pending_ep_size(),
scale_phase=ElasticEPStateManager.get_scale_phase(),
)
if (eplb_manager := self.tp_worker.model_runner.eplb_manager) is not None:
eplb_manager.disable_rebalance("elastic EP scale-up is pending")
logger.debug(
"[Elastic EP][scale] scale requested: target_ep_size=%d; "
"waiting for a joining cohort",