Stop reading cur_batch in is_fully_idle and abort_request (#29406)

python/sglang/srt/managers/scheduler.py

@@ -3580,7 +3580,6 @@ class Scheduler(
             and self.chunked_req is None
             and not self.dllm_manager.any_staging_reqs()
             and (self.last_batch is None or self.last_batch.is_empty())
-            and (self.cur_batch is None or self.cur_batch.is_empty())
             and (not self.enable_overlap or len(self.result_queue) == 0)
             and self._pp_microbatches_drained()
         )
@@ -3967,7 +3966,7 @@ class Scheduler(
 
         # Delete requests in the running batch
         if self.ps.pp_size == 1:
-            inflight_batches = [self.running_batch, self.cur_batch]
+            inflight_batches = [self.running_batch, self.last_batch]
         else:
             inflight_batches = [*self.running_mbs, *self.mbs]
 

python/sglang/srt/managers/scheduler_components/batch_result_processor.py

@@ -16,8 +16,8 @@ import torch
 from sglang.srt.disaggregation.utils import DisaggregationMode
 from sglang.srt.environ import envs
 from sglang.srt.layers.logits_processor import LogitsProcessorOutput
-from sglang.srt.managers.io_struct import AbortReq
 from sglang.srt.managers.schedule_batch import (
+    FINISH_ABORT,
     Req,
     ScheduleBatch,
 )
@@ -492,7 +492,7 @@ class SchedulerBatchResultProcessor:
             logger.error(
                 f"Grammar accept_token failed for req {req.rid} with token {next_token_id}: {e}"
             )
-            self.abort_request(AbortReq(rid=req.rid))
+            req.to_finish = FINISH_ABORT()
         req.grammar.finished = req.finished()
 
     def _apply_chunked_prefill_logprobs(
@@ -611,7 +611,7 @@ class SchedulerBatchResultProcessor:
                 f"Grammar accept_token failed for req {req.rid} with token "
                 f"{tokens}: {e}"
             )
-            self.abort_request(AbortReq(rid=req.rid))
+            req.to_finish = FINISH_ABORT()
         return retained
 
     def process_batch_result_idle(