[Bug]: under long hour testing, MTP with enable_schedule_overlap=false shows request prompt is too long

[phantomlei3]

Your environment

MLU
commit id:3011eb4c7a0f4722204878298bafe04c66164ea0

🐛 Describe the bug

test model: GLM-5-W8A8
start command:

for ((i = 0; i < NNODES; i++)); do
    DEVICE=$((START_DEVICE + i))
    LOG_FILE="${LOG_DIR}/node_${i}.log"
    # node_rank = server_rank * NNODES + i
    NODE_RANK=$((SERVER_RANK * NNODES + i))
    xllm \
        --model "${MODEL_PATH}" \
        --devices="mlu:${DEVICE}" \
	--draft_model "${MODEL_PATH}" --draft_devices="mlu:${DEVICE}" --num_speculative_tokens 1 \
	--port "${PORT}" \
        --host="0.0.0.0" \
        --master_node_addr="${MASTER_NODE_ADDR}" \
        --nnodes="${WORLD_SIZE}" \
        --max_memory_utilization=0.84 \
        --max_tokens_per_batch="${max_tokens_per_batch}" \
        --max_seqs_per_batch="${max_seqs_per_batch}" \
        --block_size=16 \
        --max_cache_size=0 \
        --enable_prefix_cache=true \
        --enable_chunked_prefill=true \
        --enable_schedule_overlap=false \
        --enable_prefill_sp=false \
        --node_rank="${NODE_RANK}" \
        --enable_shm=false \
        --enable_graph=false \
        --random_seed=42 \
        --reasoning_parser glm5 \
        --tool_call_parser glm5 \
        --expert_parallel_degree=2 \
	--dp_size=4 \
        --ep_size=${WORLD_SIZE} \
    > "${LOG_FILE}" 2>&1 &
done

description
There is no request but prefill_only_scheduler consistently shows "Request prompt is too long, no enough memory to schedule a single sequence." after long-hour testing even I send a small prompt

message:

I20260329 21:33:44.348765 28607 request.cpp:92] x-request-id: , x-request-time: , request_id: chatcmpl-7722415046857869456-MKtu3aAF95WPirBSMLMrGj, sequence 0, max_tokens: 32768, temperature: 1, finish_reason: stop, prompt_tokens: 32792, generated_tokens: 24, ttft: 25476.0ms, total_latency: 26504.9ms, avg tpot: 44.7ms, generation speed: 23.3 tokens/s
I20260329 21:34:05.048291 28608 request.cpp:92] x-request-id: , x-request-time: , request_id: chatcmpl-7722415046857869456-ZwUgB4NC5EFmck9V5YPobM, sequence 0, max_tokens: 32768, temperature: 1, finish_reason: stop, prompt_tokens: 13723, generated_tokens: 514, ttft: 19702.0ms, total_latency: 40283.4ms, avg tpot: 40.1ms, generation speed: 25.0 tokens/s
I20260329 21:34:31.200860 28609 request.cpp:92] x-request-id: , x-request-time: , request_id: chatcmpl-7722415046857869456-RqJFEM7XjtCorrue24fhGz, sequence 0, max_tokens: 32768, temperature: 1, finish_reason: stop, prompt_tokens: 90582, generated_tokens: 631, ttft: 38378.0ms, total_latency: 63884.9ms, avg tpot: 40.5ms, generation speed: 24.7 tokens/s
E20260329 21:34:33.941910 28640 prefill_only_scheduler.cpp:222] Request prompt is too long, no enough memory to schedule a single sequence.
E20260329 21:34:33.942020 28640 prefill_only_scheduler.cpp:222] Request prompt is too long, no enough memory to schedule a single sequence.
E20260329 21:34:33.942056 28640 prefill_only_scheduler.cpp:222] Request prompt is too long, no enough memory to schedule a single sequence.
I20260329 21:34:33.942041 28610 request.cpp:118] x-request-id: , x-request-time: , request_id: chatcmpl-7722415046857869456-xRTKq8FZFo92LDQD5tbz7A, sequence 0, max_tokens: 32768, temperature: 1, prompt_tokens: 5720, status_code : 5, status_msg : No enough memory to schedule single sequence
E20260329 21:34:33.942098 28640 prefill_only_scheduler.cpp:222] Request prompt is too long, no enough memory to schedule a single sequence.
I20260329 21:34:33.942086 28608 request.cpp:118] x-request-id: , x-request-time: , request_id: chatcmpl-7722415046857869456-q7LGa6hn46ZBZyd8x9AQsY, sequence 0, max_tokens: 32768, temperature: 1, prompt_tokens: 30746, status_code : 5, status_msg : No enough memory to schedule single sequence
E20260329 21:34:33.942142 28640 prefill_only_scheduler.cpp:222] Request prompt is too long, no enough memory to schedule a single sequence.
I20260329 21:34:33.942102 28609 request.cpp:118] x-request-id: , x-request-time: , request_id: chatcmpl-7722415046857869456-HUyfF9NJH7DndYQDT49Gue, sequence 0, max_tokens: 32768, temperature: 1, prompt_tokens: 33069, status_code : 5, status_msg : No enough memory to schedule single sequence
E20260329 21:34:33.942168 28640 prefill_only_scheduler.cpp:222] Request prompt is too long, no enough memory to schedule a single sequence.
E20260329 21:34:33.942193 28640 prefill_only_scheduler.cpp:222] Request prompt is too long, no enough memory to schedule a single sequence.
I20260329 21:34:33.942723 28607 request.cpp:92] x-request-id: , x-request-time: , request_id: chatcmpl-7722415046857869456-bAz5smbowDGcwYhymnVjsm, sequence 0, max_tokens: 32768, temperature: 1, finish_reason: stop, prompt_tokens: 8142, generated_tokens: 61, ttft: 62676.0ms, total_latency: 65106.9ms, avg tpot: 40.5ms, generation speed: 25.1 tokens/s
I20260329 21:34:33.943156 28610 request.cpp:118] x-request-id: , x-request-time: , request_id: chatcmpl-7722415046857869456-8aJPb3GxWAKoK6nnjdQwbR, sequence 0, max_tokens: 32768, temperature: 1, prompt_tokens: 24306, status_code : 5, status_msg : No enough memory to schedule single sequence
I20260329 21:34:33.943763 28609 request.cpp:118] x-request-id: , x-request-time: , request_id: chatcmpl-7722415046857869456-tq7MeL6F2pB3i5TRQHqSDe, sequence 0, max_tokens: 32768, temperature: 1, prompt_tokens: 91555, status_code : 5, status_msg : No enough memory to schedule single sequence

[RobbieLeung]

#1146 will fix it.