scheduler.py 18.5 KB
Newer Older
1
2
import enum
import time
3
from typing import Dict, List, Optional, Tuple
Woosuk Kwon's avatar
Woosuk Kwon committed
4

Woosuk Kwon's avatar
Woosuk Kwon committed
5
6
7
8
9
10
11
from vllm.config import CacheConfig, SchedulerConfig
from vllm.core.block_manager import BlockSpaceManager
from vllm.core.policy import PolicyFactory
from vllm.logger import init_logger
from vllm.sequence import (Sequence, SequenceData, SequenceGroup,
                           SequenceGroupMetadata, SequenceOutputs,
                           SequenceStatus)
Woosuk Kwon's avatar
Woosuk Kwon committed
12

Woosuk Kwon's avatar
Woosuk Kwon committed
13
logger = init_logger(__name__)
14

15
_LOGGING_INTERVAL_SEC = 5
Woosuk Kwon's avatar
Woosuk Kwon committed
16
17


18
19
20
21
22
23
24
25
26
27
28
29
30
class PreemptionMode(enum.Enum):
    """Preemption modes.

    1. Swapping: Swap out the blocks of the preempted sequences to CPU memory
    and swap them back in when the sequences are resumed.
    2. Recomputation: Discard the blocks of the preempted sequences and
    recompute them when the sequences are resumed, treating the sequences as
    new prompts.
    """
    SWAP = enum.auto()
    RECOMPUTE = enum.auto()


31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
class SchedulerOutputs:

    def __init__(
        self,
        blocks_to_swap_in: Dict[int, int],
        blocks_to_swap_out: Dict[int, int],
        blocks_to_copy: Dict[int, List[int]],
    ) -> None:
        self.blocks_to_swap_in = blocks_to_swap_in
        self.blocks_to_swap_out = blocks_to_swap_out
        self.blocks_to_copy = blocks_to_copy
        # Swap in and swap out should never happen at the same time.
        assert not (blocks_to_swap_in and blocks_to_swap_out)

    def is_empty(self) -> bool:
46
        return (not self.blocks_to_swap_in and not self.blocks_to_swap_out
47
48
49
                and not self.blocks_to_copy)


Woosuk Kwon's avatar
Woosuk Kwon committed
50
51
class Scheduler:

Woosuk Kwon's avatar
Woosuk Kwon committed
52
    def __init__(
Woosuk Kwon's avatar
Woosuk Kwon committed
53
        self,
54
55
        scheduler_config: SchedulerConfig,
        cache_config: CacheConfig,
Woosuk Kwon's avatar
Woosuk Kwon committed
56
        log_stats: bool,
Woosuk Kwon's avatar
Woosuk Kwon committed
57
    ) -> None:
58
59
        self.scheduler_config = scheduler_config
        self.cache_config = cache_config
Woosuk Kwon's avatar
Woosuk Kwon committed
60
        self.log_stats = log_stats
Woosuk Kwon's avatar
Woosuk Kwon committed
61

62
        # Instantiate the scheduling policy.
63
        self.policy = PolicyFactory.get_policy(policy_name="fcfs")
Woosuk Kwon's avatar
Woosuk Kwon committed
64
        # Create the block space manager.
Woosuk Kwon's avatar
Woosuk Kwon committed
65
        self.block_manager = BlockSpaceManager(
66
67
68
            block_size=self.cache_config.block_size,
            num_gpu_blocks=self.cache_config.num_gpu_blocks,
            num_cpu_blocks=self.cache_config.num_cpu_blocks,
Woosuk Kwon's avatar
Woosuk Kwon committed
69
70
        )

71
72
73
        # Sequence groups in the WAITING state.
        self.waiting: List[SequenceGroup] = []
        # Sequence groups in the RUNNING state.
74
        self.running: List[SequenceGroup] = []
75
        # Sequence groups in the SWAPPED state.
Woosuk Kwon's avatar
Woosuk Kwon committed
76
        self.swapped: List[SequenceGroup] = []
Woosuk Kwon's avatar
Woosuk Kwon committed
77

Woosuk Kwon's avatar
Woosuk Kwon committed
78
79
80
        self.last_logging_time: float = 0.0
        # List[timestamp, num_tokens]
        self.num_input_tokens: List[Tuple[float, int]] = []
81

82
    def add_seq_group(self, seq_group: SequenceGroup) -> None:
83
        # Add sequence groups to the waiting queue.
84
        self.waiting.append(seq_group)
Woosuk Kwon's avatar
Woosuk Kwon committed
85

86
87
88
89
90
91
92
93
94
95
96
97
    def abort_seq_group(self, request_id: str) -> None:
        for state_queue in [self.waiting, self.running, self.swapped]:
            for seq_group in state_queue:
                if seq_group.request_id == request_id:
                    # Remove the sequence group from the state queue.
                    state_queue.remove(seq_group)
                    for seq in seq_group.seqs:
                        if seq.is_finished():
                            continue
                        self.free_seq(seq, SequenceStatus.FINISHED_ABORTED)
                    return

98
99
100
    def has_unfinished_seqs(self) -> bool:
        return self.waiting or self.running or self.swapped

101
102
103
    def get_num_unfinished_seq_groups(self) -> int:
        return len(self.waiting) + len(self.running) + len(self.swapped)

104
105
    def _schedule(
            self) -> Tuple[SchedulerOutputs, List[str], List[SequenceGroup]]:
106
107
108
        # Blocks that need to be swaped or copied before model execution.
        blocks_to_swap_in: Dict[int, int] = {}
        blocks_to_swap_out: Dict[int, int] = {}
109
        blocks_to_copy: Dict[int, List[int]] = {}
Lily Liu's avatar
Lily Liu committed
110
        ignored_seq_groups: List[SequenceGroup] = []
111

112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
        # Fix the current time.
        now = time.time()

        # NOTE(woosuk): We prioritize the sequence groups in the RUNNING state
        # in order to minimize the preemption overheads.
        # Preemption happens only when there is no available slot to keep all
        # the sequence groups in the RUNNING state.
        # In this case, the policy is responsible for deciding which sequence
        # groups to preempt.
        self.running = self.policy.sort_by_priority(now, self.running)

        # Reserve new token slots for the running sequence groups.
        running: List[SequenceGroup] = []
        preempted: List[SequenceGroup] = []
        while self.running:
            seq_group = self.running.pop(0)
128
            while not self.block_manager.can_append_slot(seq_group):
129
130
131
132
133
134
135
136
137
138
                if self.running:
                    # Preempt the lowest-priority sequence groups.
                    victim_seq_group = self.running.pop(-1)
                    self._preempt(victim_seq_group, blocks_to_swap_out)
                    preempted.append(victim_seq_group)
                else:
                    # No other sequence groups can be preempted.
                    # Preempt the current sequence group.
                    self._preempt(seq_group, blocks_to_swap_out)
                    preempted.append(seq_group)
Woosuk Kwon's avatar
Woosuk Kwon committed
139
140
                    break
            else:
141
                # Append new slots to the sequence group.
142
                self._append_slot(seq_group, blocks_to_copy)
143
144
145
146
147
                running.append(seq_group)
        self.running = running

        # Swap in the sequence groups in the SWAPPED state if possible.
        self.swapped = self.policy.sort_by_priority(now, self.swapped)
148
        while self.swapped and not blocks_to_swap_out:
149
150
151
152
153
154
            seq_group = self.swapped[0]
            # If the sequence group has been preempted in this step, stop.
            if seq_group in preempted:
                break
            # If the sequence group cannot be swapped in, stop.
            if not self.block_manager.can_swap_in(seq_group):
Woosuk Kwon's avatar
Woosuk Kwon committed
155
156
                break

157
158
            # The total number of sequences in the RUNNING state should not
            # exceed the maximum number of sequences.
159
            num_new_seqs = seq_group.num_seqs(status=SequenceStatus.SWAPPED)
160
161
162
            num_curr_seqs = sum(
                seq_group.num_seqs(status=SequenceStatus.RUNNING)
                for seq_group in self.running)
163
164
            if (num_curr_seqs + num_new_seqs >
                    self.scheduler_config.max_num_seqs):
165
166
                break

167
168
            seq_group = self.swapped.pop(0)
            self._swap_in(seq_group, blocks_to_swap_in)
169
            self._append_slot(seq_group, blocks_to_copy)
170
            self.running.append(seq_group)
171

172
173
        num_batched_tokens = sum(
            seq_group.num_seqs(status=SequenceStatus.RUNNING)
174
            for seq_group in self.running)
175

176
        # Join waiting sequences if possible.
177
        prompt_group_ids: List[str] = []
178
179
180
181
        # NOTE(woosuk): The sequence groups in the SWAPPED state are strictly
        # prioritized over the sequence groups in the WAITING state.
        # This is because we want to bound the amount of CPU memory taken by
        # the swapped sequence groups.
Woosuk Kwon's avatar
Woosuk Kwon committed
182
        if not self.swapped:
183
184
185
            # Optimization: We do not sort the waiting queue since the preempted
            # sequence groups are added to the front and the new sequence groups
            # are added to the back.
186
187
188
189
190
            while self.waiting:
                seq_group = self.waiting[0]
                # If the sequence group has been preempted in this step, stop.
                if seq_group in preempted:
                    break
Lily Liu's avatar
Lily Liu committed
191
192

                num_prompt_tokens = seq_group.get_seqs()[0].get_len()
193
                if num_prompt_tokens > self.scheduler_config.max_seq_len:
194
                    logger.warning(
Lily Liu's avatar
Lily Liu committed
195
196
197
198
199
200
201
202
203
                        f"Input prompt ({num_prompt_tokens} tokens) is too long"
                        " and exceeds limit of "
                        f"{self.scheduler_config.max_seq_len}")
                    for seq in seq_group.get_seqs():
                        seq.status = SequenceStatus.FINISHED_IGNORED
                    ignored_seq_groups.append(seq_group)
                    self.waiting.pop(0)
                    break

204
205
206
207
208
                # If the sequence group cannot be allocated, stop.
                if not self.block_manager.can_allocate(seq_group):
                    break

                # If the number of batched tokens exceeds the limit, stop.
209
210
                if (num_batched_tokens + num_prompt_tokens >
                        self.scheduler_config.max_num_batched_tokens):
211
212
                    break

213
214
                # The total number of sequences in the RUNNING state should not
                # exceed the maximum number of sequences.
215
216
                num_new_seqs = seq_group.num_seqs(
                    status=SequenceStatus.WAITING)
217
218
219
                num_curr_seqs = sum(
                    seq_group.num_seqs(status=SequenceStatus.RUNNING)
                    for seq_group in self.running)
220
221
                if (num_curr_seqs + num_new_seqs >
                        self.scheduler_config.max_num_seqs):
222
223
                    break

224
225
226
227
                seq_group = self.waiting.pop(0)
                self._allocate(seq_group)
                self.running.append(seq_group)
                num_batched_tokens += num_prompt_tokens
228
                prompt_group_ids.append(seq_group.request_id)
Woosuk Kwon's avatar
Woosuk Kwon committed
229

230
231
232
233
234
        scheduler_outputs = SchedulerOutputs(
            blocks_to_swap_in=blocks_to_swap_in,
            blocks_to_swap_out=blocks_to_swap_out,
            blocks_to_copy=blocks_to_copy,
        )
Woosuk Kwon's avatar
Woosuk Kwon committed
235
        if not self.log_stats:
Lily Liu's avatar
Lily Liu committed
236
            return scheduler_outputs, prompt_group_ids, ignored_seq_groups
Woosuk Kwon's avatar
Woosuk Kwon committed
237

Zhuohan Li's avatar
Zhuohan Li committed
238
        # TODO(woosuk): Move the below code to the engine.
Woosuk Kwon's avatar
Woosuk Kwon committed
239
240
241
242
243
244
        now = time.time()
        if num_batched_tokens > 0:
            self.num_input_tokens.append((now, num_batched_tokens))
        elapsed_time = now - self.last_logging_time
        if elapsed_time > _LOGGING_INTERVAL_SEC:
            self.last_logging_time = now
245
246
            self.num_input_tokens = [(t, n) for t, n in self.num_input_tokens
                                     if now - t < _LOGGING_INTERVAL_SEC]
Woosuk Kwon's avatar
Woosuk Kwon committed
247
            if len(self.num_input_tokens) > 1:
248
249
                total_num_tokens = sum(n
                                       for _, n in self.num_input_tokens[:-1])
Woosuk Kwon's avatar
Woosuk Kwon committed
250
251
252
253
254
                window = now - self.num_input_tokens[0][0]
                avg_throughput = total_num_tokens / window
            else:
                avg_throughput = 0.0

255
            total_num_gpu_blocks = self.cache_config.num_gpu_blocks
Woosuk Kwon's avatar
Woosuk Kwon committed
256
            num_free_gpu_blocks = self.block_manager.get_num_free_gpu_blocks()
257
258
259
260
261
            num_used_gpu_blocks = total_num_gpu_blocks - num_free_gpu_blocks
            gpu_cache_usage = num_used_gpu_blocks / total_num_gpu_blocks

            total_num_cpu_blocks = self.cache_config.num_cpu_blocks
            if total_num_cpu_blocks > 0:
262
263
                num_free_cpu_blocks = (
                    self.block_manager.get_num_free_cpu_blocks())
264
265
                num_used_cpu_blocks = total_num_cpu_blocks - num_free_cpu_blocks
                cpu_cache_usage = num_used_cpu_blocks / total_num_cpu_blocks
Woosuk Kwon's avatar
Woosuk Kwon committed
266
267
268
            else:
                cpu_cache_usage = 0.0

269
270
271
272
273
274
            logger.info(f"Throughput: {avg_throughput:.1f} tokens/s, "
                        f"Running: {len(self.running)} reqs, "
                        f"Swapped: {len(self.swapped)} reqs, "
                        f"Pending: {len(self.waiting)} reqs, "
                        f"GPU KV cache usage: {gpu_cache_usage * 100:.1f}%, "
                        f"CPU KV cache usage: {cpu_cache_usage * 100:.1f}%")
Lily Liu's avatar
Lily Liu committed
275
        return scheduler_outputs, prompt_group_ids, ignored_seq_groups
Woosuk Kwon's avatar
Woosuk Kwon committed
276

277
278
279
280
    def schedule(
        self
    ) -> Tuple[List[SequenceGroupMetadata], SchedulerOutputs,
               List[SequenceGroup]]:
281
282
283
        # Schedule sequence groups.
        # This function call changes the internal states of the scheduler
        # such as self.running, self.swapped, and self.waiting.
284
285
        (scheduler_outputs, prompt_group_ids,
         ignored_seq_groups) = self._schedule()
286
287

        # Create input data structures.
288
        seq_group_metadata_list: List[SequenceGroupMetadata] = []
289
        for seq_group in self.running:
290
            is_prompt = seq_group.request_id in prompt_group_ids
291

292
            seq_data: Dict[int, List[SequenceData]] = {}
293
294
            block_tables: Dict[int, List[int]] = {}
            for seq in seq_group.get_seqs(status=SequenceStatus.RUNNING):
295
                seq_id = seq.seq_id
296
                seq_data[seq_id] = seq.data
297
                block_tables[seq_id] = self.block_manager.get_block_table(seq)
298

299
            seq_group_metadata = SequenceGroupMetadata(
300
                request_id=seq_group.request_id,
301
                is_prompt=is_prompt,
302
                seq_data=seq_data,
303
                sampling_params=seq_group.sampling_params,
304
305
                block_tables=block_tables,
            )
306
            seq_group_metadata_list.append(seq_group_metadata)
Lily Liu's avatar
Lily Liu committed
307
        return seq_group_metadata_list, scheduler_outputs, ignored_seq_groups
308

309
    def update(
Woosuk Kwon's avatar
Woosuk Kwon committed
310
        self,
311
        seq_outputs: Dict[int, SequenceOutputs],
312
    ) -> List[SequenceGroup]:
Woosuk Kwon's avatar
Woosuk Kwon committed
313
        # Update the running sequences and free blocks.
314
        for seq_group in self.running:
315
316
            # Process beam search results before processing the new tokens.
            for seq in seq_group.get_seqs(status=SequenceStatus.RUNNING):
317
318
                output = seq_outputs[seq.seq_id]
                if seq.seq_id != output.parent_seq_id:
319
320
                    # The sequence is a fork of the parent sequence (beam
                    # search). Free the current sequence.
Woosuk Kwon's avatar
Woosuk Kwon committed
321
322
                    self.block_manager.free(seq)
                    # Fork the parent sequence.
323
324
                    parent_seq = seq_group.find(output.parent_seq_id)
                    parent_seq.fork(seq)
Woosuk Kwon's avatar
Woosuk Kwon committed
325
326
                    self.block_manager.fork(parent_seq, seq)

327
328
            # Process the new tokens.
            for seq in seq_group.get_seqs(status=SequenceStatus.RUNNING):
Woosuk Kwon's avatar
Woosuk Kwon committed
329
                # Append a new token to the sequence.
330
                output = seq_outputs[seq.seq_id]
331
                seq.append_token_id(output.output_token, output.logprobs)
Zhuohan Li's avatar
Zhuohan Li committed
332
333
        # Return a shallow copy of the running queue to prevent the queue
        # from being modified by the caller.
334
        return self.running.copy()
Woosuk Kwon's avatar
Woosuk Kwon committed
335

Zhuohan Li's avatar
Zhuohan Li committed
336
337
    def free_seq(self, seq: Sequence, finish_status: SequenceStatus) -> None:
        seq.status = finish_status
338
        self.block_manager.free(seq)
Woosuk Kwon's avatar
Woosuk Kwon committed
339

340
341
342
343
344
    def free_finished_seq_groups(self) -> None:
        self.running = [
            seq_group for seq_group in self.running
            if not seq_group.is_finished()
        ]
Woosuk Kwon's avatar
Woosuk Kwon committed
345

346
347
    def _allocate(self, seq_group: SequenceGroup) -> None:
        self.block_manager.allocate(seq_group)
348
        for seq in seq_group.get_seqs():
349
350
            seq.status = SequenceStatus.RUNNING

351
    def _append_slot(
352
353
354
355
356
        self,
        seq_group: SequenceGroup,
        blocks_to_copy: Dict[int, List[int]],
    ) -> None:
        for seq in seq_group.get_seqs(status=SequenceStatus.RUNNING):
357
            ret = self.block_manager.append_slot(seq)
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
            if ret is not None:
                src_block, dst_block = ret
                if src_block in blocks_to_copy:
                    blocks_to_copy[src_block].append(dst_block)
                else:
                    blocks_to_copy[src_block] = [dst_block]

    def _preempt(
        self,
        seq_group: SequenceGroup,
        blocks_to_swap_out: Dict[int, int],
        preemption_mode: Optional[PreemptionMode] = None,
    ) -> None:
        # If preemption mode is not specified, we determine the mode as follows:
        # We use recomputation by default since it incurs lower overhead than
        # swapping. However, when the sequence group has multiple sequences
        # (e.g., beam search), recomputation is not supported. In such a case,
        # we use swapping instead.
        # FIXME(woosuk): This makes our scheduling policy a bit bizarre.
        # As swapped sequences are prioritized over waiting sequences,
        # sequence groups with multiple sequences are implicitly prioritized
        # over sequence groups with a single sequence.
        # TODO(woosuk): Support recomputation for sequence groups with multiple
        # sequences. This may require a more sophisticated CUDA kernel.
        if preemption_mode is None:
            seqs = seq_group.get_seqs(status=SequenceStatus.RUNNING)
            if len(seqs) == 1:
                preemption_mode = PreemptionMode.RECOMPUTE
            else:
                preemption_mode = PreemptionMode.SWAP
        if preemption_mode == PreemptionMode.RECOMPUTE:
            self._preempt_by_recompute(seq_group)
        elif preemption_mode == PreemptionMode.SWAP:
            self._preempt_by_swap(seq_group, blocks_to_swap_out)
        else:
393
            assert False, "Invalid preemption mode."
394
395
396
397
398
399
400
401
402
403

    def _preempt_by_recompute(
        self,
        seq_group: SequenceGroup,
    ) -> None:
        seqs = seq_group.get_seqs(status=SequenceStatus.RUNNING)
        assert len(seqs) == 1
        for seq in seqs:
            seq.status = SequenceStatus.WAITING
            self.block_manager.free(seq)
404
405
406
        # NOTE: For FCFS, we insert the preempted sequence group to the front
        # of the waiting queue.
        self.waiting.insert(0, seq_group)
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433

    def _preempt_by_swap(
        self,
        seq_group: SequenceGroup,
        blocks_to_swap_out: Dict[int, int],
    ) -> None:
        seqs = seq_group.get_seqs(status=SequenceStatus.RUNNING)
        for seq in seqs:
            seq.status = SequenceStatus.SWAPPED
        self._swap_out(seq_group, blocks_to_swap_out)
        self.swapped.append(seq_group)

    def _swap_in(
        self,
        seq_group: SequenceGroup,
        blocks_to_swap_in: Dict[int, int],
    ) -> None:
        mapping = self.block_manager.swap_in(seq_group)
        blocks_to_swap_in.update(mapping)
        for seq in seq_group.get_seqs(status=SequenceStatus.SWAPPED):
            seq.status = SequenceStatus.RUNNING

    def _swap_out(
        self,
        seq_group: SequenceGroup,
        blocks_to_swap_out: Dict[int, int],
    ) -> None:
434
435
436
437
438
439
        if not self.block_manager.can_swap_out(seq_group):
            # FIXME(woosuk): Abort the sequence group instead of aborting the
            # entire engine.
            raise RuntimeError(
                "Aborted due to the lack of CPU swap space. Please increase "
                "the swap space to avoid this error.")
440
441
442
443
        mapping = self.block_manager.swap_out(seq_group)
        blocks_to_swap_out.update(mapping)
        for seq in seq_group.get_seqs(status=SequenceStatus.RUNNING):
            seq.status = SequenceStatus.SWAPPED