[mem_cache] Skip the release-time insert on optimistic prefill requeue; make refresh_fill_ids public (#42194)
build-sglang-image / build (push) Successful in 25m53s

(cherry picked from commit 07e1a4f082)
This commit is contained in:
Liangsheng Yin
2026-10-09 11:18:17 +08:00
committed by minke.yu
parent 6ba34507f1
commit 9ef65b51c2
9 changed files with 37 additions and 18 deletions
+5 -4
View File
@@ -1638,10 +1638,11 @@ class SchedulerDisaggregationPrefillMixin:
)
)
self._release_aborted_request(req)
# Mamba insertion donates the checkpoint and clears its sequence marker.
release_kv_cache(
req, self.tree_cache, is_insert=not self.tree_cache.supports_mamba()
)
# The checkpoint above (maybe_cache_unfinished_req) already handed the
# prefill KV to the tree; the request is not finished, so the release
# only frees the rest. (#42194 semantics; our tree predates #41520's
# mamba-donation rework, mamba models are not run on this branch.)
release_kv_cache(req, self.tree_cache, is_insert=False)
req.reset_for_retract()
req.output_ids = array("q")
req.start_send_idx = 0
+6 -6
View File
@@ -1034,12 +1034,12 @@ class Req(ReqDllmMixin):
else self.origin_input_ids
) # Before image padding
# Each decode stage's output ids. Append-only by contract:
# _refresh_fill_ids infers how many output tokens are already in
# refresh_fill_ids infers how many output tokens are already in
# full_untruncated_fill_ids from lengths alone, so in-place rewrites
# that preserve length would silently corrupt fill_ids.
self.output_ids = array("q")
# Full untruncated sequence: origin + output (+ DLLM mask block).
# Kept in sync by _refresh_fill_ids; admission only updates
# Kept in sync by refresh_fill_ids; admission only updates
# extend_range, never mutates this array's length.
self.full_untruncated_fill_ids = array("q")
self.extend_range: Optional[Range] = None
@@ -1533,7 +1533,7 @@ class Req(ReqDllmMixin):
def get_fill_ids(self) -> array:
return self.full_untruncated_fill_ids[: self.extend_range.end]
def _refresh_fill_ids(self) -> None:
def refresh_fill_ids(self) -> None:
"""Keep full_untruncated_fill_ids == origin_input_ids + output_ids by
appending only the new output tokens.
@@ -1562,7 +1562,7 @@ class Req(ReqDllmMixin):
self._init_fill_ids_for_dllm()
self.determine_dllm_phase()
else:
self._refresh_fill_ids()
self.refresh_fill_ids()
input_len = len(self.full_untruncated_fill_ids)
@@ -3110,7 +3110,7 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
for req, prefix_len in zip(
running_batch.reqs, running_prefix_lens, strict=True
):
req._refresh_fill_ids()
req.refresh_fill_ids()
req.set_extend_range(prefix_len, prefix_len + 1)
# Decode tokens of the running portion live in future_map.output_tokens_buf.
@@ -3188,7 +3188,7 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
# cannot stand in for it. Rows past bs are a beam tail, not requests.
seq_lens = self.seq_lens_cpu[:bs].tolist()
for req, seq_len in zip(self.reqs, seq_lens, strict=True):
req._refresh_fill_ids()
req.refresh_fill_ids()
# end runs one past full_untruncated_fill_ids while output_ids
# trails; safe only while decoding_reqs suppresses cache_unfinished_req.
req.set_extend_range(seq_len - 1, seq_len)
@@ -157,7 +157,7 @@ class Session:
or carry_fill is input_ids_unpadded
):
# Unexpected type or aliased with an origin array (extending it
# below would double-append): let _refresh_fill_ids rebuild.
# below would double-append): let refresh_fill_ids rebuild.
carry_fill = None
else:
del carry_fill[self.committed_fill_len :]
@@ -28,7 +28,7 @@ class _FakeReq:
"""Carries the fill-id state convert_decode_to_extend touches, with the
real Req methods so the array bookkeeping is not re-implemented here."""
_refresh_fill_ids = Req._refresh_fill_ids
refresh_fill_ids = Req.refresh_fill_ids
set_extend_range = Req.set_extend_range
def __init__(self, *, num_prompt_tokens: int, num_output_tokens: int):
@@ -132,7 +132,7 @@ class _FakeReq:
self.full_untruncated_fill_ids = list(range(origin_len + output_len))
self.extend_range = None
def _refresh_fill_ids(self):
def refresh_fill_ids(self):
self.full_untruncated_fill_ids = self.origin_input_ids + self.output_ids
@property
@@ -103,6 +103,19 @@ class MockReq:
def get_fill_ids(self):
return self.full_untruncated_fill_ids[: self.extend_range.end]
rid = "mock-req"
skip_radix_cache_insert = False
finished_reason = None
def finished(self):
return self.finished_reason is not None
def refresh_fill_ids(self):
pass
def owned_kv_len(self):
return self.kv.kv_committed_len
def _make_req(fill_ids, req_pool_idx=0, cache_protected_len=0, last_node=None):
return MockReq(fill_ids, req_pool_idx, cache_protected_len, last_node)
@@ -641,6 +641,11 @@ class _GraftReq:
def get_fill_ids(self):
return array("q", self.fill_ids)
def refresh_fill_ids(self):
pass # fill ids are fixed for the stand-in
def finished(self):
return self.finished_reason is not None
class TestRotationGraftDecline(CustomTestCase):
"""The overlap disagg-prefill loop plans batch t+1 before batch t's radix
@@ -167,7 +167,7 @@ class TestSharedPrefillAdmission(unittest.TestCase):
with get_parallel().override(attn_dcp_size=1):
scheduler.init_req_max_new_tokens(req)
self.assertEqual(req.sampling_params.max_new_tokens, 1)
req._refresh_fill_ids()
req.refresh_fill_ids()
cache = SimpleNamespace(
sliding_window_size=page_size,
@@ -78,7 +78,7 @@ class TestSessionTokenShare(CustomTestCase):
if baked is None:
baked = len(output)
req.output_ids.extend(output[:baked])
req._refresh_fill_ids()
req.refresh_fill_ids()
req.output_ids.extend(output[baked:])
self.session.finish_req(req)
@@ -114,7 +114,7 @@ class TestSessionTokenShare(CustomTestCase):
r2 = self._create("r2", [50, 51])
self.assertEqual(list(r2.origin_input_ids), in1 + out1 + [50, 51])
r2.output_ids.extend([6, 7])
r2._refresh_fill_ids()
r2.refresh_fill_ids()
self.session.abort_req()
self.assertEqual(self.session.committed_origin_len, len(in1))
@@ -149,7 +149,7 @@ class TestSessionTokenShare(CustomTestCase):
# array, then output trimmed to finished_len (like _trim_overshoot)
# before finish.
r1.output_ids.extend([1, 2, 3, 4, 5, 6])
r1._refresh_fill_ids()
r1.refresh_fill_ids()
del r1.output_ids[4:]
self.session.finish_req(r1)
self.assertEqual(
@@ -161,7 +161,7 @@ class TestSessionTokenShare(CustomTestCase):
r2 = self._create("r2", [50])
self.assertEqual(list(r2.origin_input_ids), in1 + [1, 2, 3, 4] + [50])
self.assertEqual(len(r2.full_untruncated_fill_ids), 0) # carry skipped
r2._refresh_fill_ids()
r2.refresh_fill_ids()
self.assertEqual(list(r2.full_untruncated_fill_ids), list(r2.origin_input_ids))