[Core][Metrics][BugFix] Replace num_cached_tokens/num_external_computed_tokens with PrefillStats (#37460)

Related to `Counters can only be incremented by non-negative amounts`
error with the `vllm:prompt_tokens_by_source_total` metric.

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Co-authored-by: Or Ozeri <or@ozery.com>
This commit is contained in:
Mark McLoughlin
2026-04-14 09:00:45 +01:00
committed by GitHub
co-authored by Or Ozeri
parent 25b3242d8b
commit d3af8c1831
9 changed files with 187 additions and 102 deletions
+9 -5
View File
@@ -20,6 +20,7 @@ from vllm.v1.engine import (
EngineCoreRequest,
FinishReason,
)
from vllm.v1.metrics.stats import PrefillStats
from vllm.v1.structured_output.request import StructuredOutputRequest
from vllm.v1.utils import ConstantList
@@ -145,9 +146,6 @@ class Request:
self.all_token_ids = ConstantList(self._all_token_ids)
# trace_headers
self.trace_headers = trace_headers
# State
# The number of tokens with prefix cache hits.
self.num_cached_tokens = -1
# True if this request is scheduled as a non-final prefill chunk.
self.is_prefill_chunk = False
@@ -159,8 +157,7 @@ class Request:
# The number of times this request has been preempted by the scheduler.
self.num_preemptions = 0
# The number of tokens that have been computed remotely.
self.num_external_computed_tokens = 0
self.prefill_stats: PrefillStats | None = PrefillStats()
self.block_hashes: list[BlockHash] = []
# Store the block hasher without binding self to avoid creating a
@@ -278,6 +275,13 @@ class Request:
events, self.events = self.events, []
return events
def take_prefill_stats(self) -> PrefillStats | None:
if self.prefill_stats is None:
return None
prefill_stats = self.prefill_stats
self.prefill_stats = None
return prefill_stats
def __lt__(self, other: "Request") -> bool:
"""
Compare two requests based on priority, arrival time, and request ID.