Lazy load requested model candidates

This commit is contained in:
fawney19
2026-05-03 20:56:31 +08:00
parent a24e4a793d
commit 6f00cabe96
13 changed files with 1047 additions and 225 deletions
@@ -2,6 +2,7 @@ use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
use tracing::warn;
use crate::ai_serving::planner::candidate_materialization::{
build_lazy_requested_model_execution_candidate_attempt_source_with_serving,
build_local_execution_candidate_attempt_source_with_serving,
mark_skipped_local_execution_candidate, mark_skipped_local_execution_candidate_with_extra_data,
mark_skipped_local_execution_candidate_with_failure_diagnostic,
@@ -240,80 +241,72 @@ pub(crate) async fn build_local_openai_responses_candidate_attempt_source<'a>(
input.required_capabilities.as_ref(),
LocalCandidatePersistencePolicyKind::OpenAiResponsesDecision,
);
let preselection = preselect_local_execution_candidates_with_serving(
planner_state,
spec_metadata.api_format,
&input.requested_model,
spec_metadata.require_streaming,
input.required_capabilities.as_ref(),
&input.auth_snapshot,
true,
LocalCandidatePreselectionKeyMode::ProviderEndpointKeyModelAndApiFormat,
)
.await?;
Ok(build_local_execution_candidate_attempt_source_with_serving(
planner_state,
trace_id,
spec_metadata.api_format,
Some(&input.requested_model),
Some(&input.auth_snapshot),
input.required_capabilities.as_ref(),
sticky_session_token.as_deref(),
input.request_auth_channel.as_deref(),
persistence_policy,
preselection.candidates,
preselection.skipped_candidates,
LocalCandidateResolutionMode::Standard,
|eligible| {
let provider_api_format = eligible.provider_api_format.clone();
let (execution_strategy, conversion_mode) = ai_local_execution_contract_for_formats(
spec_metadata.api_format,
&provider_api_format,
);
Some(build_local_execution_candidate_contract_metadata(
LocalExecutionCandidateMetadataParts {
eligible,
provider_api_format: provider_api_format.as_str(),
client_api_format: spec_metadata.api_format,
extra_fields: serde_json::Map::new(),
},
execution_strategy,
conversion_mode,
eligible.candidate.endpoint_api_format.as_str(),
))
},
|mut skipped_candidate| {
let provider_api_format = skipped_candidate
.transport
.as_ref()
.map(|transport| transport.endpoint.api_format.trim().to_ascii_lowercase())
.unwrap_or_else(|| {
skipped_candidate
.candidate
.endpoint_api_format
.trim()
.to_ascii_lowercase()
});
let (execution_strategy, conversion_mode) = ai_local_execution_contract_for_formats(
spec_metadata.api_format,
&provider_api_format,
);
skipped_candidate.extra_data = Some(
build_local_execution_candidate_contract_metadata_for_candidate(
&skipped_candidate.candidate,
skipped_candidate.transport_ref(),
provider_api_format.as_str(),
Ok(
build_lazy_requested_model_execution_candidate_attempt_source_with_serving(
planner_state,
trace_id,
spec_metadata.api_format,
&input.requested_model,
spec_metadata.require_streaming,
&input.auth_snapshot,
input.required_capabilities.as_ref(),
sticky_session_token.as_deref(),
input.request_auth_channel.as_deref(),
persistence_policy,
true,
LocalCandidatePreselectionKeyMode::ProviderEndpointKeyModelAndApiFormat,
LocalCandidateResolutionMode::Standard,
move |eligible| {
let provider_api_format = eligible.provider_api_format.clone();
let (execution_strategy, conversion_mode) = ai_local_execution_contract_for_formats(
spec_metadata.api_format,
serde_json::Map::new(),
&provider_api_format,
);
Some(build_local_execution_candidate_contract_metadata(
LocalExecutionCandidateMetadataParts {
eligible,
provider_api_format: provider_api_format.as_str(),
client_api_format: spec_metadata.api_format,
extra_fields: serde_json::Map::new(),
},
execution_strategy,
conversion_mode,
provider_api_format.as_str(),
),
);
skipped_candidate
},
eligible.candidate.endpoint_api_format.as_str(),
))
},
move |mut skipped_candidate| {
let provider_api_format = skipped_candidate
.transport
.as_ref()
.map(|transport| transport.endpoint.api_format.trim().to_ascii_lowercase())
.unwrap_or_else(|| {
skipped_candidate
.candidate
.endpoint_api_format
.trim()
.to_ascii_lowercase()
});
let (execution_strategy, conversion_mode) = ai_local_execution_contract_for_formats(
spec_metadata.api_format,
&provider_api_format,
);
skipped_candidate.extra_data = Some(
build_local_execution_candidate_contract_metadata_for_candidate(
&skipped_candidate.candidate,
skipped_candidate.transport_ref(),
provider_api_format.as_str(),
spec_metadata.api_format,
serde_json::Map::new(),
execution_strategy,
conversion_mode,
provider_api_format.as_str(),
),
);
skipped_candidate
},
)
.await,
)
.await)
}
pub(crate) async fn mark_skipped_local_openai_responses_candidate(