Unify candidate ranking pipeline

This commit is contained in:
fawney19
2026-04-27 17:24:31 +08:00
parent 9b866a6d17
commit 3b542434a2
46 changed files with 3635 additions and 1914 deletions
@@ -2,9 +2,10 @@ use aether_data::DataLayerError;
use aether_data_contracts::repository::candidate_selection::StoredMinimalCandidateSelectionRow;
use aether_scheduler_core::{
auth_constraints_allow_api_format, build_minimal_candidate_selection,
collect_global_model_names_for_required_capability, normalize_api_format,
resolve_requested_global_model_name, BuildMinimalCandidateSelectionInput,
SchedulerAuthConstraints, SchedulerMinimalCandidateSelectionCandidate, SchedulerPriorityMode,
collect_global_model_names_for_required_capability, enumerate_minimal_candidate_selection,
normalize_api_format, resolve_requested_global_model_name, row_supports_requested_model,
BuildMinimalCandidateSelectionInput, SchedulerAuthConstraints,
SchedulerMinimalCandidateSelectionCandidate, SchedulerPriorityMode,
};
use async_trait::async_trait;
use std::collections::BTreeSet;
@@ -30,31 +31,24 @@ pub(crate) async fn read_requested_model_rows(
api_format: &str,
requested_model_name: &str,
) -> Result<Option<(String, Vec<StoredMinimalCandidateSelectionRow>)>, DataLayerError> {
let exact_rows = state
.read_minimal_candidate_selection_rows_for_api_format_and_global_model(
api_format,
requested_model_name,
)
.await?;
if !exact_rows.is_empty() {
return Ok(Some((requested_model_name.to_string(), exact_rows)));
}
let rows = state
.read_minimal_candidate_selection_rows_for_api_format(api_format)
.await?;
let rows = rows
.into_iter()
.filter(|row| row_supports_requested_model(row, requested_model_name, api_format))
.collect::<Vec<_>>();
if rows.is_empty() {
return Ok(None);
}
let Some(resolved_global_model_name) =
resolve_requested_global_model_name(&rows, requested_model_name, api_format)
else {
return Ok(None);
};
Ok(Some((
resolved_global_model_name.clone(),
rows.into_iter()
.filter(|row| row.global_model_name == resolved_global_model_name)
.collect(),
)))
Ok(Some((resolved_global_model_name, rows)))
}
pub(crate) async fn read_minimal_candidate_selection(
@@ -180,6 +174,45 @@ pub(crate) async fn read_minimal_candidate_selection_with_priority_mode_and_affi
})
}
pub(crate) async fn enumerate_minimal_candidate_selection_with_required_capabilities(
state: &(impl MinimalCandidateSelectionRowSource + Sync),
api_format: &str,
requested_model_name: &str,
require_streaming: bool,
auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
required_capabilities: Option<&serde_json::Value>,
) -> Result<Vec<SchedulerMinimalCandidateSelectionCandidate>, DataLayerError> {
let normalized_api_format = normalize_api_format(api_format);
if normalized_api_format.is_empty() {
return Ok(Vec::new());
}
if !auth_constraints_allow_api_format(
auth_snapshot.map(auth_snapshot_constraints).as_ref(),
&normalized_api_format,
) {
return Ok(Vec::new());
}
let Some((resolved_global_model_name, rows)) =
read_requested_model_rows(state, &normalized_api_format, requested_model_name).await?
else {
return Ok(Vec::new());
};
let auth_constraints = auth_snapshot.map(auth_snapshot_constraints);
enumerate_minimal_candidate_selection(BuildMinimalCandidateSelectionInput {
rows,
normalized_api_format: &normalized_api_format,
requested_model_name,
resolved_global_model_name: resolved_global_model_name.as_str(),
require_streaming,
required_capabilities,
auth_constraints: auth_constraints.as_ref(),
affinity_key: None,
priority_mode: SchedulerPriorityMode::Provider,
})
}
pub(crate) async fn read_global_model_names_for_required_capability(
state: &(impl MinimalCandidateSelectionRowSource + Sync),
api_format: &str,