feat: improve model catalog aggregation

This commit is contained in:
2026-05-11 17:44:57 +08:00
parent ec87816c95
commit 0431cb8157
41 changed files with 4745 additions and 550 deletions
+13 -7
View File
@@ -17,7 +17,7 @@ SELECT p.id::text, p.platform_key, p.name, p.provider,
p.retry_policy, p.rate_limit_policy,
COALESCE(p.dynamic_priority, p.priority) AS effective_priority,
m.id::text, COALESCE(m.base_model_id::text, ''), COALESCE(b.canonical_model_key, ''),
COALESCE(b.provider_model_name, ''), m.model_name, COALESCE(m.model_alias, ''),
COALESCE(NULLIF(m.provider_model_name, ''), m.model_name), m.model_name, COALESCE(m.model_alias, ''),
$2 AS requested_model_type, m.display_name, m.capabilities, m.capability_override,
COALESCE(b.base_billing_config, '{}'::jsonb), m.billing_config, m.billing_config_override,
m.pricing_mode, COALESCE(m.discount_factor, 0)::float8, COALESCE(m.pricing_rule_set_id::text, ''),
@@ -32,17 +32,22 @@ LEFT JOIN model_catalog_providers cp ON cp.provider_key = p.provider OR cp.provi
LEFT JOIN base_model_catalog b ON b.id = m.base_model_id
LEFT JOIN model_runtime_policy_sets rp ON rp.id = COALESCE(m.runtime_policy_set_id, b.runtime_policy_set_id)
LEFT JOIN runtime_client_states s
ON s.client_id = p.platform_key || ':' || $2 || ':' || m.model_name
ON s.client_id = p.platform_key || ':' || $2 || ':' || COALESCE(NULLIF(m.provider_model_name, ''), m.model_name)
WHERE p.status = 'enabled'
AND p.deleted_at IS NULL
AND m.enabled = true
AND m.model_type @> jsonb_build_array($2)
AND (p.cooldown_until IS NULL OR p.cooldown_until <= now())
AND (
m.model_name = $1
OR m.model_alias = $1
OR b.canonical_model_key = $1
OR b.provider_model_name = $1
(COALESCE(m.model_alias, '') <> '' AND m.model_alias = $1)
OR (
COALESCE(m.model_alias, '') = ''
AND (
m.model_name = $1
OR b.canonical_model_key = $1
OR b.provider_model_name = $1
)
)
)
ORDER BY effective_priority ASC,
COALESCE(s.limiter_ratio, 0) ASC,
@@ -137,7 +142,8 @@ ORDER BY effective_priority ASC,
item.RuntimeRateLimitPolicy = decodeObject(runtimeRateLimitPolicy)
item.AutoDisablePolicy = decodeObject(autoDisablePolicy)
item.DegradePolicy = decodeObject(degradePolicy)
item.ClientID = fmt.Sprintf("%s:%s:%s", item.PlatformKey, item.ModelType, item.ModelName)
upstreamModelName := firstNonEmpty(item.ProviderModelName, item.ModelName)
item.ClientID = fmt.Sprintf("%s:%s:%s", item.PlatformKey, item.ModelType, upstreamModelName)
item.QueueKey = item.ClientID
items = append(items, item)
}