121 lines
4.0 KiB
Go
121 lines
4.0 KiB
Go
package store
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
)
|
|
|
|
func (s *Store) ListModelCandidates(ctx context.Context, model string, modelType string) ([]RuntimeModelCandidate, error) {
|
|
rows, err := s.pool.Query(ctx, `
|
|
SELECT p.id::text, p.platform_key, p.name, p.provider, COALESCE(p.base_url, ''),
|
|
p.auth_type, p.credentials, p.config, p.default_pricing_mode,
|
|
p.default_discount_factor::float8, COALESCE(p.pricing_rule_set_id::text, ''),
|
|
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, ''),
|
|
m.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, ''),
|
|
m.retry_policy, m.rate_limit_policy
|
|
FROM platform_models m
|
|
JOIN integration_platforms p ON p.id = m.platform_id
|
|
LEFT JOIN base_model_catalog b ON b.id = m.base_model_id
|
|
LEFT JOIN runtime_client_states s
|
|
ON s.client_id = p.platform_key || ':' || m.model_type || ':' || m.model_name
|
|
WHERE p.status = 'enabled'
|
|
AND p.deleted_at IS NULL
|
|
AND m.enabled = true
|
|
AND m.model_type = $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
|
|
)
|
|
ORDER BY effective_priority ASC,
|
|
COALESCE(s.limiter_ratio, 0) ASC,
|
|
COALESCE(s.running_count, 0) ASC,
|
|
COALESCE(s.waiting_count, 0) ASC,
|
|
COALESCE(s.last_assigned_at, to_timestamp(0)) ASC,
|
|
m.created_at ASC`, model, modelType)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
defer rows.Close()
|
|
|
|
items := make([]RuntimeModelCandidate, 0)
|
|
for rows.Next() {
|
|
var item RuntimeModelCandidate
|
|
var credentials []byte
|
|
var platformConfig []byte
|
|
var platformRetryPolicy []byte
|
|
var platformRateLimitPolicy []byte
|
|
var capabilities []byte
|
|
var capabilityOverride []byte
|
|
var baseBilling []byte
|
|
var billing []byte
|
|
var billingOverride []byte
|
|
var modelRetryPolicy []byte
|
|
var modelRateLimitPolicy []byte
|
|
if err := rows.Scan(
|
|
&item.PlatformID,
|
|
&item.PlatformKey,
|
|
&item.PlatformName,
|
|
&item.Provider,
|
|
&item.BaseURL,
|
|
&item.AuthType,
|
|
&credentials,
|
|
&platformConfig,
|
|
&item.DefaultPricingMode,
|
|
&item.DefaultDiscountFactor,
|
|
&item.PlatformPricingRuleSetID,
|
|
&platformRetryPolicy,
|
|
&platformRateLimitPolicy,
|
|
&item.PlatformPriority,
|
|
&item.PlatformModelID,
|
|
&item.BaseModelID,
|
|
&item.CanonicalModelKey,
|
|
&item.ProviderModelName,
|
|
&item.ModelName,
|
|
&item.ModelAlias,
|
|
&item.ModelType,
|
|
&item.DisplayName,
|
|
&capabilities,
|
|
&capabilityOverride,
|
|
&baseBilling,
|
|
&billing,
|
|
&billingOverride,
|
|
&item.PricingMode,
|
|
&item.DiscountFactor,
|
|
&item.ModelPricingRuleSetID,
|
|
&modelRetryPolicy,
|
|
&modelRateLimitPolicy,
|
|
); err != nil {
|
|
return nil, err
|
|
}
|
|
item.Credentials = decodeObject(credentials)
|
|
item.PlatformConfig = decodeObject(platformConfig)
|
|
item.PlatformRetryPolicy = decodeObject(platformRetryPolicy)
|
|
item.PlatformRateLimitPolicy = decodeObject(platformRateLimitPolicy)
|
|
item.Capabilities = decodeObject(capabilities)
|
|
item.CapabilityOverride = decodeObject(capabilityOverride)
|
|
item.BaseBillingConfig = decodeObject(baseBilling)
|
|
item.BillingConfig = decodeObject(billing)
|
|
item.BillingConfigOverride = decodeObject(billingOverride)
|
|
item.ModelRetryPolicy = decodeObject(modelRetryPolicy)
|
|
item.ModelRateLimitPolicy = decodeObject(modelRateLimitPolicy)
|
|
item.ClientID = fmt.Sprintf("%s:%s:%s", item.PlatformKey, item.ModelType, item.ModelName)
|
|
item.QueueKey = item.ClientID
|
|
items = append(items, item)
|
|
}
|
|
if err := rows.Err(); err != nil {
|
|
return nil, err
|
|
}
|
|
if len(items) == 0 {
|
|
return nil, ErrNoModelCandidate
|
|
}
|
|
return items, nil
|
|
}
|