feat: add runner failover rules and cache affinity

2026-06-28 20:50:23 +08:00
parent 41bb3a6525
commit 0e675e259c
31 changed files with 2939 additions and 470 deletions
@@ -0,0 +1,46 @@
+ALTER TABLE IF EXISTS gateway_runner_policies
+  ADD COLUMN IF NOT EXISTS cache_affinity_policy jsonb NOT NULL DEFAULT '{
+    "enabled": true,
+    "modelTypes": ["chat", "text_generate", "responses"],
+    "minSamples": 3,
+    "minInputTokens": 512,
+    "staleAfterSeconds": 86400,
+    "maxPriorityBoost": 20,
+    "emaAlpha": 0.35
+  }'::jsonb;
+
+UPDATE gateway_runner_policies
+SET cache_affinity_policy = COALESCE(NULLIF(cache_affinity_policy, '{}'::jsonb), '{
+      "enabled": true,
+      "modelTypes": ["chat", "text_generate", "responses"],
+      "minSamples": 3,
+      "minInputTokens": 512,
+      "staleAfterSeconds": 86400,
+      "maxPriorityBoost": 20,
+      "emaAlpha": 0.35
+    }'::jsonb),
+    metadata = COALESCE(metadata, '{}'::jsonb) || jsonb_build_object('cacheAffinityPolicy', '0055_cache_affinity_routing'),
+    updated_at = now();
+
+CREATE TABLE IF NOT EXISTS gateway_cache_affinity_stats (
+  client_id text NOT NULL,
+  cache_affinity_key text NOT NULL,
+  platform_id uuid REFERENCES integration_platforms(id) ON DELETE SET NULL,
+  platform_model_id uuid REFERENCES platform_models(id) ON DELETE SET NULL,
+  model_type text NOT NULL,
+  request_count bigint NOT NULL DEFAULT 0,
+  input_tokens bigint NOT NULL DEFAULT 0,
+  cached_input_tokens bigint NOT NULL DEFAULT 0,
+  ema_hit_ratio numeric NOT NULL DEFAULT 0,
+  last_hit_ratio numeric NOT NULL DEFAULT 0,
+  last_observed_at timestamptz NOT NULL DEFAULT now(),
+  created_at timestamptz NOT NULL DEFAULT now(),
+  updated_at timestamptz NOT NULL DEFAULT now(),
+  PRIMARY KEY (client_id, cache_affinity_key)
+);
+
+CREATE INDEX IF NOT EXISTS idx_gateway_cache_affinity_key_observed
+  ON gateway_cache_affinity_stats(cache_affinity_key, last_observed_at DESC);
+
+CREATE INDEX IF NOT EXISTS idx_gateway_cache_affinity_platform_model
+  ON gateway_cache_affinity_stats(platform_model_id, cache_affinity_key);