mixdog 0.9.1 → 0.9.3

package/src/runtime/agent/orchestrator/providers/retry-classifier.mjs

@@ -126,7 +126,7 @@ export function classifyError(err) {
     || code === 'EAI_AGAIN' || code === 'ENOTFOUND' || code === 'EAI_NODATA'
     || code === 'ECONNREFUSED' || code === 'ENETUNREACH' || code === 'EHOSTUNREACH'
     || code === 'EPIPE'
-    || code === 'EPROVIDERTIMEOUT' || code === 'EGEMINITIMEOUT'
+    || code === 'EPROVIDERTIMEOUT' || code === 'EGEMINITIMEOUT' || code === 'ESTREAMSTALL'
     || code === 'EWSACQUIRETIMEOUT') {
     return 'transient'
   }
@@ -232,12 +232,11 @@ export function jitterDelayMs(ms, ratio = PROVIDER_RETRY_JITTER_RATIO, mode = 's
 }
 
 // ── Shared network-resilience interface ──────────────────────────────────────
-// One home for the logic that used to be copied per provider: mid-stream
-// classifier (WS + SSE), transport fallback predicate, stream-safety stamp
-// latches, abort-aware sleep, handshake classifier, and the retry-budget table.
-// Provider differences are passed as ARGUMENTS (policy objects), never branched
-// on a hardcoded provider name. Behavior is preserved EXACTLY — each path below
-// is the relocated original, selected by policy.mode.
+// One home for the logic shared across providers: mid-stream classifier
+// (WS + SSE), transport fallback predicate, stream-safety stamp latches,
+// abort-aware sleep, handshake classifier, and the retry-budget table.
+// Provider differences are passed as ARGUMENTS (policy objects), never
+// branched on a hardcoded provider name.
 
 // F) Retry-budget profiles as DATA. The numbers live ONLY here now.
 //    ws.transientCloseRetries (4) — ws_1006 / ws_1011 connection-loss buckets.
@@ -260,8 +259,8 @@ function _midstreamLimitFor(classifier, policy) {
   return policy.defaultRetries
 }
 
-// WS gates each classifier against its own budget (mirrors the old
-// _allowMidstreamRetry). SSE applies a single top-of-function budget gate and
+// WS gates each classifier against its own budget. SSE applies a single
+// top-of-function budget gate and
 // then returns raw classifier strings, so perClassifierGate:false returns the
 // classifier unconditionally here.
 function _allowMidstream(classifier, attemptIndex, policy) {
@@ -282,7 +281,7 @@ export function classifyMidstreamError(err, signals, policy = {}) {
 
 // Verbatim relocation of openai-oauth-ws's _classifyMidstreamError. `signals`
 // carries sawResponseCreated / sawCompleted / emittedText / emittedToolCall /
-// wsCloseCode / firstByteTimeout / firstMeaningfulTimeout / wsSendFailed /
+// wsCloseCode / firstByteTimeout / wsSendFailed /
 // userAbort / watchdogAbort / responseFailedPayload exactly as before.
 function _classifyMidstreamWs(err, state, attemptIndex, policy) {
   if (state.sawCompleted) return null
@@ -294,9 +293,6 @@ function _classifyMidstreamWs(err, state, attemptIndex, policy) {
   if (state.firstByteTimeout || err?.firstByteTimeout) {
     return _allowMidstream('first_byte_timeout', attemptIndex, policy)
   }
-  if (state.firstMeaningfulTimeout || err?.firstMeaningfulTimeout) {
-    return _allowMidstream('first_meaningful_timeout', attemptIndex, policy)
-  }
   if (err?.wsSendFailed || state.wsSendFailed) {
     return _allowMidstream('ws_send_failed', attemptIndex, policy)
   }
@@ -315,7 +311,12 @@ function _classifyMidstreamWs(err, state, attemptIndex, policy) {
 
   const name = err?.name || ''
   if (name === 'AgentStallAbortError') return _allowMidstream('agent_stall', attemptIndex, policy)
-  if (name === 'StreamStalledAbortError') return _allowMidstream('stream_stalled', attemptIndex, policy)
+  if (name === 'StreamStalledAbortError' || name === 'StreamStalledError' || err?.code === 'ESTREAMSTALL' || err?.streamStalled === true) {
+    // A stall that fired AFTER a tool call was emitted is unsafe to replay
+    // (double side-effect); surface it as terminal so the turn is not retried.
+    if (err?.unsafeToRetry === true) return null
+    return _allowMidstream('stream_stalled', attemptIndex, policy)
+  }
 
   if (state.watchdogAbort === 'AgentStallAbortError') return _allowMidstream('agent_stall', attemptIndex, policy)
   if (state.watchdogAbort === 'StreamStalledAbortError') return _allowMidstream('stream_stalled', attemptIndex, policy)
@@ -357,7 +358,12 @@ function _classifyMidstreamSse(err, state, attemptIndex, policy) {
 
   const name = err?.name || ''
   if (name === 'AgentStallAbortError') return 'agent_stall'
-  if (name === 'StreamStalledAbortError') return 'stream_stalled'
+  if (name === 'StreamStalledAbortError' || name === 'StreamStalledError' || err?.code === 'ESTREAMSTALL' || err?.streamStalled === true) {
+    // A stall AFTER a tool emit is unsafe to replay (double side-effect) →
+    // terminal (null), no mid-stream retry. Otherwise route to stream_stalled.
+    if (err?.unsafeToRetry === true) return null
+    return 'stream_stalled'
+  }
   if (state.watchdogAbort === 'AgentStallAbortError') return 'agent_stall'
   if (state.watchdogAbort === 'StreamStalledAbortError') return 'stream_stalled'
 
@@ -379,7 +385,7 @@ function _classifyMidstreamSse(err, state, attemptIndex, policy) {
 //    env-flag check (caller computes the flag and passes it).
 const TRANSPORT_FALLBACK_CLASSIFIERS = new Set([
   'timeout', 'reset', 'dns', 'refused', 'network', 'acquire_timeout', 'http_5xx',
-  'first_byte_timeout', 'first_meaningful_timeout',
+  'first_byte_timeout',
   'ws_1006', 'ws_1011', 'ws_1012', 'ws_1000', 'ws_4000', 'agent_stall', 'stream_stalled',
   'response_failed_disconnected', 'response_failed_network', 'response_failed_auth_expired',
   'ws_send_failed',
@@ -403,7 +409,6 @@ export function shouldFallbackTransport(err, { signal, enabled = true } = {}) {
   if (TRANSPORT_FALLBACK_CLASSIFIERS.has(classifier)) return true
   if (/^http_5\d\d$/.test(classifier)) return true
   if (err?.firstByteTimeout) return true
-  if (err?.firstMeaningfulTimeout) return true
   const msg = String(err?.message || '')
   return /opening handshake has timed out|socket hang up|acquire timed out|no first server event|no meaningful output/i.test(msg)
 }