@adaptic/lumic-utils 1.0.23 → 1.0.25

package/dist/{index-CWoI2dXN.js → index-Bvc61lYS.js}

@@ -1983,11 +1983,19 @@ const DEFAULT_RETRY_CONFIG = {
  * @throws The last error encountered if all retry attempts fail or if a non-retryable error occurs
  */
 async function withRetry(fn, config = {}, label = 'unknown') {
-    const { maxRetries, baseDelayMs, maxDelayMs, retryableErrors, circuitBreaker } = {
+    const { maxRetries, baseDelayMs, maxDelayMs, retryableErrors, circuitBreaker, signal } = {
         ...DEFAULT_RETRY_CONFIG,
         ...config,
     };
     for (let attempt = 0; attempt <= maxRetries; attempt++) {
+        // Short-circuit on cancellation BEFORE starting a fresh attempt so
+        // the caller does not pay the cost of issuing a request whose result
+        // they have already discarded.
+        if (signal?.aborted) {
+            throw signal.reason instanceof Error
+                ? signal.reason
+                : new Error('Aborted');
+        }
         try {
             // If a circuit breaker is provided, execute through it
             const result = circuitBreaker
@@ -2000,6 +2008,13 @@ async function withRetry(fn, config = {}, label = 'unknown') {
             if (error instanceof CircuitBreakerOpenError) {
                 throw error;
             }
+            // Do not retry on cooperative cancellation — the caller has
+            // explicitly asked us to stop.
+            if (signal?.aborted) {
+                throw signal.reason instanceof Error
+                    ? signal.reason
+                    : new Error('Aborted');
+            }
             // Don't retry if we've exhausted all attempts
             if (attempt === maxRetries) {
                 throw error;
@@ -2013,7 +2028,26 @@ async function withRetry(fn, config = {}, label = 'unknown') {
             const jitter = Math.random() * 1000;
             const delay = Math.min(exponentialDelay + jitter, maxDelayMs);
             getLumicLogger().warn(`[${label}] Attempt ${attempt + 1} failed, retrying in ${Math.round(delay)}ms`);
-            await new Promise((resolve) => setTimeout(resolve, delay));
+            // Sleep but bail early if signal aborts mid-backoff.
+            await new Promise((resolve, reject) => {
+                const timer = setTimeout(() => {
+                    if (signal)
+                        signal.removeEventListener('abort', onAbort);
+                    resolve();
+                }, delay);
+                const onAbort = () => {
+                    clearTimeout(timer);
+                    reject(signal?.reason instanceof Error ? signal.reason : new Error('Aborted'));
+                };
+                if (signal) {
+                    if (signal.aborted) {
+                        clearTimeout(timer);
+                        reject(signal.reason instanceof Error ? signal.reason : new Error('Aborted'));
+                        return;
+                    }
+                    signal.addEventListener('abort', onAbort, { once: true });
+                }
+            });
         }
     }
     // This should never be reached, but TypeScript needs it for type safety
@@ -2506,11 +2540,20 @@ async function createCompletion(content, responseFormat, options = DEFAULT_OPTIO
     }
     let completion;
     try {
-        completion = await withRetry(() => openai.chat.completions.create(queryOptions), {
+        completion = await withRetry(
+        // Pass the AbortSignal into the OpenAI SDK request-options object.
+        // The OpenAI SDK honours `signal` to cancel the in-flight fetch
+        // immediately when it fires — this is what makes caller-driven
+        // cancel-and-restart (e.g. signal-monitoring's materiality
+        // re-trigger path in the engine) actually release the HTTP
+        // connection and response-body buffer rather than just discarding
+        // the resolved value at the engine boundary.
+        () => openai.chat.completions.create(queryOptions, options.signal ? { signal: options.signal } : undefined), {
             maxRetries: 3,
             baseDelayMs: 2000,
             maxDelayMs: 30000,
             retryableErrors: isRetryableLLMError,
+            signal: options.signal,
         }, `OpenAI:${normalizedModel}`);
     }
     catch (error) {
@@ -2668,19 +2711,20 @@ const makeResponsesAPICall = async (input, options = {}) => {
         apiKey: apiKey,
         timeout: options.timeout ?? LLM_TIMEOUT_MS,
     });
-    // Remove apiKey, model, and timeout from options before creating request body
-    const { apiKey: _, model: __, timeout: ___, ...cleanOptions } = options;
+    // Remove apiKey, model, timeout, and signal from options before creating request body
+    const { apiKey: _, model: __, timeout: ___, signal: ____, ...cleanOptions } = options;
     const requestBody = {
         model: normalizedModel,
         input,
         ...cleanOptions,
     };
     // Make the API call to the Responses endpoint
-    const response = await withRetry(() => openai.responses.create(requestBody), {
+    const response = await withRetry(() => openai.responses.create(requestBody, options.signal ? { signal: options.signal } : undefined), {
         maxRetries: 3,
         baseDelayMs: 2000,
         maxDelayMs: 30000,
         retryableErrors: isRetryableLLMError,
+        signal: options.signal,
     }, `OpenAI-Responses:${normalizedModel}`);
     // Responses API exposes cached input tokens under `input_tokens_details.cached_tokens`
     // (the equivalent of Chat Completions' `prompt_tokens_details.cached_tokens`).
@@ -8253,23 +8297,32 @@ async function makeAnthropicCall(content, responseFormat = 'text', options = {})
     if (systemParts.length > 0) {
         requestParams.system = systemParts.join('\n\n');
     }
-    // Add temperature if specified
-    if (options.temperature !== undefined) {
-        requestParams.temperature = options.temperature;
-    }
-    if (options.top_p !== undefined) {
-        requestParams.top_p = options.top_p;
-    }
+    // Sampling controls are intentionally disabled for the Anthropic provider.
+    //
+    // The claude-*-4.x model family rejects any request that specifies BOTH
+    // `temperature` and `top_p` ("`temperature` and `top_p` cannot both be
+    // specified for this model. Please use only one." — HTTP 400), which aborts
+    // the entire caller flow (e.g. the engine's account decision session).
+    // Callers across the org pass both knobs generically, so rather than make
+    // each one provider-aware we drop both here and let Anthropic use its model
+    // defaults. `options.temperature` / `options.top_p` are intentionally not
+    // forwarded. This is the Anthropic adapter only — the OpenAI, DeepSeek, and
+    // OpenAI-compatible adapters still honor `temperature` and `top_p`.
     // Translate and add tools if present
     if (options.tools && options.tools.length > 0) {
         requestParams.tools = translateToolsToAnthropic(options.tools);
     }
     try {
-        const response = await withRetry(() => client.messages.create(requestParams), {
+        const response = await withRetry(
+        // Pass AbortSignal into the Anthropic SDK request-options object
+        // so caller-driven cancellation interrupts the in-flight HTTP
+        // request immediately. See LLMOptions.signal JSDoc for context.
+        () => client.messages.create(requestParams, options.signal ? { signal: options.signal } : undefined), {
             maxRetries: 3,
             baseDelayMs: 2000,
             maxDelayMs: 30000,
             retryableErrors: isRetryableAnthropicError,
+            signal: options.signal,
         }, `Anthropic:${model}`);
         const inputTokens = response.usage.input_tokens;
         const outputTokens = response.usage.output_tokens;
@@ -8462,11 +8515,16 @@ async function makeOpenAICompatibleCall(content, responseFormat = 'text', option
         queryOptions.tools = options.tools;
     }
     try {
-        const completion = await withRetry(() => openai.chat.completions.create(queryOptions), {
+        const completion = await withRetry(
+        // Pass AbortSignal into the SDK request-options object so
+        // caller-driven cancellation interrupts the in-flight HTTP
+        // request immediately. See LLMOptions.signal JSDoc for context.
+        () => openai.chat.completions.create(queryOptions, options.signal ? { signal: options.signal } : undefined), {
             maxRetries: 3,
             baseDelayMs: 2000,
             maxDelayMs: 30000,
             retryableErrors: isRetryableError,
+            signal: options.signal,
         }, `${providerConfig.name}:${model}`);
         const promptTokens = completion.usage?.prompt_tokens || 0;
         const completionTokens = completion.usage?.completion_tokens || 0;
@@ -22997,11 +23055,11 @@ let poolConfig = DEFAULT_POOL_CONFIG;
 async function loadApolloModules() {
     if (typeof window === "undefined" || process.env.AWS_EXECUTION_ENV) {
         // Server-side (or Lambda): load the CommonJS‑based implementation.
-        return (await import('./apollo-client.server-CJ6iRLa2.js'));
+        return (await import('./apollo-client.server-CXLiLmSz.js'));
     }
     else {
         // Client-side: load the ESM‑based implementation.
-        return (await import('./apollo-client.client-DrSy1wz-.js'));
+        return (await import('./apollo-client.client-DLW-p46Y.js'));
     }
 }
 /**
@@ -81525,4 +81583,4 @@ const lumic = {
 };
 
 export { GraphQLInterfaceType as $, print as A, getNamedType as B, isInputType as C, isRequiredArgument as D, isNamedType as E, GraphQLError as F, GraphQLNonNull as G, isOutputType as H, isRequiredInputField as I, isCompositeType as J, Kind as K, getNullableType as L, getEnterLeaveForKind as M, isNode as N, OperationTypeNode as O, didYouMean as P, naturalCompare as Q, suggestionList as R, specifiedScalarTypes as S, keyMap as T, isType as U, isNullableType as V, visit as W, visitInParallel as X, keyValMap as Y, assertObjectType as Z, GraphQLScalarType as _, isListType as a, validateGoogleSheetsRange as a$, GraphQLUnionType as a0, GraphQLInputObjectType as a1, assertNullableType as a2, assertInterfaceType as a3, mapValue as a4, isSpecifiedScalarType as a5, isPrintableAsBlockString as a6, printBlockString as a7, BREAK as a8, GRAPHQL_MAX_INT as a9, printSourceLocation as aA, resolveObjMapThunk as aB, resolveReadonlyArrayThunk as aC, valueFromASTUntyped as aD, version$4 as aE, versionInfo as aF, getAugmentedNamespace as aG, isDigit$1 as aH, isNameStart as aI, dedentBlockStringLines as aJ, isNameContinue as aK, setLumicLogger as aL, getLumicLogger as aM, sanitizeForLog as aN, sanitizeError as aO, sanitizeAWSAuth as aP, sanitizeObject as aQ, getSecrets as aR, resetSecrets as aS, requireSecret as aT, withRetry as aU, CircuitBreaker as aV, CircuitBreakerState as aW, CircuitBreakerOpenError as aX, DEFAULT_CIRCUIT_BREAKER_CONFIG as aY, validateSlackChannel as aZ, validateS3Key as a_, GRAPHQL_MIN_INT as aa, GraphQLFloat as ab, GraphQLInt as ac, Location as ad, Token as ae, assertAbstractType as af, assertCompositeType as ag, assertEnumType as ah, assertEnumValueName as ai, assertInputObjectType as aj, assertInputType as ak, assertLeafType as al, assertListType as am, assertNamedType as an, assertNonNullType as ao, assertOutputType as ap, assertScalarType as aq, assertType as ar, assertUnionType as as, assertWrappingType as at, formatError as au, getLocation as av, getVisitFn as aw, isWrappingType as ax, printError as ay, printLocation as az, isAbstractType as b, PDFError as b$, LLMCostTracker as b0, getLLMCostTracker as b1, setLLMCostTracker as b2, resetLLMCostTracker as b3, setMetricsCollector as b4, getMetricsCollector as b5, resetMetricsCollector as b6, withMetrics as b7, generateCorrelationId as b8, getCorrelationId as b9, openAIChatCompletionSchema as bA, openAIImageResponseSchema as bB, validateOpenAIChatCompletion as bC, safeValidateOpenAIChatCompletion as bD, perplexityResponseSchema as bE, validatePerplexityResponse as bF, safeValidatePerplexityResponse as bG, googleSheetsValueRangeSchema as bH, validateGoogleSheetsResponse as bI, safeValidateGoogleSheetsResponse as bJ, s3ListObjectsSchema as bK, s3GetObjectSchema as bL, lambdaInvokeResponseSchema as bM, validateS3ListObjects as bN, safeValidateS3ListObjects as bO, validateLambdaResponse as bP, safeValidateLambdaResponse as bQ, createValidator as bR, createSafeValidator as bS, LumicError as bT, SlackError as bU, LLMError as bV, AWSLambdaError as bW, AWSS3Error as bX, GoogleSheetsError as bY, PerplexityError as bZ, JsonParseError as b_, getCorrelationContext as ba, withCorrelationId as bb, getCorrelationHeaders as bc, TokenBucketRateLimiter as bd, RATE_LIMIT_PROFILES as be, getRateLimiter as bf, resetAllRateLimiters as bg, withRateLimit as bh, checkIntegrationHealth as bi, SUPPORTED_MODELS as bj, isValidModel as bk, getModelCapabilities as bl, getModelProvider as bm, MODEL_ALIASES as bn, OPENAI_COMPATIBLE_PROVIDERS as bo, PROVIDER_DEFAULT_MODELS as bp, LLM_DEFAULT_PROVIDER as bq, LLM_MINI_PROVIDER as br, LLM_NORMAL_PROVIDER as bs, LLM_ADVANCED_PROVIDER as bt, LLM_PROVIDER as bu, LLM_MODEL_MINI as bv, LLM_MODEL_NORMAL as bw, LLM_MODEL_ADVANCED as bx, makeAnthropicCall as by, makeOpenAICompatibleCall as bz, isInterfaceType as c, ZipError as c0, SLACK_TIMEOUT_MS as c1, PERPLEXITY_TIMEOUT_MS as c2, GOOGLE_SHEETS_TIMEOUT_MS as c3, LLM_TIMEOUT_MS as c4, AWS_LAMBDA_TIMEOUT_MS as c5, AWS_S3_TIMEOUT_MS as c6, OPENWEATHER_TIMEOUT_MS as c7, GENERIC_FETCH_TIMEOUT_MS as c8, isObjectType as d, assertName as e, devAssert as f, isObjectLike as g, defineArguments as h, isNonNullType as i, argsToArgsConfig as j, GraphQLBoolean as k, lumic as l, GraphQLString as m, instanceOf as n, inspect as o, isInputObjectType as p, isLeafType as q, isEnumType as r, GraphQLID as s, toObjMap as t, invariant as u, GraphQLObjectType as v, GraphQLEnumType as w, GraphQLList as x, isScalarType as y, isUnionType as z };
-//# sourceMappingURL=index-CWoI2dXN.js.map
+//# sourceMappingURL=index-Bvc61lYS.js.map