feat(core): Add enableTruncation option to LangChain integration (#20182)

This PR adds an `enableTruncation` option to the LangChain integration
that allows users to disable input message truncation. It defaults to
`true` to preserve existing behavior.

Also fixes missing truncation for LLM string prompts in
extractLLMRequestAttributes and refactors to use the shared
getTruncatedJsonString/getJsonString utilities.

Closes: #20138

---------

Co-authored-by: Nicolas Hrubec <nicolas.hrubec@outlook.com>

Author: andreiborza

dev-packages/node-integration-tests/suites/tracing/langchain/instrument-no-truncation.mjs

@@ -0,0 +1,24 @@
+import * as Sentry from '@sentry/node';
+import { loggingTransport } from '@sentry-internal/node-integration-tests';
+
+Sentry.init({
+  dsn: 'https://public@dsn.ingest.sentry.io/1337',
+  release: '1.0',
+  tracesSampleRate: 1.0,
+  sendDefaultPii: true,
+  transport: loggingTransport,
+  integrations: [
+    Sentry.langChainIntegration({
+      enableTruncation: false,
+      recordInputs: true,
+      recordOutputs: true,
+    }),
+  ],
+  beforeSendTransaction: event => {
+    // Filter out mock express server transactions
+    if (event.transaction.includes('/v1/messages') || event.transaction.includes('/v1/embeddings')) {
+      return null;
+    }
+    return event;
+  },
+});

dev-packages/node-integration-tests/suites/tracing/langchain/scenario-no-truncation.mjs

@@ -0,0 +1,56 @@
+import { ChatAnthropic } from '@langchain/anthropic';
+import * as Sentry from '@sentry/node';
+import express from 'express';
+
+function startMockAnthropicServer() {
+  const app = express();
+  app.use(express.json({ limit: '10mb' }));
+
+  app.post('/v1/messages', (req, res) => {
+    res.json({
+      id: 'msg_no_truncation_test',
+      type: 'message',
+      role: 'assistant',
+      content: [{ type: 'text', text: 'Response' }],
+      model: req.body.model,
+      stop_reason: 'end_turn',
+      stop_sequence: null,
+      usage: { input_tokens: 10, output_tokens: 5 },
+    });
+  });
+
+  return new Promise(resolve => {
+    const server = app.listen(0, () => {
+      resolve(server);
+    });
+  });
+}
+
+async function run() {
+  const server = await startMockAnthropicServer();
+  const baseUrl = `http://localhost:${server.address().port}`;
+
+  await Sentry.startSpan({ op: 'function', name: 'main' }, async () => {
+    const model = new ChatAnthropic({
+      model: 'claude-3-5-sonnet-20241022',
+      apiKey: 'mock-api-key',
+      clientOptions: {
+        baseURL: baseUrl,
+      },
+    });
+
+    // Long content that would normally be truncated
+    const longContent = 'A'.repeat(50_000);
+    await model.invoke([
+      { role: 'user', content: longContent },
+      { role: 'assistant', content: 'Some reply' },
+      { role: 'user', content: 'Follow-up question' },
+    ]);
+  });
+
+  await Sentry.flush(2000);
+
+  server.close();
+}
+
+run();

dev-packages/node-integration-tests/suites/tracing/langchain/test.ts

@@ -549,4 +549,37 @@ describe('LangChain integration', () => {
         .completed();
     });
   });
+
+  const longContent = 'A'.repeat(50_000);
+
+  const EXPECTED_TRANSACTION_NO_TRUNCATION = {
+    transaction: 'main',
+    spans: expect.arrayContaining([
+      expect.objectContaining({
+        data: expect.objectContaining({
+          [GEN_AI_INPUT_MESSAGES_ATTRIBUTE]: JSON.stringify([
+            { role: 'user', content: longContent },
+            { role: 'assistant', content: 'Some reply' },
+            { role: 'user', content: 'Follow-up question' },
+          ]),
+          [GEN_AI_INPUT_MESSAGES_ORIGINAL_LENGTH_ATTRIBUTE]: 3,
+        }),
+      }),
+    ]),
+  };
+
+  createEsmAndCjsTests(
+    __dirname,
+    'scenario-no-truncation.mjs',
+    'instrument-no-truncation.mjs',
+    (createRunner, test) => {
+      test('does not truncate input messages when enableTruncation is false', async () => {
+        await createRunner()
+          .ignore('event')
+          .expect({ transaction: EXPECTED_TRANSACTION_NO_TRUNCATION })
+          .start()
+          .completed();
+      });
+    },
+  );
 });

packages/core/.oxlintrc.json

@@ -16,6 +16,12 @@
       "rules": {
         "sdk/no-unsafe-random-apis": "off"
       }
+    },
+    {
+      "files": ["src/tracing/langchain/utils.ts"],
+      "rules": {
+        "max-lines": "off"
+      }
     }
   ],
   "ignorePatterns": ["rollup.npm.config.mjs"]

packages/core/src/tracing/langchain/index.ts

@@ -34,6 +34,7 @@ import {
  */
 export function createLangChainCallbackHandler(options: LangChainOptions = {}): LangChainCallbackHandler {
   const { recordInputs, recordOutputs } = resolveAIRecordingOptions(options);
+  const enableTruncation = options.enableTruncation ?? true;
 
   // Internal state - single instance tracks all spans
   const spanMap = new Map<string, Span>();
@@ -89,6 +90,7 @@ export function createLangChainCallbackHandler(options: LangChainOptions = {}):
         llm as LangChainSerialized,
         prompts,
         recordInputs,
+        enableTruncation,
         invocationParams,
         metadata,
       );
@@ -127,6 +129,7 @@ export function createLangChainCallbackHandler(options: LangChainOptions = {}):
         llm as LangChainSerialized,
         messages as LangChainMessage[][],
         recordInputs,
+        enableTruncation,
         invocationParams,
         metadata,
       );

packages/core/src/tracing/langchain/types.ts

@@ -13,6 +13,12 @@ export interface LangChainOptions {
    * @default false (respects sendDefaultPii option)
    */
   recordOutputs?: boolean;
+
+  /**
+   * Enable or disable truncation of recorded input messages.
+   * Defaults to `true`.
+   */
+  enableTruncation?: boolean;
 }
 
 /**

packages/core/src/tracing/langchain/utils.ts

@@ -26,8 +26,7 @@ import {
   GEN_AI_USAGE_TOTAL_TOKENS_ATTRIBUTE,
 } from '../ai/gen-ai-attributes';
 import { isContentMedia, stripInlineMediaFromSingleMessage } from '../ai/mediaStripping';
-import { truncateGenAiMessages } from '../ai/messageTruncation';
-import { extractSystemInstructions } from '../ai/utils';
+import { extractSystemInstructions, getJsonString, getTruncatedJsonString } from '../ai/utils';
 import { LANGCHAIN_ORIGIN, ROLE_MAP } from './constants';
 import type { LangChainLLMResult, LangChainMessage, LangChainSerialized } from './types';
 
@@ -284,6 +283,7 @@ export function extractLLMRequestAttributes(
   llm: LangChainSerialized,
   prompts: string[],
   recordInputs: boolean,
+  enableTruncation: boolean,
   invocationParams?: Record<string, unknown>,
   langSmithMetadata?: Record<string, unknown>,
 ): Record<string, SpanAttributeValue> {
@@ -295,7 +295,11 @@ export function extractLLMRequestAttributes(
   if (recordInputs && Array.isArray(prompts) && prompts.length > 0) {
     setIfDefined(attrs, GEN_AI_INPUT_MESSAGES_ORIGINAL_LENGTH_ATTRIBUTE, prompts.length);
     const messages = prompts.map(p => ({ role: 'user', content: p }));
-    setIfDefined(attrs, GEN_AI_INPUT_MESSAGES_ATTRIBUTE, asString(messages));
+    setIfDefined(
+      attrs,
+      GEN_AI_INPUT_MESSAGES_ATTRIBUTE,
+      enableTruncation ? getTruncatedJsonString(messages) : getJsonString(messages),
+    );
   }
 
   return attrs;
@@ -314,6 +318,7 @@ export function extractChatModelRequestAttributes(
   llm: LangChainSerialized,
   langChainMessages: LangChainMessage[][],
   recordInputs: boolean,
+  enableTruncation: boolean,
   invocationParams?: Record<string, unknown>,
   langSmithMetadata?: Record<string, unknown>,
 ): Record<string, SpanAttributeValue> {
@@ -334,8 +339,11 @@ export function extractChatModelRequestAttributes(
     const filteredLength = Array.isArray(filteredMessages) ? filteredMessages.length : 0;
     setIfDefined(attrs, GEN_AI_INPUT_MESSAGES_ORIGINAL_LENGTH_ATTRIBUTE, filteredLength);
 
-    const truncated = truncateGenAiMessages(filteredMessages as unknown[]);
-    setIfDefined(attrs, GEN_AI_INPUT_MESSAGES_ATTRIBUTE, asString(truncated));
+    setIfDefined(
+      attrs,
+      GEN_AI_INPUT_MESSAGES_ATTRIBUTE,
+      enableTruncation ? getTruncatedJsonString(filteredMessages) : getJsonString(filteredMessages),
+    );
   }
 
   return attrs;

packages/core/test/lib/tracing/langchain-utils.test.ts

@@ -237,7 +237,7 @@ describe('extractChatModelRequestAttributes with multimodal content', () => {
       ],
     ];
 
-    const attrs = extractChatModelRequestAttributes(serialized, messages, true);
+    const attrs = extractChatModelRequestAttributes(serialized, messages, true, true);
     const inputMessages = attrs[GEN_AI_INPUT_MESSAGES_ATTRIBUTE] as string | undefined;
 
     expect(inputMessages).toBeDefined();