langgraph-api 0.5.35__tar.gz → 0.6.25__tar.gz

{langgraph_api-0.5.35 → langgraph_api-0.6.25}/Makefile

@@ -2,7 +2,7 @@
 
 # Environment variables
 FF_USE_CORE_API ?= false
-LANGSERVE_GRAPHS_ALL = '{"agent": {"path": "./tests/graphs/agent.py:graph", "description": "agent"}, "custom_lifespan": "./tests/graphs/my_router.py:graph", "single_node": "./tests/graphs/single_node.py:graph", "benchmark": "./tests/graphs/benchmark.py:graph", "config_graph": "./tests/graphs/config_graph.py:graph", "other": "./tests/graphs/other.py:make_graph", "weather": "./tests/graphs/weather.py:mk_weather_graph", "searchy": "./tests/graphs/searchy.py:graph", "agent_simple": "./tests/graphs/agent_simple.py:graph", "simple_runtime": "./tests/graphs/simple_runtime.py:graph", "agent_interrupt": "./tests/graphs/agent_interrupt.py:graph", "message_type_test": "./tests/graphs/message_type_test.py:graph", "remote_subgraph_parent": "./tests/graphs/remote_subgraph_parent.py:graph", "simple_remote": "./tests/graphs/simple_remote.py:graph", "nested_subgraphs": "./tests/graphs/nested_subgraphs.py:graph", "functional_fibonacci": "./tests/graphs/functional_fibonacci.py:fibonacci", "state_graph_fibonacci": "./tests/graphs/state_graph_fibonacci.py:fibonacci"}'
+LANGSERVE_GRAPHS_ALL = '{"agent": {"path": "./tests/graphs/agent.py:graph", "description": "agent"}, "custom_lifespan": "./tests/graphs/my_router.py:graph", "single_node": "./tests/graphs/single_node.py:graph", "benchmark": "./tests/graphs/benchmark.py:graph", "config_graph": "./tests/graphs/config_graph.py:graph", "other": "./tests/graphs/other.py:make_graph", "weather": "./tests/graphs/weather.py:mk_weather_graph", "searchy": "./tests/graphs/searchy.py:graph", "agent_simple": "./tests/graphs/agent_simple.py:graph", "simple_runtime": "./tests/graphs/simple_runtime.py:graph", "agent_interrupt": "./tests/graphs/agent_interrupt.py:graph", "message_type_test": "./tests/graphs/message_type_test.py:graph", "remote_subgraph_parent": "./tests/graphs/remote_subgraph_parent.py:graph", "simple_remote": "./tests/graphs/simple_remote.py:graph", "nested_subgraphs": "./tests/graphs/nested_subgraphs.py:graph", "functional_fibonacci": "./tests/graphs/functional_fibonacci.py:fibonacci", "state_graph_fibonacci": "./tests/graphs/state_graph_fibonacci.py:fibonacci", "max_concurrency_graph": "./tests/graphs/max_concurrency_graph.py:graph"}'
 LANGSERVE_GRAPHS_AUTH = '{"agent": {"path": "./tests/graphs/agent.py:graph", "description": "agent"}, "config_graph": "./tests/graphs/config_graph.py:graph", "other": "./tests/graphs/other.py:make_graph", "weather": "./tests/graphs/weather.py:mk_weather_graph", "searchy": "./tests/graphs/searchy.py:graph", "agent_simple": "./tests/graphs/agent_simple.py:graph", "simple_runtime": "./tests/graphs/simple_runtime.py:graph", "functional_fibonacci": "./tests/graphs/functional_fibonacci.py:fibonacci", "state_graph_fibonacci": "./tests/graphs/state_graph_fibonacci.py:fibonacci"}'
 
 # Go server management
@@ -66,7 +66,7 @@ LANGGRAPH_ENCRYPTION ?=
 
 test-license-oss:
 	@if [ "$(FF_USE_CORE_API)" = "true" ]; then \
-		LANGGRAPH_RUNTIME_EDITION=inmem LANGGRAPH_HTTP='$(HTTP_CONFIG)' LANGGRAPH_STORE='$(STORE_CONFIG)' LANGGRAPH_ENCRYPTION='$(LANGGRAPH_ENCRYPTION)' REDIS_URI=_FAKE DATABASE_URI=:memory: MIGRATIONS_PATH=__inmem__ FF_USE_CORE_API=true uv run pytest -m grpc -v $(TEST); \
+		LANGGRAPH_RUNTIME_EDITION=inmem LANGGRAPH_HTTP='$(HTTP_CONFIG)' LANGGRAPH_STORE='$(STORE_CONFIG)' LANGGRAPH_ENCRYPTION='$(LANGGRAPH_ENCRYPTION)' REDIS_URI=_FAKE DATABASE_URI=:memory: MIGRATIONS_PATH=__inmem__ FF_USE_CORE_API=true uv run pytest -m sqlite_grpc -v $(TEST); \
 	else \
 		LANGGRAPH_RUNTIME_EDITION=inmem LANGGRAPH_HTTP='$(HTTP_CONFIG)' LANGGRAPH_STORE='$(STORE_CONFIG)' LANGGRAPH_ENCRYPTION='$(LANGGRAPH_ENCRYPTION)' REDIS_URI=_FAKE DATABASE_URI=:memory: MIGRATIONS_PATH=__inmem__ uv run pytest -v $(TEST); \
 	fi

{langgraph_api-0.5.35 → langgraph_api-0.6.25}/PKG-INFO

@@ -1,20 +1,21 @@
 Metadata-Version: 2.4
 Name: langgraph-api
-Version: 0.5.35
+Version: 0.6.25
 Author-email: Will Fu-Hinthorn <will@langchain.dev>, Josh Rogers <josh@langchain.dev>, Parker Rule <parker@langchain.dev>
 License: Elastic-2.0
 License-File: LICENSE
 Requires-Python: >=3.11
 Requires-Dist: cloudpickle>=3.0.0
-Requires-Dist: cryptography<45.0,>=42.0.0
+Requires-Dist: cryptography<47.0,>=42.0.0
+Requires-Dist: grpcio-health-checking<2.0.0,>=1.75.0
 Requires-Dist: grpcio-tools==1.75.1
 Requires-Dist: grpcio<2.0.0,>=1.75.0
 Requires-Dist: httpx>=0.25.0
 Requires-Dist: jsonschema-rs<0.30,>=0.20.0
 Requires-Dist: langchain-core>=0.3.64
-Requires-Dist: langgraph-checkpoint<4,>=3.0.1
-Requires-Dist: langgraph-runtime-inmem<0.20.0,>=0.19.0
-Requires-Dist: langgraph-sdk>=0.2.10
+Requires-Dist: langgraph-checkpoint<5,>=3.0.1
+Requires-Dist: langgraph-runtime-inmem<0.23.0,>=0.22.0
+Requires-Dist: langgraph-sdk>=0.3.0
 Requires-Dist: langgraph<2,>=0.4.10
 Requires-Dist: langsmith>=0.3.45
 Requires-Dist: opentelemetry-api>=1.37.0

{langgraph_api-0.5.35 → langgraph_api-0.6.25}/benchmark/Makefile

@@ -1,5 +1,6 @@
 # Benchmark commands
 BASE_URL ?= https://benchmark-dr-s-2799835ad04b501a95044223ae72ced7.staging.langgraph.app
+SLOWEST_N ?= 2
 RAMP_START ?= 40
 RAMP_END ?= 1000
 RAMP_MULTIPLIER ?= 2
@@ -23,6 +24,7 @@ benchmark-capacity:
 	rm -f capacity_summary_t*.json capacity_raw_t*.json capacity_histogram_*.json
 	npm install
 	BASE_URL=$(BASE_URL) \
+	SLOWEST_N=$(SLOWEST_N) \
 	RAMP_START=$(RAMP_START) \
 	RAMP_END=$(RAMP_END) \
 	RAMP_MULTIPLIER=$(RAMP_MULTIPLIER) \
@@ -50,3 +52,38 @@ benchmark-clean:
 
 benchmark-clean-charts:
 	rm -f *_chart_*.png
+
+# Mixed workload benchmark - tests quick runs competing with long runs
+QUICK_VUS ?= 250
+QUICK_ITERATIONS ?= 1000
+QUICK_STEPS ?= 5
+QUICK_DATA_SIZE ?= 100
+QUICK_MAX_WAIT_SECONDS ?= 120
+QUICK_POLL_INTERVAL ?= 2
+LONG_VUS ?= 5
+LONG_ITERATIONS ?= 10
+LONG_STEPS ?= 50
+LONG_DELAY ?= 1
+LONG_DATA_SIZE ?= 10000
+LONG_MAX_WAIT_SECONDS ?= 200
+LONG_POLL_INTERVAL ?= 5
+
+benchmark-mixed:
+	rm -f mixed_workload_*.json
+	npm install
+	BASE_URL=$(BASE_URL) \
+	SLOWEST_N=$(SLOWEST_N) \
+	QUICK_VUS=$(QUICK_VUS) \
+	QUICK_ITERATIONS=$(QUICK_ITERATIONS) \
+	QUICK_STEPS=$(QUICK_STEPS) \
+	QUICK_DATA_SIZE=$(QUICK_DATA_SIZE) \
+	QUICK_MAX_WAIT_SECONDS=$(QUICK_MAX_WAIT_SECONDS) \
+	QUICK_POLL_INTERVAL=$(QUICK_POLL_INTERVAL) \
+	LONG_VUS=$(LONG_VUS) \
+	LONG_ITERATIONS=$(LONG_ITERATIONS) \
+	LONG_STEPS=$(LONG_STEPS) \
+	LONG_DELAY=$(LONG_DELAY) \
+	LONG_DATA_SIZE=$(LONG_DATA_SIZE) \
+	LONG_MAX_WAIT_SECONDS=$(LONG_MAX_WAIT_SECONDS) \
+	LONG_POLL_INTERVAL=$(LONG_POLL_INTERVAL) \
+	node mixed_workload_runner.mjs

{langgraph_api-0.5.35 → langgraph_api-0.6.25}/benchmark/README.md

@@ -7,7 +7,7 @@ K6 is a modern load testing tool that allows you to test the performance and rel
 ### Available Tests
 
 There are two modes of testing available:
-1. `Burst` - Kick off a burst of /run/wait requests. 
+1. `Burst` - Kick off a burst of /run/wait requests.
     Available Params:
         BURST_SIZE - How many requests to run. Default: 100
 2. `Ramp` - Scale up the number of /run/wait requests and then plateau.
@@ -71,4 +71,4 @@ Charts can be created from the run locally using the `make benchmark-charts` com
 
 - [K6 Documentation](https://k6.io/docs/)
 - [K6 JavaScript API](https://k6.io/docs/javascript-api/)
-- [Performance Testing Best Practices](https://k6.io/docs/testing-guides/) 
+- [Performance Testing Best Practices](https://k6.io/docs/testing-guides/)

{langgraph_api-0.5.35 → langgraph_api-0.6.25}/benchmark/capacity_k6.js

@@ -13,6 +13,7 @@ const successfulRuns = new Counter('successful_runs');
 const failedRuns = new Counter('failed_runs');
 const capacitySuccessRate = new Rate('capacity_success_rate');
 
+
 // Env
 const BASE_URL = __ENV.BASE_URL;
 const LANGSMITH_API_KEY = __ENV.LANGSMITH_API_KEY;
@@ -59,6 +60,7 @@ function buildPayload() {
   });
 }
 
+
 export default function () {
   // one-shot per iteration: create → wait → poll status once
   const payload = buildPayload();
@@ -130,7 +132,7 @@ export default function () {
         }
         const durMs = new Date(run.updated_at).getTime() - createdAt;
         if (!Number.isNaN(durMs) && durMs >= 0) {
-          runDuration.add(durMs);
+          runDuration.add(durMs, { runId, threadId });
         }
         const pickupDurMs = new Date(t.values.start_time).getTime() - new Date(run.created_at).getTime();
         if (!Number.isNaN(pickupDurMs) && pickupDurMs >= 0) {

{langgraph_api-0.5.35 → langgraph_api-0.6.25}/benchmark/capacity_runner.mjs

@@ -75,16 +75,24 @@ async function cleanThreads() {
   let totalDeleted = 0;
   // Loop until no more threads
   while (true) {
-    const res = await fetch(searchUrl, {
-      method: 'POST',
-      headers: hdrs,
-      body: JSON.stringify({ limit: 1000 }),
-    });
-    if (!res.ok) {
-      console.error(`Cleanup search failed: ${res.status} ${res.statusText}`);
+    let threads;
+    try {
+      const res = await fetch(searchUrl, {
+        method: 'POST',
+        headers: hdrs,
+        // Only request thread_id to minimize response size
+        body: JSON.stringify({ limit: 1000, select: ['thread_id'] }),
+      });
+      if (!res.ok) {
+        console.error(`Cleanup search failed: ${res.status} ${res.statusText}`);
+        break;
+      }
+      threads = await res.json();
+    } catch (e) {
+      // Response too large or other error - skip cleanup
+      console.error(`Cleanup search error: ${e?.message || e}`);
       break;
     }
-    const threads = await res.json();
     if (!Array.isArray(threads) || threads.length === 0) break;
     for (const t of threads) {
       try {
@@ -104,6 +112,8 @@ async function cleanThreads() {
   console.log(`Cleanup completed. Deleted ~${totalDeleted} threads.`);
 }
 
+const SLOWEST_N = envInt('SLOWEST_N', 2);
+
 function runK6(target) {
   const env = {
     ...process.env,
@@ -138,7 +148,13 @@ function loadSummaryForTarget(target) {
   }
   const latest = files[files.length - 1];
   const content = readFileSync(join(process.cwd(), latest), 'utf-8');
-  return JSON.parse(content);
+  return { filename: latest, data: JSON.parse(content) };
+}
+
+function updateSummaryWithSlowestRuns(summaryFilename, summaryData, slowestRuns) {
+  summaryData.slowestRuns = slowestRuns;
+  writeFileSync(join(process.cwd(), summaryFilename), JSON.stringify(summaryData, null, 2));
+  console.log(`Updated ${summaryFilename} with ${slowestRuns.length} slowest runs`);
 }
 
 async function main() {
@@ -155,14 +171,18 @@ async function main() {
     let stageTimestamp = null;
     let runDurationHistogram = null;
     let histogramSampleCount = 0;
+    let slowestRuns = [];
+    let rawOut = null;
 
     try {
-      const { rawOut, ts } = runK6(currentTarget);
-      stageTimestamp = ts;
+      const result = runK6(currentTarget);
+      rawOut = result.rawOut;
+      stageTimestamp = result.ts;
       try {
-        const histogramResult = await generateHistogramsForStage(rawOut);
+        const histogramResult = await generateHistogramsForStage(result.rawOut);
         runDurationHistogram = histogramResult?.runDurationHistogramSeconds ?? null;
         histogramSampleCount = histogramResult?.sampleCount ?? 0;
+        slowestRuns = histogramResult?.slowestRuns ?? [];
       } catch (e) {
         console.error(`Failed to generate histograms for N=${currentTarget}:`, e?.message || e);
       }
@@ -171,9 +191,17 @@ async function main() {
       // Treat as failure for this step
     }
 
+    let summaryFilename = null;
     let summary = null;
     try {
-      summary = loadSummaryForTarget(currentTarget);
+      const result = loadSummaryForTarget(currentTarget);
+      summaryFilename = result.filename;
+      summary = result.data;
+
+      // Update summary with slowest runs extracted from raw JSON
+      if (slowestRuns.length > 0) {
+        updateSummaryWithSlowestRuns(summaryFilename, summary, slowestRuns);
+      }
     } catch (e) {
       console.error(`Failed to read summary for N=${currentTarget}:`, e?.message || e);
     }
@@ -190,7 +218,7 @@ async function main() {
       console.log(`Step N=${currentTarget} summary incomplete`);
     }
 
-    const effectiveTimestamp = summary?.timestamp ?? stageTimestamp ?? slugifyTimestamp();
+    const effectiveTimestamp = summary?.timestamp ?? stageTimestamp ?? slugifyTimestamp(null);
 
     if (Array.isArray(runDurationHistogram) && runDurationHistogram.length > 0) {
       const histogramLog = {
@@ -279,6 +307,7 @@ function buildHistogramBuckets(valuesSeconds, bucketCount = 12) {
 }
 
 // Build and save histogram charts for one stage from raw K6 JSON
+// Also extracts slowest runs from run_duration metric tags
 async function generateHistogramsForStage(rawFile) {
   // Parse streaming JSONL from k6 --out json
   const metrics = {
@@ -288,6 +317,8 @@ async function generateHistogramsForStage(rawFile) {
     run_insertion_duration: [],
     run_oss_duration: [],
   };
+  // Collect run_duration entries with tags for slowest runs
+  const runDurationWithTags = [];
 
   await new Promise((resolve, reject) => {
     const rl = readline.createInterface({ input: createReadStream(join(process.cwd(), rawFile), { encoding: 'utf-8' }), crlfDelay: Infinity });
@@ -300,6 +331,18 @@ async function generateHistogramsForStage(rawFile) {
             const v = entry?.data?.value;
             if (Number.isFinite(v)) metrics[name].push(v);
           }
+          // Capture run_duration entries with runId/threadId tags
+          if (name === 'run_duration') {
+            const tags = entry?.data?.tags;
+            const v = entry?.data?.value;
+            if (tags?.runId && tags?.threadId && Number.isFinite(v)) {
+              runDurationWithTags.push({
+                durationMs: v,
+                runId: tags.runId,
+                threadId: tags.threadId,
+              });
+            }
+          }
         }
       } catch (_) {
         // ignore parse errors
@@ -309,9 +352,20 @@ async function generateHistogramsForStage(rawFile) {
     rl.on('error', reject);
   });
 
+  // Get slowest N runs
+  const slowestRuns = runDurationWithTags
+    .sort((a, b) => b.durationMs - a.durationMs)
+    .slice(0, SLOWEST_N)
+    .map((r) => ({
+      durationSeconds: r.durationMs / 1000,
+      runId: r.runId,
+      threadId: r.threadId,
+    }));
+
   const runDurationSeconds = metrics.run_duration.map((v) => v / 1000);
   return {
     runDurationHistogramSeconds: buildHistogramBuckets(runDurationSeconds),
     sampleCount: runDurationSeconds.length,
+    slowestRuns,
   };
 }

{langgraph_api-0.5.35 → langgraph_api-0.6.25}/benchmark/clean.js

@@ -28,7 +28,7 @@ async function cleanAssistants() {
     let totalDeleted = 0;
 
     console.log('Starting assistant cleanup...');
-    
+
     while (true) {
         // Get the next page of assistants
         console.log('Searching for assistants...');
@@ -48,7 +48,7 @@ async function cleanAssistants() {
         }
 
         const assistants = await searchResponse.json();
-        
+
         // If no assistants found, we're done
         if (!assistants || assistants.length === 0) {
             console.log('No more assistants found.');
@@ -93,7 +93,7 @@ async function cleanThreads() {
     let totalDeleted = 0;
 
     console.log('Starting thread cleanup...');
-    
+
     while (true) {
         // Get the next page of threads
         console.log('Searching for threads...');
@@ -101,7 +101,8 @@ async function cleanThreads() {
             method: 'POST',
             headers,
             body: JSON.stringify({
-                limit: 1000
+                limit: 1000,
+                select: ['thread_id']
             })
         });
 
@@ -110,7 +111,7 @@ async function cleanThreads() {
         }
 
         const threads = await searchResponse.json();
-        
+
         // If no threads found, we're done
         if (!threads || threads.length === 0) {
             console.log('No more threads found.');
@@ -147,4 +148,4 @@ async function cleanThreads() {
 clean().catch(error => {
     console.error('Unhandled error:', error.message);
     process.exit(1);
-});
+});

langgraph_api-0.6.25/benchmark/continuous/README.md

@@ -0,0 +1,38 @@
+# Continuous Load Test
+
+Maintains a target pool of streaming threads to exercise long-running HTTP SSE streams.
+
+## Local Usage
+
+```bash
+# Install dependencies
+uv sync
+
+# Set environment variables
+export LANGSMITH_API_KEY=your-langsmith-key
+export DD_API_KEY=your-datadog-key
+
+# Run with defaults (1000 threads)
+uv run python runner.py
+
+# Or with custom options
+# use --raw-sse flag to send without SDK (which would mask retries etc)
+uv run python runner.py \
+  --base-url https://your-deployment.langgraph.app \
+  --assistant-id long_running \
+  --num-threads 1000
+  --raw-sse
+```
+
+## Metrics
+
+Metrics are shipped to Datadog every 60 seconds. Per-run metrics are sent once when each run completes:
+
+- `langsmith_deployment.continuous.threads.active` - Active streaming threads (gauge)
+- `langsmith_deployment.continuous.threads.failed` - Failed threads (gauge)
+- `langsmith_deployment.continuous.events` - SSE events received per run (count)
+- `langsmith_deployment.continuous.failures` - Connection failures per run (count)
+- `langsmith_deployment.continuous.threads.faults.dropped_chunks` - Dropped messages per run (count)
+- `langsmith_deployment.continuous.threads.faults.out_of_order_runs` - Runs with out-of-order messages (count)
+- `langsmith_deployment.continuous.threads.faults.malformed_chunks` - Malformed chunks per run (count)
+- `langsmith_deployment.continuous.threads.latency` - End-to-end latency per message in seconds (distribution)

langgraph_api-0.6.25/benchmark/continuous/pyproject.toml

@@ -0,0 +1,13 @@
+[project]
+name = "langgraph-continuous-test"
+version = "0.1.0"
+description = "Continuous stress test for Agent Server"
+requires-python = ">=3.13"
+dependencies = [
+    "langgraph-sdk>=0.2.10",
+    "click>=8.0.0",
+    "structlog>=24.1.0",
+    "datadog-api-client>=2.0.0",
+    "python-dateutil>=2.8.0",
+    "httpx>=0.27.0",
+]