feat(js-sdk)!: drop undici 7 and the in-flight cap, depend on undici@^8 only

.changeset/drop-undici7.md

@@ -0,0 +1,11 @@
+---
+'e2b': minor
+'@e2b/cli': minor
+'@e2b/code-interpreter': minor
+'@e2b/desktop': minor
+---
+
+Require Node.js 22.19.0 or newer and depend on `undici@^8` only (undici 7 is dropped).
+
+- undici 8 respects the server's HTTP/2 `SETTINGS_MAX_CONCURRENT_STREAMS` and opens more connections when one is saturated.
+- Breaking: the SDK-level in-flight request cap is removed, along with the `E2B_API_INFLIGHT_REQUESTS`, `E2B_ENVD_INFLIGHT_REQUESTS` and `E2B_ENVD_RPC_INFLIGHT_REQUESTS` env vars. Concurrency is governed by undici's connection pools (`E2B_API_CONNECTIONS`, `E2B_ENVD_RPC_CONNECTIONS`).

codegen.Dockerfile

@@ -24,7 +24,7 @@ ENV PATH="/go/bin:${PATH}"
 RUN pip install black==26.3.1 e2b-openapi-python-client==0.26.2 datamodel-code-generator==0.64.0 protoc-gen-connectrpc==0.11.1 protoc-gen-py==0.1.1
 
 # Install Node.js (pinned to match .tool-versions)
-ENV NODE_VERSION=22.18.0
+ENV NODE_VERSION=22.23.3
 RUN ARCH=$(uname -m) && \
     case "$ARCH" in \
         x86_64) NODE_ARCH="x64" ;; \

packages/cli/package.json

@@ -89,6 +89,6 @@
     "yup": "^1.7.1"
   },
   "engines": {
-    "node": ">=22.9.0"
+    "node": ">=22.19.0"
   }
 }

packages/code-interpreter-js/package.json

@@ -64,7 +64,7 @@
     "vm"
   ],
   "engines": {
-    "node": ">=22"
+    "node": ">=22.19.0"
   },
   "browserslist": [
     "defaults"

packages/desktop-js/package.json

@@ -36,7 +36,7 @@
     "package.json"
   ],
   "engines": {
-    "node": ">=22"
+    "node": ">=22.19.0"
   },
   "browserslist": [
     "defaults"

packages/js-sdk/package.json

@@ -97,13 +97,10 @@
     "openapi-fetch": "^0.17.0",
     "platform": "^1.3.6",
     "tar": "^7.5.22",
-    "undici": "^7.30.0"
-  },
-  "optionalDependencies": {
-    "undici8": "npm:undici@8.11.2"
+    "undici": "^8.11.2"
   },
   "engines": {
-    "node": ">=22"
+    "node": ">=22.19.0"
   },
   "browserslist": [
     "defaults"

packages/js-sdk/src/api/http2.ts

@@ -1,16 +1,13 @@
 import { runtime } from '../utils'
 import { DEFAULT_HTTP_VERSION, type HttpVersion } from '../connectionConfig'
-import { parseInflightLimitEnv, parsePositiveIntEnv } from './metadata'
+import { parsePositiveIntEnv } from './metadata'
 import {
   buildDispatchedFetch,
   createRuntimeFetch,
   type UndiciModule,
 } from '../undici'
 
 const DEFAULT_API_CONNECTION_LIMIT = 100
-// 1000 = ~10 streams per connection (with the 100-conn default).
-// Override via env if your workload needs different.
-const DEFAULT_API_INFLIGHT_LIMIT = 1000
 
 // Fetchers are cached per proxy and HTTP version so requests without a proxy
 // keep sharing a single dispatcher while each distinct proxy URL gets its own.
@@ -42,7 +39,6 @@ export function createApiFetchForRuntime(
   currentRuntime = runtime,
   options: {
     connectionLimit?: number
-    inflightLimit?: number
     proxy?: string
     httpVersion?: HttpVersion
     loadUndici?: () => Promise<UndiciModule | undefined>
@@ -53,7 +49,6 @@ export function createApiFetchForRuntime(
   return createRuntimeFetch(currentRuntime, () =>
     buildDispatchedFetch({
       connections: options.connectionLimit ?? getApiConnectionLimit(),
-      inflightLimit: options.inflightLimit ?? getApiInflightLimit(),
       proxy: options.proxy,
       httpVersion: options.httpVersion,
       loadUndici: options.loadUndici,
@@ -67,17 +62,3 @@ export function getApiConnectionLimit(): number {
     DEFAULT_API_CONNECTION_LIMIT
   )
 }
-
-/**
- * Returns the configured max number of API requests that can be in flight at
- * once, or `0` to disable the cap.
- *
- * Defaults to `1000` ({@link DEFAULT_API_INFLIGHT_LIMIT}). Override via
- * `E2B_API_INFLIGHT_REQUESTS` env var; set to `0` to disable the cap entirely.
- */
-export function getApiInflightLimit(): number {
-  return parseInflightLimitEnv(
-    'E2B_API_INFLIGHT_REQUESTS',
-    DEFAULT_API_INFLIGHT_LIMIT
-  )
-}

packages/js-sdk/src/api/metadata.ts

@@ -62,24 +62,3 @@ export function parsePositiveIntEnv(
 
   return parsed
 }
-
-/**
- * Parse an inflight-limit env var. Returns `0` to disable the cap (documented
- * opt-out) or a positive integer to cap concurrency. Throws on non-integer or
- * negative values so misconfiguration is surfaced loudly rather than silently
- * removing the cap. A return value of `0` is recognized by
- * {@link limitConcurrency} as "no cap".
- */
-export function parseInflightLimitEnv(
-  name: string,
-  defaultValue: number
-): number {
-  const parsed = parseIntEnv(name, defaultValue)
-  if (parsed < 0) {
-    throw new Error(
-      `Invalid ${name}=${parsed}: expected a non-negative integer ` +
-        '(use 0 to disable the cap).'
-    )
-  }
-  return parsed
-}

packages/js-sdk/src/envd/http2.ts

@@ -1,7 +1,7 @@
 import { runtime } from '../utils'
 import { DEFAULT_HTTP_VERSION, type HttpVersion } from '../connectionConfig'
 import type { FetchOpts } from '../api/http2'
-import { parseInflightLimitEnv, parsePositiveIntEnv } from '../api/metadata'
+import { parsePositiveIntEnv } from '../api/metadata'
 import {
   buildDispatchedFetch,
   createRuntimeFetch,
@@ -10,7 +10,6 @@ import {
 
 type EnvdFetchOptions = {
   connectionLimit?: number
-  inflightLimit?: number
   proxy?: string
   httpVersion?: HttpVersion
   loadUndici?: () => Promise<UndiciModule | undefined>
@@ -22,8 +21,6 @@ const envdFetchers = new Map<string, typeof fetch>()
 const envdRpcFetchers = new Map<string, typeof fetch>()
 const DEFAULT_ENVD_CONNECTION_LIMIT = 10
 const DEFAULT_ENVD_RPC_CONNECTION_LIMIT = 200
-const DEFAULT_ENVD_INFLIGHT_LIMIT = 2000
-const DEFAULT_ENVD_RPC_INFLIGHT_LIMIT = 2000
 
 export function createEnvdFetchForRuntime(
   currentRuntime = runtime,
@@ -32,7 +29,6 @@ export function createEnvdFetchForRuntime(
   return createRuntimeFetch(currentRuntime, () =>
     buildDispatchedFetch({
       connections: options.connectionLimit ?? DEFAULT_ENVD_CONNECTION_LIMIT,
-      inflightLimit: options.inflightLimit ?? 0,
       proxy: options.proxy,
       httpVersion: options.httpVersion,
       loadUndici: options.loadUndici,
@@ -61,7 +57,6 @@ export function createEnvdFetch({
   // Keep one origin connection for short envd REST calls. If ALPN falls back
   // to h1, this favors connection pressure over per-sandbox throughput.
   const envdFetch = createEnvdFetchForRuntime(runtime, {
-    inflightLimit: getEnvdInflightLimit(),
     proxy,
     httpVersion,
   })
@@ -83,7 +78,6 @@ export function createEnvdRpcFetch({
 
   const envdRpcFetch = createEnvdFetchForRuntime(runtime, {
     connectionLimit: getEnvdRpcConnectionLimit(),
-    inflightLimit: getEnvdRpcInflightLimit(),
     proxy,
     httpVersion,
   })
@@ -98,35 +92,3 @@ export function getEnvdRpcConnectionLimit(): number {
     DEFAULT_ENVD_RPC_CONNECTION_LIMIT
   )
 }
-
-/**
- * Returns the configured max number of envd REST requests (e.g.
- * `files.read`/`files.write`) that can be in flight at once across all
- * sandboxes in this SDK process, or `0` to disable the cap.
- *
- * Defaults to `2000` ({@link DEFAULT_ENVD_INFLIGHT_LIMIT}). Override
- * via `E2B_ENVD_INFLIGHT_REQUESTS` env var; set to `0` to disable the cap
- * entirely.
- */
-export function getEnvdInflightLimit(): number {
-  return parseInflightLimitEnv(
-    'E2B_ENVD_INFLIGHT_REQUESTS',
-    DEFAULT_ENVD_INFLIGHT_LIMIT
-  )
-}
-
-/**
- * Returns the configured max number of envd RPC requests that
- * can be in flight at once across all sandboxes in this SDK process,
- * or `0` to disable the cap.
- *
- * Defaults to `2000` ({@link DEFAULT_ENVD_RPC_INFLIGHT_LIMIT}). Override
- * via `E2B_ENVD_RPC_INFLIGHT_REQUESTS` env var; set to `0` to disable the cap
- * entirely.
- */
-export function getEnvdRpcInflightLimit(): number {
-  return parseInflightLimitEnv(
-    'E2B_ENVD_RPC_INFLIGHT_REQUESTS',
-    DEFAULT_ENVD_RPC_INFLIGHT_LIMIT
-  )
-}

packages/js-sdk/src/template/buildApi.ts

@@ -183,10 +183,9 @@ async function putFileStream(
 ): Promise<{ ok: boolean; statusText: string }> {
   // Prefer undici's fetch: it honors the explicit Content-Length on stream
   // bodies on every runtime, while Deno's native fetch ignores the header and
-  // falls back to chunked. Where undici isn't resolvable (e.g. bundled apps),
-  // fall back to the global fetch — Node's and Bun's native fetch also honor
-  // Content-Length on stream bodies. loadUndici tries undici 8 first and
-  // falls back to undici 7 where 8 doesn't import (Bun).
+  // falls back to chunked. Where undici doesn't load (e.g. bundled apps),
+  // the global fetch is used — Node's and Bun's native fetch also honor
+  // Content-Length on stream bodies.
   const undici = await loadUndici()
   const fetchImpl = (undici?.fetch as typeof fetch | undefined) ?? fetch