cloudflare/workers-sdk miniflare@5.20261006.0-alpha

on GitHub

Minor Changes

• #15998 b75421f Thanks @dario-piotrowicz! - Add assets.base_path support to Workers Assets

  Serve an asset directory from a public URL prefix without changing its on-disk layout:

  {
    "assets": {
      "directory": "./public",
      "base_path": "/docs"
    }
  }

  Wrangler, preview, Miniflare, and generated build configuration preserve the explicitly selected value, while the Asset Worker normalizes it and strips the prefix only for asset lookup. Requests passed to a user Worker, request-facing headers, and redirects retain the public path. Relative pathname inputs are interpreted as root-relative prefixes, URL-shaped values are rejected, and omitting the option preserves existing root-path behavior.

  Authored _headers and _redirects rules continue to match full public paths. In particular, both the source and destination of an authored 200 asset rewrite must include the configured public prefix; Asset Worker-generated redirects are prefixed automatically.

• #15330 f8cdcb9 Thanks @akshitsinha! - Manage local Flagship flags in Local Explorer

  Bound Flagship apps now appear in Local Explorer. You can create, edit, toggle, delete, and evaluate flags against the same local store used by your Worker, including targeting conditions and percentage rollouts.

  Explorer requests are routed to the development process that owns each app, so Flagship management also works across multiple local Workers.

Patch Changes

• #16081 0ec13b7 Thanks @petebacondarwin! - Authenticate dev registry updates and internal loopback requests

  Require per-instance credentials for dev registry updates and internal loopback requests, including WebSocket upgrades. Authenticate callers before parsing registry updates or dispatching privileged loopback operations, while preserving legitimate shared-storage peers.

• #16014 c492d63 Thanks @dependabot! - Update dependencies of "miniflare", "wrangler"

  The following dependency versions have been updated:

  Dependency	From	To
  @cloudflare/workers-types	^5.20261001.1	^5.20261005.1
  workerd	1.20261001.1	1.20261005.1

• #16079 ba52118 Thanks @dependabot! - Update dependencies of "miniflare", "wrangler"

  The following dependency versions have been updated:

  Dependency	From	To
  @cloudflare/workers-types	^5.20261005.1	^5.20261006.1
  workerd	1.20261005.1	1.20261006.1

• #14921 946aaa7 Thanks @Mohith26! - Prevent local D1 session bookmark errors from crashing the development server

  Session bookmark lookup failures, including SQLite errors when another connection holds the database write lock, now reach the Worker as catchable D1_ERRORs. SQL execution and bookmark retrieval share a transaction, so a failed lookup rolls back the queries and retrying cannot duplicate their writes. This applies to local D1 through Miniflare, Wrangler, the Vite plugin, and the Vitest plugin.

  Fixes #14916

• #15781 48f3c04 Thanks @Wichtowski! - Reduce dispatchFetch() connection exhaustion under sustained local and CI workloads

  Repeated dispatches now reuse runtime connections for all HTTP methods, including POST, PUT, DELETE, and PATCH, instead of creating a new connection for every request. This prevents read-heavy and write-heavy Miniflare test suites from exhausting the host's available ephemeral ports. Transport failures are surfaced without automatically replaying requests, since Worker handlers can have side effects even for GET and HEAD. Idle runtime connections now close after one second, before workerd's five-second idle timeout can race with reuse.

• #16033 5606a74 Thanks @Pduhard! - Remove a 40 ms delay from Hyperdrive queries in local dev

  Miniflare's local Hyperdrive proxy left Nagle's algorithm on for its sockets. A Postgres driver that sends one query in several small writes, such as pg for every query with parameters, had the later writes held back until the database acknowledged the first, which took about 40 ms per query. Large results were held back the same way on the way back to the Worker.

  The proxy now turns on noDelay for the connection from the Worker and for the connection to the database. Connection strings using sslmode=disable are unaffected, since that mode connects directly and skips the proxy.

• #16050 e44cf6b Thanks @acchou! - Serve each Worker's own static assets when several Workers with assets run together

  When several Workers with static assets ran in one Miniflare instance, such as wrangler dev with multiple -c configs or the test harness, every Worker read its assets from the same Worker's directory. Other Workers got 404s or that Worker's file at the same path. Each Worker now reads its own assets directory.

• #16063 0b51fec Thanks @Cherry! - Start the synchronous proxy worker before returning proxies

  getBindings(), getDurableObjectNamespace() and the other proxy getters now wait for the worker that serves synchronous proxy calls to start, instead of the first synchronous call blocking Node's main thread while it boots. That block also stalled every other Miniflare instance served from the same process, such as Vitest pool workers running test files in parallel.