fix(miniflare): Skip creating hyperdrive proxy server for local database

[Ltadrian]

Skip creating a local TCP proxy server for Hyperdrive bindings when SSL is not needed (sslmode disabled or not specified), connecting directly to the database instead. This avoids connection refused errors caused by firewall rules or proxy port binding issues on Windows/macOS.

Also fixes two bugs in HyperdriveProxyController:

• Add missing error handler on the proxy net.Server to prevent silent crashes
• Fix missing return in dispose() that caused servers not to be awaited on cleanup

Fixes: #11556

---

• Tests
  • Tests included/updated
  • Automated tests not possible - manual testing has been completed as follows:

On Windows 10 Machine

1. pnpm run build --filter wrangler
2. Run custom build of wrangler in worker with Hyperdrive pointing at local postgres
3. node ..\workers-sdk\packages\wrangler\wrangler-dist\cli.js dev
4. Confirm that connecting to localhost works as expected

• Additional testing not necessary because:
• Public documentation
  • Cloudflare docs PR(s):
  • Documentation not necessary because: This is expected behavior, and attempting to fix customers that run into this issue.

A picture of a cute animal (not mandatory, but encouraged)

[changeset-bot]

🦋 Changeset detected

Latest commit: b29b302

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[pkg-pr-new]

create-cloudflare

npm i https://pkg.pr.new/create-cloudflare@13390

@cloudflare/kv-asset-handler

npm i https://pkg.pr.new/@cloudflare/kv-asset-handler@13390

miniflare

npm i https://pkg.pr.new/miniflare@13390

@cloudflare/pages-shared

npm i https://pkg.pr.new/@cloudflare/pages-shared@13390

@cloudflare/unenv-preset

npm i https://pkg.pr.new/@cloudflare/unenv-preset@13390

@cloudflare/vite-plugin

npm i https://pkg.pr.new/@cloudflare/vite-plugin@13390

@cloudflare/vitest-pool-workers

npm i https://pkg.pr.new/@cloudflare/vitest-pool-workers@13390

@cloudflare/workers-editor-shared

npm i https://pkg.pr.new/@cloudflare/workers-editor-shared@13390

@cloudflare/workers-utils

npm i https://pkg.pr.new/@cloudflare/workers-utils@13390

wrangler

npm i https://pkg.pr.new/wrangler@13390

commit: b29b302

[workers-devprod]

Codeowners approval required for this PR:

• @cloudflare/wrangler

Show detailed file reviewers

• packages/miniflare/src/index.ts: [@cloudflare/wrangler]
• packages/miniflare/src/plugins/hyperdrive/hyperdrive-proxy.ts: [@cloudflare/wrangler]
• packages/miniflare/src/plugins/hyperdrive/index.ts: [@cloudflare/wrangler]
• packages/miniflare/test/plugins/hyperdrive/index.spec.ts: [@cloudflare/wrangler]

[petebacondarwin]

Tidy PR. Thanks!
Please can you update the changeset and address the devin comment about re-assigning the log object during setOptions.

[Ltadrian]

Tidy PR. Thanks! Please can you update the changeset and address the devin comment about re-assigning the log object during setOptions.

@petebacondarwin thank you! I have updated these, and the tests failing seem unrelated to my current PR

[workers-devprod]

Codeowners approval required for this PR:

• ✅ @cloudflare/wrangler

Show detailed file reviewers

[workers-devprod]

Codeowners reviews satisfied

[petebacondarwin]

Investigating the CI failures...

[petebacondarwin]

Hi @Ltadrian — pushed a follow-up commit (b29b302) to address the failing Wrangler E2E jobs (Linux shard 1/4, macOS shard 1/4, Windows shard 1/4, Windows shard 4/4).

What was failing

In every failing job the affected test itself passed (all assertions green), but Vitest then caught two unhandled ECONNRESET exceptions and failed the run.

Job	Test
Linux/macOS/Windows shard 1	e2e/dev.test.ts:819 — hyperdrive dev tests > connects to a socket
Windows shard 4	e2e/get-platform-proxy.test.ts:472 — Hyperdrive > can connect to a TCP socket via the hyperdrive connect method

Why this PR triggered it

Before this PR: worker → miniflare hyperdrive proxy (net.Server) → test mock server. When wrangler dev was force-killed via treeKill (WranglerLongLivedCommand.stop()), the proxy's clientSocket.pipe(dbSocket) semantics caused dbSocket.end() (a graceful FIN) to be sent to the test mock. So the mock saw a clean close.

After this PR (when sslmode=disable, the default): worker → test mock server directly. treeKill of wrangler tears down workerd's outbound TCP connection with RST instead of FIN. The mock sockets had no error listener, so Node escalated to uncaughtException and Vitest reported it.

The proxy was effectively absorbing the abrupt-close edge case for these tests.

The fix

Attach an ECONNRESET-tolerant socket.on('error', …) listener to the mock server sockets in the two affected files. This mirrors the pattern already in use at dev.test.ts:947 for the immediately-following sibling test — it just wasn't applied to the spots that started failing under the new direct-connect path.

function ignoreEconnreset(err: NodeJS.ErrnoException): void {
  if (err.code !== 'ECONNRESET') {
    throw err;
  }
}

No production code changes — the miniflare behaviour change in this PR is intentional and good. The unit tests in LocalRuntimeController.test.ts were unaffected because they tear down via controller.teardown() (graceful) rather than treeKill.

Verification

Ran both affected tests locally (after rebuilding miniflare + wrangler from this branch) and they pass cleanly with no Unhandled Errors:

• e2e/dev.test.ts -t \"connects to a socket\" — 2 passed (postgresql + mysql)
• e2e/get-platform-proxy.test.ts -t \"can connect to a TCP socket via the hyperdrive connect method\" — 3 passed

Format / lint / type-check all clean. No changeset added — the change is test-only and the existing quiet-houses-attack.md covers the production change.