Socket hang up with Lambda with JS client

I have a Lambda@Edge function (origin request) that uses @libsql/client to query Turso.

It initializes the client in the global scope (i.e. not in the handler) as is common practice in Lambdas so you can re-use the client across subsequent invocations.

e.g.

const dbClient = createClient({
  url: DB_URL,
  authToken: DB_AUTH_TOKEN,
});

export async function handler(
  event: CloudFrontRequestEvent
): Promise<CloudFrontRequestResult> {
  // Use `dbClient` here
}

Testing just now, my first request was processed correctly but on the second request I got:

2023-09-30T02:44:21.779Z	cdd6ee8d-b573-40e8-8871-224a0551d468	ERROR	Invoke Error 	
{
    "errorType": "FetchError",
    "errorMessage": "request to https://jpdict-dev-birchill.turso.io/v2/pipeline failed, reason: socket hang up",
    "code": "ECONNRESET",
    "message": "request to https://jpdict-dev-birchill.turso.io/v2/pipeline failed, reason: socket hang up",
    "type": "system",
    "errno": "ECONNRESET",
    "stack": [
        "FetchError: request to https://jpdict-dev-birchill.turso.io/v2/pipeline failed, reason: socket hang up",
        "    at ClientRequest.<anonymous> (/var/task/index.js:7054:18)",
        "    at ClientRequest.emit (node:events:514:28)",
        "    at TLSSocket.socketCloseListener (node:_http_client:474:11)",
        "    at TLSSocket.emit (node:events:526:35)",
        "    at node:net:323:12",
        "    at TCP.done (node:_tls_wrap:588:7)"
    ]
}

Is there some lifetime to the connection I need to be aware of? Some way to force a persistent connection? Or should I just create a new client for each invocation?

For what it's worth I am using version 0.3.5 but I never saw this error with version 0.3.4 so I suppose it might be related to using libsql instead of better-sqlite3.

Start by reproducing the second-invocation failure in the Lambda@Edge handler with a globally initialized @libsql/client, then compare behavior between versions 0.3.5 and 0.3.4. Inspect the bundled index.js stack trace and client connection lifecycle; done means identifying whether connection reuse causes ECONNRESET and documenting the supported invocation pattern or required fix.