Client handles abort signals incorrectly in high availability configuration

[jonatansberg]

Description

Using SearchClient with the cluster configuration suggested by the docs (both LB + nodes) together with an AbortSignal causes the client to bypass the LB and call the nodes directly if/when the signal fires.

Steps to reproduce

• Use SearchClient in cluster configuration suggested by the docs (both LB + nodes)
• Pass an abort controller signal in SearchOptions

let controller: AbortController | null = null;
const client = new SearchClient({
  nearestNode: { host: 'xxx.a1.typesense.net', port: 443, protocol: 'https' },
  nodes: [
    { host: 'xxx-1.a1.typesense.net', port: 443, protocol: 'https' },
    { host: 'xxx-2.a1.typesense.net', port: 443, protocol: 'https' },
    { host: 'xxx-3.a1.typesense.net', port: 443, protocol: 'https' },
  ],
  apiKey: '<API_KEY>',
  connectionTimeoutSeconds: 2
});

function search(q: string) {
  if (controller) controller.abort();
  controller = new AbortController();
  const abortSignal = controller.signal;

  client
    .collections('<collection>')
    .documents()
    .search(
      { 
        q
        // ...rest of query
      },
      { abortSignal }
    );
}

Expected Behavior

Request is aborted when signal fires.

Actual Behavior

Request is aborted, and this makes the client cycle through the nodes. Initially from the LB -> first node, and then proceeding through the nodes in order every time a request is interrupted, wrapping around to the first one at the end.

Metadata

Typesense Version: 1.8.2, 1.7.2
OS: MacOS

[jasonbosco]

Thank you for reporting this issue.

The fix for this is now in v2.0.0-13. Could you give it a shot and let us know how it goes?

CC: @tharropoulos