Software Engineering Glossary

Request Waiting List

Also known as: Pending Requests Map Waiting List Callback Registry

Definition

A request waiting list is a structure a cluster node keeps to hold client requests that cannot be answered yet because they are waiting on responses from other nodes. Each entry maps a key, such as a correlation id or a log index, to a callback that decides when the request is ready to complete. As responses arrive asynchronously, the node looks up the matching entry and fires the callback once its condition, often a majority quorum, is met.

Key Takeaways

  • It exists because replication is asynchronous. A node must reply to other nodes and collect enough responses before it can answer the client, so the client request has to wait somewhere.
  • Each entry is a key mapped to a callback. The key is chosen to match the arriving responses, and the callback holds the logic for when the request is done.
  • The key is usually a correlation id for point-to-point messages, or the high watermark log index when waiting for a replicated entry to commit.
  • The callback fires only when its criteria are satisfied, most often when a majority quorum of acknowledgements has arrived.
  • Entries need expiry. A timeout sweeps stale requests and fails them so a lost response does not leak memory or hang a client forever.

How It Works

  1. A node receives a client request that needs agreement from other nodes before it can respond.
  2. It sends the work to the other nodes asynchronously and adds an entry to the waiting list, keyed on a correlation id or a log index, with a callback.
  3. Responses from other nodes come back out of order and the node looks up the matching entry by key.
  4. The callback counts responses and, once a quorum is reached, completes the client request and removes the entry.
  5. A background timeout expires entries that never gather enough responses, failing those requests cleanly.

Where It Is Used

  • Raft and etcd hold client proposals until the log entry they created is committed by a majority, then answer the caller.
  • The Kafka producer keeps produce requests pending until the required acks arrive from the in-sync replicas.
  • A Cassandra or Dynamo-style coordinator waits for a read or write quorum of replica responses before replying to the client.

Related glossary terms