Pub/Sub Best Practices: Publishing

Publishing settings determine how long messages wait before being sent and how much work a Pub/Sub client can handle. Chloe explains the tradeoff between batching for throughput and sending ordered messages one at a time.

Key points

• Batching sends more messages together, but makes them wait: Larger batches can increase throughput while adding delay before publication. Chloe recommends choosing settings from message size, publisher throughput, cost, and acceptable latency 00:00–01:03(opens in a new tab).
• Ordering changes the batching recommendation: An ordering key identifies messages whose order must be preserved. Retrying a failed batch can disrupt publication order, so the episode recommends one-message batches and waiting for the server to receive one message before publishing the next 00:00–02:06(opens in a new tab).
• Its ordered-delivery recommendation is regional: It says ordering is guaranteed only within a region and recommends a regional endpoint so messages using ordering keys enter the same region 01:03–02:06(opens in a new tab).
• Publishers have client-side resource limits: Bandwidth, network latency, request rate, CPU, memory, and threads constrain throughput, while flow control can bound pressure 02:06–03:10(opens in a new tab).
• Timed-out requests are retried by the client library: A publish timeout produces a deadline-exceeded error. Chloe points to the initial timeout, retry multiplier, and total timeout as settings that can be adjusted rather than always accepting the defaults 02:06–03:53(opens in a new tab).

Full video: https://www.youtube.com/watch?v=Oo5PXA4idI8(opens in a new tab)