Retry backoff narrated tour

Retry backoff narrated tour

animation
http
protocols

Retry backoff narrated tour

Play silently with captions on provisional storyboard timing; narration audio is pending. Production pacing is not protocol time. Choose Reduced motion for a complete still at each beat.

Play the tour to open the workbench. Pause, replay a beat, or read the full transcript below. Situations marked CANNOT SHOW keep the existing frame and provide captions only.

Full transcript — Retry backoff (8 beats)

B1 — Reconnect storm
You keep a fleet of connected sensor dashboards running. After a server restart, their connections fail together. Everyone tries to reconnect together, and the burst can crash the recovering server. Your users need live readings to return. How should their dashboards choose when to try again?

Model caption: Dashboard connections fail together. Matching timers can send a burst back to a recovering server.

B2 — Exponential waiting
Backoff means waiting before another attempt. Exponential backoff makes successive waits longer by multiplying them. Jitter gives clients different waits, spreading their attempts over time. In the retry tuner, start with a four-second base and a multiplier of two. Leave jitter aside while calculating the capped wait.

Model caption: Start with four seconds. Multiply successive waits by two. Apply the forty-five-second cap before adding any jitter behavior.

B3 — Capped fifth wait
Suppose a device still needs its fifth retry. Starting from four seconds, double four to eight, then double eight to sixteen. Double sixteen to thirty-two, then double thirty-two to sixty-four. The maximum delay is forty-five seconds, so choose the smaller value: forty-five seconds. That is the wait before the fifth retry, rather than the total time since the outage.

Model caption: Hypothetical fifth wait: four doubles to eight, sixteen, thirty-two, then sixty-four seconds. Choose the smaller of sixty-four and forty-five: forty-five seconds. A wait is not elapsed outage time.

B4 — Full jitter spread
Now add the tuner's full jitter at eighty percent spread. Subtract eighty percent from one hundred percent, leaving twenty percent. Twenty percent of the first four-second wait is zero point eight seconds. The remaining three point two seconds provide the spread. Simulated clients get waits from zero point eight up to, but less than, four seconds.

Model caption: One hundred minus eighty leaves twenty percent. Twenty percent of four seconds is zero point eight; the remaining spread is three point two. First waits are at least zero point eight and less than four seconds.

B5 — Server backoff hint
The tuner also offers a switch to respect a server backoff hint. Retry-After means the server supplies a minimum wait. The selected reconnect scenario has no such hint, so it does not change our calculation. When a hint applies, the tuner uses it to extend a shorter local wait.

Model caption: Respect the server minimum when supplied. The selected reconnect preset supplies no hint, so the worked wait is unchanged.

B6 — Retry safety and budget
The tuner's safety notes limit which failures deserve another attempt. Permanent errors should not be retried. An idempotent operation can be repeated without creating duplicate effects. Only retry operations that are safe to repeat. Keep the retry budget enabled too: it stops attempts at the configured limit or deadline, with success also ending retries.

Model caption: Retry eligible failures with repetition safety. Stop on success, the retry limit, or the deadline. The operation switch assumes safety; it does not demonstrate duplicate prevention.

B7 — Fixed wait misconception
The misconception is that giving everyone the same fixed wait spreads the load. Clients that fail together and wait equally still return together. The burst moves later. For your dashboards, different waits spread reconnection attempts across time while allowing them to reconnect automatically.

Model caption: The same fixed wait shifts a shared burst later. Different waits spread automatic reconnection attempts across time.

B8 — Bounded recovery
Give recovery room: spread safe retries across time, and stop within a budget.

Model caption: Give recovery room: spread safe retries across time, and stop within a budget.

Open the Retry Backoff Tuner

Back to the chapter