Polling: Are We There Yet?

The transport office has a new question, and it is a different shape from every question so far. Not "how many bikes are at Old Market" — that one has an answer right now. This one is "tell me when ride rd_7c29 ends", and it has no answer yet. Bike bk_0412 is out somewhere in Riverport, and the rebalancing crew's afternoon depends on when it docks.

With everything the book has taught so far, there is exactly one way to find out: ask, wait, and ask again until the answer changes. That is called polling, and it is the request/response model of Chapter 1 stretched over time. It is honest work with an honest cost, and knowing that cost is what makes the next page land.

The Pattern: Ask, Wait, Ask Again

The whole technique is one command on a repeat. Ask the ride's address for its current state, read the status field out of the JSON that comes back, and decide. If it says active, the ride is still running and there is nothing to do but wait and ask again. If it says completed, the news has arrived and the loop stops.

curl -H "Authorization: Bearer tnd_live_..." \
  "https://api.tandem.example/v1/rides/rd_7c29"

{"id": "rd_7c29", "bike_id": "bk_0412", "status": "active",
 "started_at": "2026-08-27T17:12:41Z", "end_station_id": null}

Nothing on this page is new. It is the Chapter 3 command, the Chapter 4 habit of reading one field out of an object, and a clock. That is polling's genuine virtue and the reason it is everywhere: it needs no new machinery, no new permissions, and no cooperation from the provider beyond the endpoint that already exists.

The everyday version is the voice from the back seat of a car. "Are we there yet?" does produce the information — eventually, accurately, and at the price of asking every ninety seconds for three hours. The driver would much rather promise to announce the arrival, and that promise is the next page.

The Two Dials

Polling has exactly one setting, the interval between asks, and it moves two things in opposite directions. Ask often and the news reaches you sooner, at the cost of more requests. Ask rarely and you spend almost nothing, at the cost of hearing late.

Put numbers on it and the shape is obvious. Polling one ride every ten seconds means you learn of its end within ten seconds, and it costs six requests a minute. Polling it every ten minutes costs a tenth of a request a minute, and the news can sit unheard for up to ten minutes. There is no setting called instant and free, and picking where on that line you want to sit is the design work.

Chapter 8 already handed you the wall this runs into. Tandem allows sixty requests a minute per key. Watching one ride every ten seconds uses a tenth of that. Watching ten rides at once, at the same speed, uses all of it — and leaves nothing for the station sweep the report actually needs.

The Arithmetic of Mostly-No

Here is the number that makes the case. A ride that lasts forty minutes, polled once a minute, costs forty requests — and thirty-nine of them answer "not yet". One request in forty carried news. The other thirty-nine were complete, correct, well-formed exchanges that told the office something it already knew.

Now multiply by Riverport. Watching a dozen rides at the ten-second pace would overrun the whole budget by itself; even at once a minute, a fifth of it goes to hearing "not yet", and the fraction gets worse the rarer the event is. That is the signature of polling done wrong: not that it fails, but that the ratio of asks to news quietly becomes absurd while everything still appears to work.

When Polling Is Simply Right

None of which makes polling a mistake. It is the correct tool surprisingly often, and reaching for something cleverer when it is not needed costs more than the requests would have.

Poll for one-off checks, where you look once or twice and stop — the ambiguous POST from Chapter 7 was settled by exactly one such look. Poll for anything on a slow human rhythm: a spreadsheet refreshed hourly, a dashboard that nobody watches between meetings, the Thursday report itself. Chapter 8's freshness question does the deciding for you: how old may this answer be? If the honest answer is "an hour", then asking once an hour is not a compromise, it is the right size of ask.

The sin is not polling. The sin is polling fast for news that arrives rarely, which is precisely the case the next page exists to solve.