The Ferryman's Leaking Bailer: On the Tool That Counts the Absent Water

Years ago, before I ever thought about pings or percentiles, I spent a summer helping an old man run a small passenger ferry. It was a wooden skiff with an outboard motor, shuttling hikers and birdwatchers across a narrow but deep tidal inlet. The boat was sound, but it took on water. Not a lot—just a persistent seepage through the planks, a quiet gathering in the bilge that was part of its character. The old ferryman’s solution wasn’t to caulk it perfectly dry. It was a rusted tin bailer, its bottom pocked with a dozen tiny holes.

He kept it lashed to a bench with a bit of frayed rope. Every few trips, without a word, he’d lean over, scoop the bailer into the inch of water sloshing below our feet, and heave its contents over the side. The ritual was as much a part of the journey as the pull of the tide. I asked him once, foolishly, why he didn’t just get a bucket that held water. He gave a dry chuckle. “This one tells me how busy I am,” he said. “A whole bailer means I’ve been chatting too long at the dock. A half-bailer means we’re on schedule. If it’s barely damp, I’ve been running non-stop and should check the engine for fever.”

The Metric of the Leak

I think of that leaking bailer now when I stare at dashboards of uptime and latency. We build our services to be as watertight as possible, and yet there is always a seepage—the incremental load, the background garbage collection, the slow queries. The instinct is to panic at any sign of ‘water,’ to chase a state of perfect, static dryness. But the ferryman understood something deeper: the value wasn’t in the absence of water, but in the rate of its arrival. The leak was a constant, gentle meter. The flawed tool, with its known, measurable inefficiency, was his gauge for the health of the entire system—boat, engine, tide, and his own attention.

In our world, we often build perfect, silent buckets for monitoring. They give us a binary: empty or full. They alarm only when disaster is already sloshing over the gunwales. But the ferryman’s wisdom was in instrumenting the normal, expected flaw. The holes in his bailer normalized the measurement. Each scoop wasn’t just maintenance; it was a data point. A sudden need for three scoops in a single crossing would have told him more about a worsening leak or an unbalanced load than any static inspection of the hull ever could.

Our health checks and synthetic transactions are often like that bailer. They are not the service itself; they are deliberately imperfect, artificial journeys taken to measure the environment. Their value is in their consistent, repeated failure to find water—or in the subtle increase in how much they bring up. The latency spike, the missed packet, the slow database connection—these are the scoops of water we count. A perfect, dry bailer would mean the instrument itself had failed, that we’d stopped checking.

I lost touch with the ferryman. The skiff, the motor, even the inlet may have changed. But the lesson remains etched: reliability isn’t the absence of all issues. It is the deep, rhythmic familiarity with their predictable accumulation, and having a tool—intentionally imperfect, always in hand—that turns that accumulation into knowledge. We are not just builders of boats. We are the keepers of the leak, and the counters of the absent water.

Notes & further reading

A few pages I came back to while writing this: