A collection of arcane, and perhaps at times the most inane of my thoughts.

Fail fast, fail often, but always fail in the open

A recent watch reminded me of an old discussion about silent failures. If I remember right, this is from my earlier times at crisp. One of those was with Santosh, my colleague (we spoke less tech and more about life towards the end).

Cutting to the chase - a practical and unfortunate case study in why systems should never fail silent.

1994, a Moscow-HongKong Aeroflot flight, had a pilot joined by his children. As providence would have it, the pilot had his children enter the cockpit. As a certain violation, he let his children be at the controls.

Come to think of it, that was a costly price to pay for isolating a bug that none had thus far known of in development. The safety net he operated with was the autopilot. Since it disables manual control when enabled, he was sure it would be fine to.

And that's where the bane of system design failures crept its sad, ugly face up.

When his son, following the daughter sat at the controls eager for his turn at it, an unseen situation comes up. He figures it is only an apparition of control and it is the autopilot handling things. But what he does next is something that none in there expected. Or, in earnest, none of the engineers who designed the system, too.

Looking at the show and nothing beyond, the lad of 16 exerts a force to push things, assuming it might work. What it does unfortunately, in a silent manner, disengage in part.

By design, the autopilot was to yield control to the pilots when the controls receive force. And it did, disengaging itself off the ailerons, under the control of the lad, at the wheel.

This would sound rather familiar; this is what the spec was, is a common statement after all. Not to mention, the original point I started with, silent failure/disengagement here.

What makes it painful are the bleak indications; a non-standard response for most. If one were to look at failure indications for too low altitude and the likes, this is nowhere sound (pardon the pun).

While lives of all 75 on board got lost, there's indeed something to take away as a grim reminder. Iterating and failing early is a good thing to have, be it a start-up, an upstart, or anything better. What's better or even the best, is never be one of the five men around the elephant. And, when the elephant at home might be your best chum, remembering what it could be in the wild, would save a lot.

Engineering, to use the Sanskrit term, must be sarvatomukha (gathering from all directions).

August 2021 (3)July 2021 (1)June 2021 (1)May 2021 (4)April 2021 (6)March 2021 (6)February 2021 (1)November 2020 (11)