The One-Bit Lie

"'Is it online?' is the wrong question. The right one is 'is it what it should be?'"

The most natural fleet field in the world is online: true. It feels complete — the machine is up or it isn't. It is also a lie, because the same bit means opposite things depending on which machine happens to be wearing it.

Same bit, opposite meaning

An always-on server reading online: false is an alarm — something worth waking up for. A travel laptop reading online: false is a Tuesday — it's in a bag, asleep, exactly as planned. The boolean cannot tell those two apart. Only a policy that knows the server must stay up and the laptop may sleep can turn one identical fact into two different verdicts.

What the bit quietly hides

Even when a machine is reachable, one bit erases everything worth knowing: reachable but a build behind; reachable over one path and dead on another; up but drifted; up but its backup service silently unloaded three weeks ago. online collapses a dozen distinct questions into a single answer that resolves none of them.

The fix is more axes, not a smarter bit

You don't rescue this by inventing a cleverer boolean. You rescue it by splitting the one question into two — what should this machine be? and what did we actually see? — and computing the verdict from both. That split is the rest of this track.

Health is a verdict, not a reading. No single measurement is healthy or unhealthy until a policy says what that measurement was supposed to be.

Code

The bit that means two things·python

alive = ping(host)   # True / False. Now what?

# alive == True on a server whose nightly backup died 3 weeks ago
#          -> the dashboard says "fine"
# alive == False on a laptop asleep on a plane, exactly as planned
#          -> the dashboard says "down"?
#
# The bit is identical; the meaning is opposite. You cannot repair
# that by making the bit cleverer. You repair it by asking TWO
# questions instead of one:
#   1. what should this machine be?   (desired)
#   2. what did we actually observe?  (observed)
# Health is what you compute from the pair -- never the ping alone.

Exercise

Take one machine you run. Write 'online: true' at the top. Now list five real problems that machine could have right now while that line stays literally true. How many of the five would you actually want to be told about?

Hint

up-but-full-disk, up-but-wrong-build, up-but-service-dead, reachable-on-wifi-but-not-VPN, up-but-clock-skewed. 'online' cheerfully hides all five and reports a green dot.

The One-Bit Lie

The one-bit lie

Same bit, opposite meaning

What the bit quietly hides

The fix is more axes, not a smarter bit

Code

External links

Exercise

Progress

Comments 0