The OpenTelemetry Magic Show
By the time you notice the chains, you’re locked in.
On September 21, 1912, Harry Houdini, the great magician, performed one of his most famous tricks.
He was shackled in chains, padlocks tight, lowered upside down into a glass tank filled with water.
Hidden from view by drapes, he would attempt to escape while the crowd anxiously watched.
To the audience, it was true magic: a man drowning one moment, surfacing free the next.
They gasped. He bowed. Magic is real.
But behind the curtain, there was no magic.
Just hidden locks, escape drills, and a system he’d mastered long before stepping on stage.
Today, observability vendors run this magic show.
To us, it’s pure magic: one-click setup, instant metrics, dashboards light up like your eyes watching a video of a puppy and a baby falling asleep together.
Behind the curtain, your data is transformed to match their system, often diverging from the open standard.
“What kind of sorcery is that?”, you ask.
Before you know it, you’re underwater, shackled in chains. Locked-in.
Chains Beneath the Water
As long as you stay on stage, the magic works; drop in their agent and telemetry kicks in.
It feels like observability is solved. No OpenTelemetry to fight with, no missing logs.
The trick is invisible: OpenTelemetry data is reshaped, fields are normalized to look consistent, and gaps are smoothed over before you ever notice them.
You never see the work behind the curtain. And rightfully so, this is what you paid for.
The trouble begins when you want out.
Costs creep up as data volumes grow: more instances, more hosts. Finance asks why the bill doubled last quarter. Default pipelines can’t bend to new routing needs.
The moment you try to migrate off, the chains reveal themselves.
Metrics behave differently than the standards expect, attributes drift from service to service, and dashboards that looked perfect now show conflicting numbers.
Inside the closed system, counters always seemed correct; the moment you run the same telemetry through an open backend, the totals shift, fields go missing, and queries break.
We seek control, but we’re left like a skill-less Houdini.
Shackles SDK
It starts at the entry point. Vendors say they support OpenTelemetry, and technically, they do - many contribute to the project and ecosystem.
You install their OTel SDK or drop in their agent, and telemetry begins to flow.
It feels like you’re using the standard, just wrapped in a more convenient package.
But hidden inside are extra locks: vendor SDKs are built around OTel but wired to their pipelines by default; auto-instrumentation adds their extensions; default fields follow their naming, not always the spec’s.
On paper, it’s OTel; in practice, it’s a fork of reality.
Try to migrate, scale, or correlate data outside their system, and the cracks split wide open.
The telemetry you thought was portable isn’t.
Or as Dua Lipa put it: ‘Catch me or I go, Houdini.’”
(Never thought I’d quote Dua Lipa in an observability article… but here we are.)
What started as “just an OTel wrapper” can quietly become shackles around your stack.
Illusion vs. Reality
The illusion is that observability is something you set up once. Quick and easy; no mastery needed.
Agents send data, and you assume everything is captured. Reliability feels automatic.
Outside, the reality is far messier; you’re never sure if telemetry is complete.
A collector under load silently drops logs or metrics. An attribute survives in one region but vanishes in another. Different services report the same field in three different ways.
You think k8s is instrumented, only to realize the metrics you needed never appeared.
Cloud logs that looked easy on paper demand weeks of wiring before they’re reliable.
None of it is automatic. All of it takes work.
Migration off of vendors magnifies the uncertainty. How long will it take to move? How do you know every service is instrumented? When do the bills stop? How do you prove nothing was lost?
Even the “vendor-neutral” data pipelines that promise freedom can become another set of chains; once everything flows through them, the control is theirs - another illusion.
Observability with OpenTelemetry isn’t just capturing and visualizing data.
It’s relentless work; tracking every field, every label, every metric, ensuring nothing vital gets lost between the source and your dashboard.
The Great Escape
Houdini never escaped by magic.
He stepped into the tank, chained and breathless, just like the audience saw.
What they didn’t see was the hours spent studying every lock, rehearsing every move, preparing for the moment when the curtain dropped.
That’s the reality for anyone inside the observability water tank. From the outside, it looks like magic. From the inside, it feels like drowning.
It’s the observability vendors’ magic show; wave their hands, tell the tale, the crowd applauds.
If we wish to get out of that water tank, we have to know your system end-to-end; own and see everything so well that you can break free. Embrace and master the complexity.
From the seats, it looks like the chains are gone.
But we’re not in the seats. We’re Houdini underwater.
OpenTelemetry is the way out.
Marketing Magic (ugh I know, read only if you’re looking for a solution)
What if adopting OpenTelemetry felt like magic? The chains fell away as you surfaced free to the crowd’s applause like Houdini himself. Lawrence helps you manage all your OpenTelemetry deployments at scale. So you don’t end up shackled underwater. Get into the private beta!
P.S. Houdini did not die doing this trick - it’s a myth!
P.P.S. We love everyone.