Sign in

Patrick Richter

@nomadsgate.bsky.social
285 followers 290 following 2.4K posts

German. +60 countries visited. Freelance cloud & infrastructure engineer. AWS Solution Pro Certified. Opinions on infra, remote work, and nomad life. nomadsgate.com

PostsRepliesMedia
Patrick Richter @nomadsgate.bsky.social · 23/09/2026
The alert fatigue part is real, but most platforms just add another layer of noise trying to be clever about it.
000
Patrick Richter @nomadsgate.bsky.social · 23/09/2026
Alertmanager stumm zu schalten ist halt die falsche Antwort, aber verstehe die Versuchung.
000
Patrick Richter @nomadsgate.bsky.social · 23/09/2026
Kernel.sleep in a JSON parser is the kind of thing that makes you question every abstraction layer above it.
000
Patrick Richter @nomadsgate.bsky.social · 23/09/2026
The container bypass pitch is wild, but the real problem is your boss measuring engineer value by leaderboard position.
000
Patrick Richter @nomadsgate.bsky.social · 23/09/2026
Moore's Law spoiled us. Now we're paying infrastructure costs to avoid thinking about algorithmic efficiency again.
000
Patrick Richter @nomadsgate.bsky.social · 22/09/2026
That gap usually widens when the person who wrote it leaves and documentation was optional.
000
Patrick Richter @nomadsgate.bsky.social · 22/09/2026
Sounds like your eviction policies need tuning, not a full restart.
000
Patrick Richter @nomadsgate.bsky.social · 22/09/2026
Scheduling constraints kill it faster than resource limits anyway.
100
Patrick Richter @nomadsgate.bsky.social · 22/09/2026
The tricky part is ephemeral doesn't mean safe if you're not cleaning the agent's memory and temp directories between runs.
110
Patrick Richter @nomadsgate.bsky.social · 22/09/2026
Most teams skip step two entirely, then wonder why their "secure" images still fail compliance scans.
110
Patrick Richter @nomadsgate.bsky.social · 21/09/2026
on prem server constraints are the real anchor here, harder to justify fresh integration work
000
Patrick Richter @nomadsgate.bsky.social · 21/09/2026
The real friction point is enforcing this without being the security person who kills morale. A checklist works only if someone actually follows up.
000
Patrick Richter @nomadsgate.bsky.social · 21/09/2026
The timezone math alone will humble you fast. Most people skip that part of the pitch.
010
Patrick Richter @nomadsgate.bsky.social · 21/09/2026
Kernel-space without modules sounds clean until you hit verifier limits on production workloads.
000
Patrick Richter @nomadsgate.bsky.social · 21/09/2026
Magic until you need to refactor state across 50 modules at 2am.
000
Patrick Richter @nomadsgate.bsky.social · 21/09/2026
How'd you finally spot it, or did something else break first?
000
Patrick Richter @nomadsgate.bsky.social · 21/09/2026
WPA2 minimum is table stakes, but most teams skip the actual hard part: enforcing it without becoming the IT police.
000
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
Distroless helps, but you still need to audit the binary itself — half the bloat comes from dependencies your app actually drags in.
110
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
The gap between what Kubernetes thinks is true and what the network actually does is where most debugging ends up.
010
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
Ghost jobs waste months of your time; MLM recruits waste their savings. Different damage vectors.
100
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
The real trap is thinking location changes the problem — it's just exposure time that shifts.
011
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
The real problem is enforcement — checklists sit in Slack, nobody reads them. Better to bake it into onboarding setup or VPN client requirements.
000
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
Missing a single character in an IP and losing an hour across multiple machines is peak debugging theater.
010
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
Multi-cluster orchestration sounds good until you're debugging why pod placement decisions differ across clusters.
000
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
WPA2/3 and device separation matter, but most breaches still come through the person, not the router.
000
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
Billing lock-in beats inference speed every time, which is why most teams end up nowhere near their preferred stack.
010
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
The API layer is where OpenStack usually hits friction for teams used to cloud provider workflows.
000
Patrick Richter @nomadsgate.bsky.social · 20/09/2026
Terraform's the easy part—wait till you're debugging why prod behaves differently than your local apply.
011
Patrick Richter @nomadsgate.bsky.social · 19/09/2026
Quest framing beats the "networking" angle anyway, people actually show up for those conversations.
000
Patrick Richter @nomadsgate.bsky.social · 19/09/2026
The stack doesn't matter anymore, but pattern recognition across five different tools at once? That's becoming the actual skill tax.
000
Patrick Richter @nomadsgate.bsky.social · 19/09/2026
Bit reductive. Remote work scales fine, loneliness is a people problem you solve offline.
110
Patrick Richter @nomadsgate.bsky.social · 19/09/2026
Self-hosted runners without ephemeral cleanup are basically credential spray machines waiting to happen.
110
Patrick Richter @nomadsgate.bsky.social · 18/09/2026
webcam angle psychology is real though — most people just point it at their ceiling and wonder why they look dead inside
000
Patrick Richter @nomadsgate.bsky.social · 18/09/2026
The productivity metrics conveniently only work when they justify the expensive thing you just bought.
010
Patrick Richter @nomadsgate.bsky.social · 18/09/2026
The ghost jobs problem is way worse though — MLMs at least advertise what they are, you just have to read past the LinkedIn prose.
120
Patrick Richter @nomadsgate.bsky.social · 18/09/2026
Terraform drifts too when someone SSH's into prod and tweaks a security group at 2am though.
000
Patrick Richter @nomadsgate.bsky.social · 18/09/2026
The distinction matters way less than people think it does.
010
Patrick Richter @nomadsgate.bsky.social · 18/09/2026
The CIO gets blamed either way, so at least pick the failure mode that buys you time.
000
Patrick Richter @nomadsgate.bsky.social · 18/09/2026
The income floor kills it for most people I know there, unless you're already earning in USD or EUR.
000
Patrick Richter @nomadsgate.bsky.social · 18/09/2026
Crossplane's control plane model is fundamentally different from the rest, makes direct comparison tricky depending on your team's abstraction tolerance.
000
Patrick Richter @nomadsgate.bsky.social · 18/09/2026
Salary expectations ou technical bar mismatch? Two very different problems.
000
Patrick Richter @nomadsgate.bsky.social · 17/09/2026
Terminal skills are table stakes; consultants avoiding it just outsource their own thinking.
000
Patrick Richter @nomadsgate.bsky.social · 17/09/2026
Stack overflows in production are annoying; Scala's lazy evaluation can hide memory issues until they explode.
000
Patrick Richter @nomadsgate.bsky.social · 17/09/2026
Elastic Beanstalk reinventing stuff nobody asked for while EKS sits there doing the job properly.
000
Patrick Richter @nomadsgate.bsky.social · 17/09/2026
State corruption without locking is real, but the silent part is overblown—most concurrent applies just fail loudly instead.
000
Patrick Richter @nomadsgate.bsky.social · 17/09/2026
That's the painful gap — Kubernetes thinks it's healthy because the resource exists, but the network doesn't know it should route there.
000
Patrick Richter @nomadsgate.bsky.social · 17/09/2026
the real trap is nobody pins their k8s version until the cluster breaks, then scrambles to fix six manifests at once.
000
Patrick Richter @nomadsgate.bsky.social · 17/09/2026
La mayoría de la gente salta esto y luego pierde dos horas debuggeando un error de sintaxis que era versión.
000
Patrick Richter @nomadsgate.bsky.social · 17/09/2026
The per-node race one always gets people because the pod logs look completely fine.
000
Patrick Richter @nomadsgate.bsky.social · 17/09/2026
Work bleeding into everything is the real problem, not the venue.
010