Need an Azure learning path for an internal transfer to DevOps (2-3 month timeline)
https://redd.it/1uqjb9j
@r_devops

Optimising workspace for multiple laptops?

What setup have you put in place on your desk to manage being on multiple laptops?

https://redd.it/1uqm1f2
@r_devops

Why are intermittent production bugs so hard to reproduce?

I had one of those bugs last month that made the whole team question reality for about four days. A checkout service would randomly throw 500s, maybe 3-4 times a day, always on the same endpoint, never on a predictable schedule sometimes it happened during peak traffic, sometimes at 3am with almost nobody hitting the system. Logs showed the error but the stack trace pointed to a null reference in a place that should have been impossible given our validation logic. We tried the obvious stuff first: replaying the same request payload from the failed logs, same user, same cart contents, down to the byte worked fine every single time locally and in staging couldn't reproduce it on demand no matter how hard we tried. Turned out three things had to line up at once to trigger it, a specific cache eviction timing on one node, a race condition between two async calls that only mattered under a certain concurrency level and a stale feature flag value that got cached slightly differently depending on which pod served the request none of that shows up in a single request replay because the bug wasn't really about the request, it was about the state of the system at that exact moment. This is why intermittent bugs are so brutal to chase down the failure depends on system state not just input, things like cache state, connection pool exhaustion, background job timing or which pod handled the request all shift constantly and aren't captured by a normal log line. Concurrency and timing windows are nearly impossible to force, since race conditions might only trigger under load patterns that don't exist in staging or in a single manual test and by the time you are looking at it, the state that caused it is already gone. logs tell you what happened not the exact conditions leading up to it, so you are reconstructing a crime scene from a few photos instead of watching it happen. We eventually solved it by adding much more granular tracing around the cache and flag evaluation logic then waiting for it to happen again with all that extra visibility on took another two days after that just to catch it in the act. How do other people approach this class of bug, better tracing and waiting to catch it live or actual techniques for forcing race conditions to surface faster before they become customer-facing?




https://redd.it/1uqm2hr
@r_devops

What are you guys using for KYC API for identity verification and onboarding?

Getting straight to the problem: our current KYC API is hitting its limits. Slower verification times, too many false rejections on international documents and support goes quiet the moment something breaks in production.Every vendor sounds identical on paper so we fail to understand what does AI powered verification actually mean when an edge case hits at 2am(lol) and thousands of verifications are queued up? Looking for something API first with good docs, reliable webhooks, global document support, and AML screening that is baked in. Anyone who has run more than one of these in production and can give an honest take would be really helpful.

https://redd.it/1uqookv
@r_devops

Do people actually use their GPUs as much as they expected?

When I first started looking at upgrading my setup, I was convinced I'd be using a GPU every day.

Reality turned out to be pretty different.

Some weeks I barely run anything. Then I'll have two or three days where I'm constantly launching jobs.

Now I'm wondering if most people actually use their hardware that much, or if it mostly sits there waiting for those busy days.

https://redd.it/1uqpme2
@r_devops

Could you please share a typical daily routine for a DevOps professional and outline the types of tools used for application hosting?

The objective here is to investigate the DevOps tools currently employed by various companies within the market, thereby enabling adequate preparation for a potential career transition.

Also share the year of experience do you have?

https://redd.it/1uqqjho
@r_devops

How do you use agents and LLMs in your work

I was wondering how you guys use LLMs and agents in your day to day work, do you use them to write yaml/boilerplate or beyond that?

https://redd.it/1uqsg66
@r_devops

Docs as Code implementation for Infrastructure

Hi there

Recently I was tasked to write documentation for our infrastructure in "doc as code" way but I have not very well grasped what it is

The only requirement my team leads has is that the documents should be enough for any new person to understand our infra setup and tools we are using.

They also mentioned that any changes in the documents should have a PR and only after reviewing and approving any changes should be visible.

What I understand till now is that we would have a central repository in confluence or version control with documentation files.

There should be a way to navigate to different documents

All .md files are similar in structure, how they are written

Architecture diagrams to show infrastructure

I had a look at kubernetes documentation as I get what it is everything is in markdown it is being rendered to the website and has different documents for different versions.

But I still have no idea how to start on this.

Can I know what are some common points to note down or industry standard for these kind of documentation. And how to implement it

https://redd.it/1uqwjuf
@r_devops

Looking for advice on transitioning from Sysadmin to DevOps

I'm looking to apply to devops/sre positions to change my current job and have a profesional glow up, i no longer feel challenged project from my job and i am stating to just doing app maintaenance and helpdesk tasks.

I am a sysadmin with hands on production enviroments and automatation background (scripting and low code) but where i learn and enjoy the most is in my homelab, I have around 2.5 years of professional experience.

I'd like to learn technologies such as Terraform (mainly because I see it requested in many LinkedIn job postings) and Ansible, as well as deepen my knowledge of CI/CD pipelines. I've already worked with GitHub Actions.

I've also used AI to help me create a learning roadmap and prioritize milestones. One of the strongest recommendations was to document everything on GitHub.

Beyond following a roadmap, I'd like to hear what you think are the most important things to focus on when transitioning into DevOps. If you've seen what helps people land their first DevOps role, or if you have any advice on common mistakes, skills to prioritize, or portfolio ideas, I'd really appreciate your perspective.

https://redd.it/1ur3a7j
@r_devops

Additional burden of hosing AI apps

With AI, business and product teams are creating apps left and right. They dont understand what the code is doing, no clue about security or how to host it.

This burden falls on DevOps/Engineering to now maintain it, fix it. Authors are still considered the owners of these apps. I wanted to know how are you guys handling this situation?
\- Is Engineering/DevOps the defacto owners of such apps in your company?

\- How are you deploying these - in your prod AWS or some hosted env?


TIA



https://redd.it/1ur4nwc
@r_devops

How to use my old laptop

Anyone help me what I can do useful with my old laptop related to server or something and I don't have router.

https://redd.it/1urbjdi
@r_devops

What is your optimal infrastructure

we're all striving to optimize our infra, but what would you say is the PERFECT infra configuration in your mind?

I'll go first

# Product:

micro services
each service has its own git repo following a standardized directory structure
each service has its own pipeline that scans the code and runs a full regression suite before a PR is merged to master/main
each service has a health check that is actually accurate to the state of the service
launch time for the services are minimal
daily pipeline regression suite that updates a global dashboard highlighting all the current repo statuses
SaaS ONLY (onprem is pain)

# Deployment:

terraform code base that is multiplatform, can be deployed to any cloud
deployment process is executed by a pipeline with optional parameters for configure what kind of environment to deploy
all services hosted in GKE/EKS or some other cloud hosting platform
auto scaling horizontally and vertically
all logs routed to logz.io and tagged by environment and service
all deployed infra is tagged by its environment name

# Monitoring:

new relic, all the things go to new relic. (not sponsored, i just used it in the past and it was a dream when setup correctly. was just too expensive for my previous company at the time)


one thing i did learn is that the less you "have" to do. the better. and if your company can afford to outsource something, your life will be much easier.

https://redd.it/1urdygd
@r_devops

What's up with GitHub runners lately?

https://preview.redd.it/yzae6azo76ch1.png?width=718&format=png&auto=webp&s=6a1dbf5a366b67da18cc08e2b898d8a764cf0580

Too much AI slop to build? If this continues I will probably prefer to just use my self hosted ones for all jobs.

https://redd.it/1urkyux
@r_devops

How should I stay curious and sharp and learn more? What resources should I go to?

Hello, I have been working as a dev sec ops for the last 4 years and recently as a devops engineer.

At my last job, since everything was new, and I had no idea about so many things everything was exciting. Learning Terraform, ansible, Linux, how to sys admin, Grafana monitoring so on.

Now at this new job, I realized that ok, I am not super familiar with CI/CD, I could use some work there, but the rest of the things I had to do, were needed in such a hurry that I did not have the time to even understand stuff, so as you know, AI (mostly Claude).

Anyway the thing is with all this AI here and there, how do you stay motivated to learn, where do you turn to, how should I go about it?

I feel like my brain is getting numb lately, and I want to do something about this, I do not want to end up being a mediocre human, just another prompter (not even good at it).

Thanks a bunch.

https://redd.it/1urnqrt
@r_devops

Is there an orchestration tool that doesn't force everyone on the team to learn python ?

Genuine question. Our team is 3 data engineers (python), 2 analytics engineers (SQL and dbt), and 2 infra guys (terraform / ansible). We tried airflow but the analytics and anfra pp can't contribute to pipeline definition bc everything is python DAGs. It creates a bottleneck where the 3 python devs end up writing and maintaining every single workflow.

I am looking for sth where defining run this dbt job, then this terraform apply, then notify Slack doesn't require importing fifteen python modules. Does this exist or am I dreaming?

https://redd.it/1uro23y
@r_devops

DevOps/Platform engineering vs. Software Dev, is it less creative and more rote work?

I have been a fullstack software dev for 5 years already, with some years also doing ops stuff for the team (since no one bothered to/liked doing it) like managing Jenkins IaC, pipelines, AWS CDK, K8S deployments, etc. I liked those stuff and our team was really suffering because no one bothered to take care of it so I took leadership there.

I am now looking for another job, since my contract ended. I just got an offer to work as a cloud engineer at another organization.

To be honest, I do like being a dev, but I could not really see myself being a "senior" or freelancer in this field one day. This is because I feel in software dev there is a lot of "openness" or options on how to do something and it is very highly opinionated, and it is hard to find the "correct" solution. For example with design patterns (do you need to apply patterns? do we need this abstraction/interface?), or with REST APIs (how do you design your endpoints), or with frontend design decisions (confirm button on right or left side? color? opacity? etc.).

And with DevOps, at least so far from what I see there is less "opinions" e.g. you follow the vendor's directions, if it deploys and it runs then it's good (less edge cases), there is more standardized ways of doing something/deploying something, and also it is domain-independent.

In software dev, you have to understand the domain to make business impact, and that can take away a lot of time from coding itself.

It is also easier to prove yourself for other jobs through certifications, whereas with full stack there's no such luxury.

But the disadvantage I see with DevOps is that it is more stressful than a software dev position, for example through on-calls, although you do get paid for your extra hours so I think it compensates it somewhat. And being on call I think really teaches you to be a tough person mentally, able to say no to other people, not be a cry baby, so it helps also perhaps with self development.

And also with DevOps, it can be harder to try something out (you will need to have a free AWS account to try deployments, etc.) although I might be wrong here. And since there's so much breadth, you cannot understand the root cause of everything going wrong, but I may be wrong here.

What is your opinion here? Do you see DevOps as being less "uncertain" than fullstack, or is it not the case?

https://redd.it/1urufkh
@r_devops

Best approach to deploy 40+ React apps as microservices on a single server — Docker or k3s?

Hey all,

I'm running into an architecture decision and want some real-world input before I commit.

Setup: I have 40+ React apps that I'm treating as individual microservices — each one needs to run in its own isolated environment (separate dependencies, separate runtime context, no bleed-over between apps). I have one server with decent specs to run all of this.

The core tension:

If I containerize each app individually with Docker, that's 40+ separate containers, each with its own process overhead — memory adds up fast even though each app is fairly lightweight on its own.

k3s (lightweight Kubernetes) is the other option, but I'm not sure if the control plane overhead on a single node actually buys me anything here, or if it's just extra complexity for no real benefit since I don't have multiple nodes.

What I need:

Each "microservice" (React app) needs to stay in its own isolated environment — that part isn't negotiable, so a single shared process serving all of them isn't an option for this use case

Minimize per-app memory/resource overhead as much as possible given that constraint

Reasonably simple to deploy/update individual apps without redeploying everything

I'm fine with a setup that isn't fully HA/production-grade — this is a single server, and I can tolerate occasional hiccups in exchange for lower cost and complexity

Questions:

For 40+ isolated environments on one box, is plain Docker (Compose) genuinely more efficient than k3s here, or does k3s's overhead stop mattering once you factor in things like better resource limits/QoS per pod?

Any tricks people use to cut per-container memory overhead at this kind of scale (40+ containers) — smaller base images, shared kernel tricks, resource requests/limits tuning, etc.?

Has anyone actually run something like this in production and hit a wall around a certain container count on a single node?

Would appreciate input from anyone who's actually deployed at this density on a single machine rather than theoretical takes.

Thanks!

Ps: written this paragraph with the help of gpt, I am bad with words

Also I m new so don't bully me.

https://redd.it/1urzfrj
@r_devops

Best approach to deploy 40+ React apps as microservices on a single server — Docker or k3s?

​

Hey all,

I'm running into an architecture decision and want some real-world input before I commit.

Setup: I have 40+ React apps that I'm treating as individual microservices — each one needs to run in its own isolated environment (separate dependencies, separate runtime context, no bleed-over between apps). I have one server with decent specs to run all of this.

The core tension:

If I containerize each app individually with Docker, that's 40+ separate containers, each with its own process overhead — memory adds up fast even though each app is fairly lightweight on its own.

k3s (lightweight Kubernetes) is the other option, but I'm not sure if the control plane overhead on a single node actually buys me anything here, or if it's just extra complexity for no real benefit since I don't have multiple nodes.

What I need:

Each "microservice" (React app) needs to stay in its own isolated environment — that part isn't negotiable, so a single shared process serving all of them isn't an option for this use case

Minimize per-app memory/resource overhead as much as possible given that constraint

Reasonably simple to deploy/update individual apps without redeploying everything

I'm fine with a setup that isn't fully HA/production-grade — this is a single server, and I can tolerate occasional hiccups in exchange for lower cost and complexity

Questions:

For 40+ isolated environments on one box, is plain Docker (Compose) genuinely more efficient than k3s here, or does k3s's overhead stop mattering once you factor in things like better resource limits/QoS per pod?

Any tricks people use to cut per-container memory overhead at this kind of scale (40+ containers) — smaller base images, shared kernel tricks, resource requests/limits tuning, etc.?

Has anyone actually run something like this in production and hit a wall around a certain container count on a single node?

Would appreciate input from anyone who's actually deployed at this density on a single machine rather than theoretical takes.

Thanks!

Ps: chatgpt written this paragraph I'm bad with words.

Also I am newbie so don't be harsh😔

https://redd.it/1urzdrr
@r_devops

The biggest lie in GPU tooling is that RUNNING means ready

I keep seeing this with rented GPUs.

The pod says RUNNING, so you assume the workload is ready.

But RUNNING usually just means the infra exists.

It does not mean vLLM finished loading.
It does not mean ComfyUI is reachable.
It does not mean CUDA is visible.
It does not mean the container did not crash.

That gap is where a lot of GPU UX gets painful. You think the job is live, but really you are just paying while the app maybe starts.

The fix I’ve been using is separating infra status from app readiness.
vLLM should be checked at the app level.
ComfyUI should be checked at the app level.
Dead pods should fail fast.

Users should see:
starting GPU
checking app
ready
failed with reason

I’m working on making rented GPUs less annoying for AI jobs, mainly by treating them more like jobs than servers.

Curious how others handle this. Do you trust provider status or probe the actual app?

https://redd.it/1us3ile
@r_devops

Securing Services with Rootless Containers
https://blog.coderspirit.xyz/blog/2026/07/06/securing-services-with-rootless-containers/

https://redd.it/1us43p2
@r_devops