How are you handling these AWS ECS (Fargate) issues? Planning to build an AI agent around this…

Hey Experts,

I’m exploring the idea of building an AI agent for AWS ECS (Fargate + EC2) that can help with some tricky debugging and reliability gaps — but before going too far, I’d love to hear how the community handles these today.

**Here are a few pain points I keep running into 👇**

* When a process slowly eats memory and crashes — and there’s no way to grab a heap/JVM dump *before* it dies.
* Tasks restart too fast to capture any “pre-mortem” evidence (logs, system state, etc.).
* Fargate tasks fill up ephemeral disk and just get killed, no cleanup or alert.
* Random DNS or network resolution failures that are impossible to trace because you can’t SSH in.
* A new deployment “passes health checks” but breaks runtime after a few minutes.


**I’m curious**

* Are you seeing these kinds of issues in your ECS setups?
* And if so, how are you handling them right now — noscripts, sidecars, observability tools, or just postmortems?



Would love to get insights from others who’ve wrestled with this in production. 🙏

https://redd.it/1onzb40
@r_devops

Paid Study Help us improve Virtual Machine Tools – $150 for a 60-minute interview

We’re conducting a paid research study to learn more about how professionals create, manage, and provision virtual machines (VMs) at work. Our goal is to better understand your workflows and challenges so we can make VM tools more efficient and user-friendly.



Details:



\- Compensation: $150 USD for a 60-minute 1:1 conversation



\- Format: Online interview via Zoom or Teams



\- Who we’re looking for: Anyone who creates or uses virtual machines, at any experience level or for any type of application



\- Priority: Participants with a LinkedIn profile linked to our platform will be considered first



If you’re interested, please send me a message or comment below and I’ll share the next steps.

Your feedback will directly help improve the tools used by thousands of professionals worldwide.

https://redd.it/1oo1fe7
@r_devops

Tired of applying everywhere - Looking for Fresher DevOps / Cloud Support / Linux Opportunity

Hey everyone,

I’m a recent Computer Science graduate actively looking for fresher roles in DevOps, Cloud Support, or Linux.
I’ve applied to many companies and portals, but most either ask for experience or never respond — it’s been really tough finding that first break.

I’ve learned and practiced:

Linux
AWS (EC2, S3, IAM, Lambda basics)
Docker & Kubernetes
Git/GitHub
CI/CD concepts
I’m genuinely passionate about DevOps and Cloud, and I’m just looking for that first opportunity to prove myself.
Preferably looking for roles in Pune or remote.

If anyone here knows of openings or referrals, I’d really appreciate your help 🙏

Thanks a lot for reading and supporting freshers like me!

https://redd.it/1oo3d4e
@r_devops

India's largest automaker Tata Motors showed how not to use AWS keys

guy found two exposed aws keys on public sites, which gave access to \~70tb of internal data - customer info, invoices, fleet tracking, you name it

they also had a decryptable aws key (encryption that did nothing), a backdoor in tableau where you could log in as anyone with no password, and an exposed api key that could mess with their test-drive fleet

cert-in tried to get tata to fix it, but it took months of back-and-forth before the keys were finally rotated

link: https://eaton-works.com/2025/10/28/tata-motors-hack/ and https://news.ycombinator.com/item?id=45741569

https://redd.it/1oo402w
@r_devops

Those of you who switched from DataDog to Google Observability - do you miss anything?

The company I work for is switching from DataDog to Google's own offering, mostly driven by cost reasons. At surface level the offering seems to be par - but I wonder if we will discover things missing after it's too late?

https://redd.it/1oo36h7
@r_devops

Cache Poisoning: Making Your CDN Serve Malicious Content to Everyone 🗄️

https://instatunnel.my/blog/cache-poisoning-making-your-cdn-serve-malicious-content-to-everyone

https://redd.it/1oo6g66
@r_devops

data democratization aka automation and management of data platforms

Hi folks, Are you guys aware of any platforms that can help with management of a number of users on large datalakes, what i mean by this say u have a product like databricks and we want to "user-wise" manage how much access someone has, we wanna stream line this by maybe this flow , user raises a request somehwere -> automated noscript grants access -> access revoked automatically within a set time,
also log who had what access etc etc,
ofc a custom solution is possible but i was hoping for any opinions on if anything similar to this already exists.
Thanks for yuour time have agood one

https://redd.it/1oo9x3u
@r_devops

Learning friend

Is anyone here willing to learn Devops with me? I am a beginner

https://redd.it/1oo9js7
@r_devops

EKS Node Resource Limits

I am currently undertaking the task of auditing EKS Node resource limits, comparing the limits to the requests and actual usage for around 40 applications. I have to pinpoint where resources are being wasted and propose changes to limits/requests for these nodes.

My question for you all is, what percentage above average Usage should I set the resource limits? I know we still need some wiggle room, but say that an application is using on average 531m of Memory, but the limit is at 1000m (1Gb). That limit obviously needs to come down, but where should it come down to? 600m I think would be too close. Is there a rule of thumb to go by here?

Likewise, the same service uses 10.1mcores of CPU on average, but the limit is set to 1core. I know CPU throttling won't bring down an application, but I'd like to keep wiggle room there to, I'm just not sure how close to bring the limit to the average usage. Any advice?

https://redd.it/1oo78yq
@r_devops

GitOps role composition pattern for deployments?

Is anyone utilizing or has anyone utilized a cluster role-based composition pattern for deployments? Any other patterns?

Currently spinning up ArgoCD for current org and looking at efficiently implementing this for scalability.

At my previous org, we wound up having things a bit scattered about with \~30 AppSets and 30 applications (separate from appsets, for individual clusters).

It was manageable as we didn't change things much but I could see running into scaling issues as far as effort/maintenance goes down the road.

I would appreciate getting a second set of eyes to see if this makes sense or if I'm going to run into issues I haven't thought of: https://github.com/SelfhostedPro/ArgoCD-Role-Composition

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

How a tiny DNS fault brought down AWS us-east-1 and what devops engineers can learn from it

When AWS us-east-1 went down due to a DynamoDB issue, it wasn’t really DynamoDB that failed , it was DNS. A small fault in AWS’s internal DNS system triggered a chain reaction that affected multiple services globally.

It was actually a race condition formed between various DNS enacters who were trying to modify route53

If you’re curious about how AWS’s internal DNS architecture (Enacter, Planner, etc.) actually works and why this fault propagated so widely, I broke it down in detail here:

Inside the AWS DynamoDB Outage: What Really Went Wrong in us-east-1 https://youtu.be/MyS17GWM3Dk

https://redd.it/1ooi45v
@r_devops

What guardrails do you use for feature flags when the feature uses AI?

Before any flag expands, we run a preflight: a small eval set with known failure cases, observability on outputs, and thresholds that trigger rollback. Owners are by role and not by person, and we document the path to stable.

Which signals or tools made this smoother for you?

What do you watch in the first twenty four hours?

https://redd.it/1oo5u1m
@r_devops

LeetCode style interview for DevOps role

Curious if anyone has done any LeetCode style interviews recently?

Recently interviewed for a Senior DevOps role at a FAANG adjacent company which was a 6 stage process.

I thought I was doing pretty well after going though multiple stages doing system design, architecture, reliability engineering, scenario based troubleshooting etc, and even got through some coding exercises in Python.

One of the interviewers was changed last minute. I was told it would purely be a cultural fit type of interview but it ended up being a couple of LeetCode style problems which completely threw me off and I kinda of bombed and struggled to get through them.

I'm fairly experienced with Python but never learned DSA as I don't have a software engineering background and was frustrated to get failed on this after everything.

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

Terraform + AWS Questions

So i'll try to keep this brief. I am an SDET learning Terraform as well as AWS. I think I mostly have "demo" stuff working but I wanted to just pose a list of questions off the top of my head:

1. Right now I think one s3 bucket per AWS account makes the most sense (for storing state). From my understanding the "key" is what determines both the terraform state file path as well as the LockID. However I am not sure if for example you define a backend s3.tf file, does the LockID use the key or the key+bucket name?
2. Sort of a follow up to #1, any suggestions for naming conventions when it comes to state files key? Something like environment+project+terraform/state.tf or similar?
3. When it comes to Terraform, I know there is the chicken and the egg sort of thing. What's the proper way to handle this? Some sort of bootstrap .tf file? From my understanding basically you would do that OR set up the s3 bucket manually and then import it? How does that usually go?
4. What are the main resources you think a newcomer should start focusing on as far as tracking? Right now i'm just doing the backend s3 and beanstalk (app and enviornment_ and rds currently.

https://redd.it/1ookd9a
@r_devops

Tofu/Terraform Modules for enterprise

So I'm looking to setup a tofu module repo, all the examples I can find show each module has to have its own git path to be loaded in.

Is there a way to load an entire repo of modules? Or do I have to roll a provider to do that?

I just want to put the classic stuff in place like tag requirements and sane defaults etc.

I got the backend config sorted but putting it in the pipeline templates so each init step gets the right settings. But struggling with the best way to centralize modules.

We are using tofu if that matters.

https://redd.it/1ooph4x
@r_devops

Need advice on deployment and dev ops

Built a simple wrapper around chatgpt for an internal audit my company and now they want it deployed company wide. I’ve never deployed something at a company, never even knew what a Linux box was until my IT team asked if I would be able to manage it which I obviously said yes too.

Looking for advice on how to best host and deploy because I’m going to have to be the one to manage it.

I have a python app wrapped in a fast api, that sends PDFs to OpenAI api for analysis and then returns the response on a basic streamlit UI. 2000-4000 6-10 page PDFs needs to be run through it monthly at scale. What’s the best way to get there. I’ve used render, but only on the free plan to demo it, now I’m pretty lost.

Any help would be great! My outsourced IT team says the solution is a Linux box which will take 10-14 days to set up. Company is ~90mm ARR, 300 employees.

I have no formal swe experience, I still have to ask the AI in cursor to run the commands to push things to GitHub. Please explain like I have basic knowledge, I will look up anything I don’t know.

https://redd.it/1oopug3
@r_devops

Insecure Direct Object References (IDOR): The $1 Billion Authorization Bug 🔢

https://instatunnel.my/blog/insecure-direct-object-references-idor-the-1-billion-authorization-bug

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

I wrote zigit, a tiny C program to download GitHub repos at lightning speed using aria2c

Hey everyone!
I recently made a small C tool called zigit — it’s basically a super lightweight alternative to git clone when you only care about downloading the latest source code and not the entire commit history.

zigit just grabs the ZIP directly from GitHub’s codeload endpoint using aria2c, which supports parallel and segmented downloads.

Check it out at : https://github.com/STRTSNM/zigit/

https://redd.it/1oownb2
@r_devops

Stateful or Stateless IaC?

I've been debating this topic relentlessly. What is better? Infra as Code, which maintains states or stateless that work directly with the resources?

View Poll

https://redd.it/1ooxi57
@r_devops

Feedback

We’re two founders building an AI system that automatically detects, predicts and fixes website/app errors in real time, think Tesla Autopilot for debugging in DevOps. 

We’d love to learn from you, engineers, founders or DevOps folks for 10 minutes about how you currently debug issues. 

Not selling anything, just trying to validate if this could save teams a significant amount time. 

Happy to share a summary of what we learn + offer early access! 

https://calendly.com/aarittaparia/30min 

If you don’t have time, we would appreciate if you could fill this form: https://rc60edu0zkd.typeform.com/to/YixyC7S7 

Thanks so much! 

https://redd.it/1ooyk0s
@r_devops

Any tips on places where i can train as aspiring devops?

Hi, currently working in small company and finishing my college degree in few months.

I got interested in devops around half year ago and trained linux, git, github, github actions + Jenkins, docker hub. Built pipelines on simple projets, even did some tests.
Also got my hands on deployment with kubctl but there is a lot i have to learn yet.

Back to the question. Coders have codewars and leetcode. I wonder if there is any site for devops?
I found Qwiklabs for GCP however i was wondering what about the rest? Like solving problems or using part of the knowledge to try fixing something more difficult?

I kind of want commercial experience..

https://redd.it/1oozs5r
@r_devops