270
Reddit DevOps. #devops Thanks @reddit2telegram and @r_channels
How should we evaluate AI Agent observability platforms in 2026?
Disclosure: I work at Bonree, the company behind Bonree ONE, an AI observability platform. I’m asking about the general problem of choosing a platform, and I’m not trying to pitch ours.
AI agents are moving from chat interfaces into production workflows, and teams are now picking tools to observe them. Feature lists for AI agent observability platforms look alike, so I’ve been trying to write down questions that can actually be tested instead of ticked off.
Here is my current list.
1. Does it trace at the session level? An agent conversation has several turns, each with model calls and tool calls. I want to open the whole conversation as one unit, not look at isolated requests.
2. Which frameworks does it cover, and how deeply? Teams build agents with LangChain, LangGraph, low-code platforms such as Dify, and other frameworks such as OpenClaw. I care less about the length of the list than whether the frameworks my teams use are covered automatically, or only through manual work.
3. Is the evidence separate from the reasoning? A read-only tool call with the same input should return the same data, while model output varies. If a platform records them separately, I can tell a data failure from a reasoning failure.
4. Can it attribute token cost? By model, application, team, business line and time.
5. How does it handle sensitive content? I want to choose per field whether to store, sample, redact or never store it.
6. Can AI traces be linked to the services and infrastructure underneath? A slow answer is sometimes a database problem.
7. If it offers AI agents for operations, what are the limits? What can the agents do alone, what needs approval, and is every step logged?
8. What does the bill look like at my volume?
For the proof of concept, I’d instrument two agents built with different frameworks, replay three real incidents where an agent gave a wrong or slow answer, and measure how long it takes to find the cause in each tool.
A note on criterion 7: AI agent operations platforms, where agents diagnose incidents or inspect systems on observability data, are a newer category and the hardest to evaluate. We work on this at Bonree with Bonree ONE·Sage AI, so I would particularly like to hear how others test whether an operations agent is trustworthy.
Some things I’m still unsure about:
1. Which of these criteria do you weight most heavily, and which do you ignore?
2. How do you test “trustworthy” in practice, beyond reading a vendor’s description?
3. Do you prefer one platform for AI and infrastructure, or a dedicated LLM tool next to your existing stack?
4. What did your proof of concept reveal that the feature list did not?
https://redd.it/1xd6dxa
@r_devops
Test automation after deployment
How do you all perform full automated tests after deployment? Using postsync hooks in argocd or use Argo events? Do you use ephemeral or static environments?
https://redd.it/1xd92zw
@r_devops
Enforcing SAST and DAST Tools at Enterprise Level
Hi All,
It has been my experience at a couple of different organizations that almost all engineering teams own their own CI/CD pipelines. So my question is how do you enforce or ensure that these teams are performing SAST or DAST (Snyk, Trivy, IaC, etc) on their pipelines? I read that Github has some capability of enforcement, but that doesn't seem like a sound solution. I am curious to understand this and if there is any good litereature on how mature organizations handle this please point me in the right direction.
This is my own personal research and drive to understand being a cloud security engineer.
TIA
https://redd.it/1xcyyiv
@r_devops
Has anyone found a way to automate SOC2 audits using AI?
This shit is the bane of my existence. We use Secureframe so it does automate a lot, but I am still spending more time taking screenshots than any human should be.
I feel like with all the AI stuff these days there should be a way to automate this. I've had Claude generate a few scripts, which has been helpful in putting together network maps and firewall lists.
Is there a way to automate the "screenshot-taking" aspect?
https://redd.it/1xctzwl
@r_devops
how are you confirming the fix actually reached your running containers?
disclosure: i work on radar, an open source k8s ui. we just added a php version check for this to its mcp server and want to know what it's missing.
php2shell (CVE-2026-109618 / 109619) is all over the news: php core bugs that can allow unauthenticated rce, aws reports broad exploitation, fixed in 8.2.35 / 8.3.36 / 8.4.27 / 8.5.12.
with containers it's easy to think you're patched when you're not. images built FROM php keep the old php until someone rebuilds with --pull, and trivy doesnt flag it there because php is compiled from source. crane config <image> shows the PHP_VERSION the image declares without pulling layers, but images that don't set it are a blind spot.
how are you tracking which running workloads are still on an old php?
https://redd.it/1xcrscq
@r_devops
Do you review agent PRs differently?
I was looking into how to structure our review queues better and found some stats I wasn’t expecting. Apparently only about 32% of AI-generated PRs get merged within thirty days, compared to around 84% for manually written ones.
I read a breakdown on this, and the suggestion kinda makes sense, which is to treat an agent diff as a draft rather than finished work. The difference is that with a deliverable you’re mostly asking whether the code runs, while with a draft you’re asking whether the approach even fits the system in the first place. Going by that acceptance rate, teams are already treating them as proposals whether anyone decided to or not.
Question for those shipping a lot of code written by agents, does your team review it differently, or does it all go through the same process?
https://redd.it/1xcnnzl
@r_devops
What does everyone use for GRC and compliance in 2026?
We’re currently handling most of our compliance work through spreadsheets, shared docs and a few scripts, and it’s starting to show. SOC 2 is manageable, but ISO 27001 is coming next and I really don’t want to build the same controls twice.
Small team, mostly AWS, GitHub, Google Workspace and Jira. Looking for something that can pull evidence from the stack, keep controls monitored between audits and handle the auditor workflow without becoming another admin job.
Been comparing Vanta, Drata and Scytale, has anyone tried them for SOC 2 and ISO 27001 together? Mainly interested in how well the control/evidence reuse works once you’re running more than one framework, and how much of the audit prep it really takes off the team.
What’s everyone using now?
https://redd.it/1xchh4w
@r_devops
How do you catch third-party webhook payload drift before your ingestion pipeline starts dropping events?
Inbound webhooks from third-party partners and external clients are easily the messiest part of our API pipeline.
Whenever an external partner updates their system or sends deprecated payload structures, we immediately see a spike in \*400 Bad Request\* responses on our webhook ingestion endpoints.
Standard monitoring tools alert us *that* 400s are spiking, but attributing those failures back to a specific customer ID or payload shape usually means pulling raw logs and manually comparing JSON fields line-by-line. Because we don't control the external senders, pre-deploy schema checks don't stop them from firing stale payloads in production.
How are your teams handling this at scale?
* Are you running dead-letter queues with custom log parsers to map failures back to senders?
* Do you do passive schema validation at the gateway level?
* Or do you just treat external webhook payload drift as inevitable operational noise?
Curious to hear how other backend teams handle this workflow!
https://redd.it/1xcakcr
@r_devops
Mantle for API Maintenance
Just came across this software called Mantle on Product Hunt. Apparently, it detects breaking API changes in your GitHub repo and suggests code fixes before they break your integrations in production.
Seems interesting, especially for APIs where keeping up with changes is such a waste of time. Has anyone here actually tried it? Curious how well it works.
https://redd.it/1xc2vqi
@r_devops
Railway's security incident is more serious than a normal outage. Customer environment variables may have been exposed.
I've been following the Railway situation since the outage started, and things have gotten pretty concerning.
I'm currently using Railway to host the backend of a small gaming marketplace I've been building. Like many developers, I initially thought this was another infrastructure outage affecting builds and deployments.
But on October 21, Railway published a security disclosure confirming something much more serious.
According to their statement, they detected anomalous activity involving systems responsible for storing customer environment variables.
Their investigation suggests an unauthorized third party may have accessed encrypted environment variables and, in some cases, obtained their decrypted contents.
That's a serious concern for anyone hosting production applications on Railway.
**What Railway has disclosed so far**
* Suspicious activity was detected in systems storing customer environment variables.
* An unauthorized third party may have accessed encrypted secrets and obtained some decrypted values.
* The impact may be limited to a subset of customers, although the investigation is ongoing.
* Railway is recommending that all customers rotate their stored credentials, including API keys and database credentials.
* Deployments have been paused during the investigation.
* Railway believes the incident may be connected to a broader campaign targeting platform companies and AI inference providers.
* Railway has also started automatically rotating its bucket credentials and resetting certain integration credentials.
As of the latest update I've seen, deployments remain paused.
**My biggest concern is how developers are supposed to rotate production credentials during an outage like this.**
Think about a small startup running its entire backend on Railway.
You might have PostgreSQL credentials, JWT secrets, payment API keys, SMTP credentials, webhook secrets, and other sensitive information stored as environment variables.
Now imagine being told that some of those credentials may have been exposed and that you should rotate everything immediately.
That's obviously the correct security recommendation.
But there's a problem.
If you revoke your current database credentials or payment API keys before your running backend can use the replacements, you risk breaking critical functionality.
And when deployments are paused, rolling out those changes becomes much more complicated.
At the same time, leaving potentially compromised credentials active is also risky.
For solo developers and small teams without dedicated infrastructure engineers, this is not a simple situation.
**What I've personally done so far**
I've reviewed my Railway account sessions, signed out older sessions, and inspected my existing environment-variable configuration.
Everything appears normal from my side.
But I understand that simply checking whether the variables still exist doesn't prove they weren't accessed.
An attacker could potentially copy credentials without modifying anything visible in the dashboard.
I haven't received confirmation that my particular project was affected, and I'm preparing to rotate my sensitive credentials carefully.
**The bigger issue is trust in managed infrastructure.**
One of the reasons developers use services like Railway is that they don't want to manage every part of their infrastructure manually.
We trust these platforms to handle deployment infrastructure and provide secure storage for sensitive configuration.
Of course, no platform can guarantee perfect security.
But when the infrastructure responsible for storing customer secrets is potentially compromised, the impact can extend far beyond the hosting provider itself.
A stolen database password could potentially allow unauthorized data access, depending on network restrictions and database permissions.
A stolen API key could allow unauthorized operations or unexpected charges.
A compromised authentication secret
Would you use a Go-based Ansible alternative with HCL configs?
I'm considering building an open-source server configuration tool in Go,
in the space of Ansible. Nothing is built yet, I'm checking whether the
idea makes sense.
The idea:
\- one binary, no Python, connects to servers over SSH
\- configs in HCL (like Terraform) instead of YAML + Jinja
\- `plan` shows a diff of what will change, then `apply` does it
\- later maybe a small optional agent on the server for when SSH is blocked
Target: small fleets and homelabs, not replacing Ansible for everyone.
Questions:
1. What annoys you most about Ansible?
2. Would you try something like this, or is the module ecosystem a dealbreaker?
3. Is an optional agent useful, or would you stick to SSH only?
# Honest criticism welcome. Thanks!
https://redd.it/1xbx70p
@r_devops
Learning in the AI era...
Mid level engineer , who landed and worked my way up into the role. I wont lie and pretend I know everything, some topics I'm still working on, hell I haven't even touched Kubernetes yet, so you can imagine I feel way FREAKING behind.
It seems like everything is quickly adopting and using AI, from downloading models locally, to use and play with them to using them to learn , to using the next leading edge models. How are you all having time for this....
I want to start using AI more confidently but I am not sure where to even start. Local models vs generate chat models via gui in chatgpt or Claude...are there any trainings you guys recommend?
I'm also weary of using AI to learn cause I do know it hallucinates, but also I dont want to feed it and let it spit out answers for me whenever I am just learning how to implement something.
This is developing so fast, I feel like I need to start putting guardrails up for my own sake of learning and sanity, else I will lose knowledge and become a Senior Devops Prompt Engineer lol...jokes aside open to how you are all trying to navigate this, from learning with and without, what guardrails are you implementing , and any learning material would be appreciated.
https://redd.it/1xbrzi1
@r_devops
AI Identity Crisis
I'm a senior with loads of years, i've been doing devops/sre and systems engineering in the cloud for a long time. recently i volunteered for a project that i knew would lean heavily on claude, and with my manager's enthusiasm and support it looks like the project will be a success. aside from not truly owning that success, i'm also more than a little concerned that the 10K lines of code that i've slopped out, while mostly robust and functional, may not be personally defensible or supportable by me. this is a "career first feeling" for me, and i'm wondering if others have had similar experiences recently. is this just the way it is? maybe i should just treat it like "i started a new job and inherited a bunch of new libraries" when things come up? I'm not really sure how this is going to end.
https://redd.it/1xbpjnn
@r_devops
What’s the worst production failure your monitoring didn’t catch?
Curious how other developers handle cases where everything technically reports success — HTTP 200, Lambda completes, webhook delivered — but the intended operation never actually finishes.
For example, a payment succeeds but the customer’s subscription isn’t activated, or an API call succeeds but a downstream system never gets updated.
Have you encountered anything like this in production? How did you discover it, and what tools or processes do you use to catch these failures?
I’m especially interested in cases where traditional logs, metrics, or error alerts weren’t enough.
https://redd.it/1xbbpob
@r_devops
Too many internal tools are becoming hard to maintain
We've built tens of internal dashboards, tools, and admin utilities over time. Now, vibe-coded tools are becoming the norm for us, and it feels like a maintenance nightmare.
We have to constantly update and fix them and keep them working with our ever-changing data structure. Also, It's getting hard to track what tools exist, who uses them, and what’s active vs. abandoned.
How does your team handle this? Are there specific systems or governance models you use to maintain the internal tools?
https://redd.it/1xb0b5k
@r_devops
Are you patching faster now that AI can help turn old bugs into working exploits?
Reading the php2shell research, the bit that stuck with me was how AI helped turn longstanding bug into working exploit. The bugs was already there, getting them to work reliably was the difficult part.
That make “hard to exploits” feel like a shaky reason to leave something in the backlog. For people maintain production system, has this changed your patching schedule at all? We are still on php 7.4 and the last upgrade attempt broke half of our cron jobs, so i’m curious how other team handle an urgent patch when the upgrade itself need debugging
source: Wiz’s PHP2Shell research, published october 21
https://redd.it/1xd8wob
@r_devops
Handling state verification and drift when agentic workflows execute machine-speed remediation
Standard IaC (Terraform, CloudFormation) works cleanly for static configurations, but breaks down when non-deterministic agentic workflows run at machine speed. When autonomous agents start provisioning resources and executing remediation routines, un-governed execution triggers compliance drift and cost spikes.
To solve this across enterprise AWS environments, I engineered a deterministic data plane that handles runtime state verification using a 90/8/2 autonomic triage model backed by 19 native security primitives in a Bring-Your-Own-Account (BYOA) architecture.
I built an interactive command center simulator where you can execute an unauthorized role elevation vector, watch Bedrock AI isolate the node, and inspect the automated heal.py drift correction routines restoring nominal state.
I put together a full architectural breakdown and live command center sandbox. Happy to share the link or discuss how other SRE/DevOps teams enforce deterministic state boundaries when introducing autonomous AI agents to cloud runtime environments.
https://redd.it/1xd2yek
@r_devops
Assigning production firewall access to a one-month hire, and AI changes nobody reviews
Posting this as a discussion because I want to know if other people's orgs do the same thing.
Situation: five years at one company, infrastructure/sysadmin side. I am the default after-hours person. Alerts, escalations, weekend pages. When my own incidents happen, nobody answers.
Then my manager assigned our advanced firewall work to someone who joined a month ago. Production firewalls. She asked me last week what apt clean does. I didn't touch a firewall until four years into my career. His answer: she needs to warm up, and I'll be the one teaching her. So my workload now includes walking a beginner through every command on infrastructure that can take down the network.
The part I find interesting from a process standpoint is the second half. We have people who deploy AI-generated changes they clearly never read. When those break, the incident lands on me, not on whoever shipped it. Review keeps not happening at the point of change, it happens at 3am with me looking at it. Leadership's framing is that documenting all our work will "move AI from an individual habit to a team culture."
Two things I'm asking:
1. For those who've pushed back on assigning production change access to very junior staff, what argument actually landed with management? "It's risky" doesn't seem to move anyone.
2. Has anyone made pre-merge review stick for AI-assisted changes without becoming the person who reviews everything? Our last three incidents all traced back to that gap.
I'm not looking for "just leave", that isn't available to me right now.
---
Note: I wrote this in Turkish first and only used a machine translation to get it into English. The words and the experience are mine.
https://redd.it/1xcyhk4
@r_devops
I'm a one man DevOps team and my production runs for $0.64 a day. Here's the real story.
Most people writing cost posts here are spending their company's money. Mine comes out of my own pocket. I'm a solo DevOps engineer, I run the backend for a mobile app with real users, and the AWS bill is mine. That changes how you build things.
Real setup, no dressing up. Two K3s nodes on t4g.small, one in each AZ. Two seperate database nodes, also t4g.small, running Postgres and Valkey. A tiny fck-nat box so I don't pay for a managed NAT. A bastion to get in. Backups go to S3, encrypted. Six small machines in total. I checked the AWS dashboard while it ran, it came to $0.64 a day. About $19 a month, everything included.
Before building it I priced the managed way, the way the tutorials tell you. EKS is $73 a month just for the control plane. Small RDS is $13. Load balancer $16. Managed NAT $33, and that's before data. So around $135 a month in extras, more than my whole setup costs, times seven. I get why teams pay it, they are buying sleep. I wanted sleep too, I just couldnt pay that much for it.
What broke first was not the cheap machines. It was my own monitoring. I installed the full Prometheus stack like everyone says, and it ate my 2GB nodes. Memory got so low, like 513MB free, that etcd couldnt reach consensus. The cluster was dying because of the thing that watches the cluster. I still laugh about that one. I ripped Prometheus and Grafana out, put in Uptime Kuma for health checks, it's like 100MB, and metrics-server for the basics. Lesson is simple, your monitoring also has to fit the budget.
Deploys are boring, in a good way. ArgoCD, GitOps, two replicas of the API. Bad deploy, it rolls back, I dont get a panic call.
Now the honest part. The second database node never got its replication set up. Postgres runs on one node, the other one is just sitting there. If that node dies I am restoring from backups, there is no auto failover. That's the real price of my cheap setup, and I won't pretend otherwise. First thing I fix when the money is real.
I did load test it before trusting it though. 133,722 requests, 300 people at the same time, zero failures. Small app, but it held.
Other solo guys here, what does your full setup look like and what does it cost you? Curious where my number is actually dumb.
https://redd.it/1xcpuj4
@r_devops
Career change? Need help
Hey guys I am currently an Intelligence Analyst in HMRC but want to change career as I have hit a dead end and there isn't much room to grow unfortunately. What would you guys advise as a good career to transition into? I have been heavily reading up on DevOps but seem to be seeing a 50/50 fight going on lol.
https://redd.it/1xco4n8
@r_devops
I run production on K3s on a single EC2 box for $0.64/day. Here is what actually breaks.
I got tired of paying the managed premium for boring workloads, so I moved a real production backend onto K3s on a single EC2 box. Total cost: $0.64/day, all in.
What I expected to break: everything. What actually breaks is the disk first. One box means one disk, so I run automated snapshots and I test the restores, because a backup you never restored is a hope, not a backup.
Then deploys. There is no rolling magic across nodes, so deploys are GitOps with health checks and automatic rollback. Boring on purpose.
The sneakiest one is my own complacency. The bill is so low you stop watching it, so I still run a weekly cost check, because the whole point dies the day something silently scales.
What does not break: the app itself. K3s is boring in the best way once the manifests are right.
The honest catch: this is single-AZ by design. If you need multi-region failover, this is not your architecture. Most small workloads do not, and they pay the managed premium while running everything in one region anyway.
What is the cheapest production setup you have actually trusted with real traffic?
https://redd.it/1xcmiu5
@r_devops
Guys can you rate my portfolio website.
Guys can you rate my portfolio website
https://sumit.cloud
Thanks
https://redd.it/1xcczvl
@r_devops
How to learn devops quickly?
I've been teaching myself DevOps for the past few months and I've hit a wall where I'm not sure if I'm actually making progress or just hoarding tutorials. I'd really appreciate some honest input from people who do this for a living.
A bit about my background:
I come from a non-traditional background (not a CS degree) and currently work in a cloud-adjacent role.
I'm comfortable on the Linux command line — bash scripting, file permissions, processes, basic systemd.
I understand networking fundamentals: DNS, HTTP, TCP/IP, ports, SSH.
I know Git reasonably well (branching, merging, pull requests).
I can read and write basic Python for small automation scripts.
I've built a few Docker images and used docker-compose for local setups.
What I've been doing so far: I've been following a mix of YouTube courses, roadmap.sh, and random blog posts. The problem is everything feels scattered — one tutorial says "learn Kubernetes now," another says "master Terraform first," and I keep bouncing between tools without finishing anything end-to-end.
Where I'm stuck / my actual questions:
1. Order of learning — After Docker, what should come next? Kubernetes, CI/CD, or Terraform? I keep seeing conflicting advice and I don't want to waste months on the wrong thing.
2. Depth vs. breadth — Is it better to go deep on one cloud provider (I'm leaning toward AWS) or to understand a bit of everything first? I'm worried about becoming a "jack of all tools, master of none."
3. Kubernetes — Do I need it to get a first junior/entry-level role, or is it something I can learn on the job? It feels massive and intimidating.
4. Projects — What kind of portfolio project actually impresses hiring managers? I'd rather build one solid, realistic project than ten toy demos. Any concrete suggestions?
5. CI/CD — Is GitHub Actions a good enough place to start, or should I learn Jenkins/GitLab CI because that's what companies actually use?
6. The gap no one talks about — What skills do self-taught people like me usually miss that become obvious the moment you're on a real team? (On-call? Incident response? Reading someone else's messy infrastructure?)
What I'm hoping to get out of this post: A realistic, prioritized path for the next 3–6 months, and a gut-check on whether my foundation is solid enough to start applying for junior roles — or if I'm kidding myself.
I'm genuinely putting in the hours here, I just don't want to spend them in the wrong places. Any advice, resources, or "I wish someone had told me this earlier" stories would mean a lot. Thanks in advance 🙏
https://redd.it/1xc8nf3
@r_devops
could potentially affect user sessions, depending on how the application implements authentication.
None of this automatically means these things happened to Railway customers. These are potential consequences of exposed credentials.
And that's exactly why the incident deserves serious attention.
**There are still several questions I'd like Railway to answer.**
1. What was the actual entry point used by the attacker?
2. How could encrypted environment variables potentially be decrypted?
3. Was this caused by compromised internal credentials, an infrastructure vulnerability, or something else?
4. How many customers or projects were actually affected?
5. Was there any evidence of attackers accessing customer databases directly?
6. What time period should developers investigate when reviewing suspicious activity?
7. What additional security measures will Railway introduce to prevent this from happening again?
I appreciate that Railway disclosed the incident and started providing recovery guidance.
But until the full scope is known, it's difficult for developers to understand their actual exposure.
**I'm curious how other Railway users are handling this.**
Have you already rotated all your production credentials?
Have you noticed any unusual database activity, suspicious API calls, or unexpected service charges?
Has Railway contacted you directly about your project being affected?
Are you waiting for deployments to resume before rotating certain credentials, or have you temporarily disabled critical functionality?
And probably the biggest question: **Has this incident changed your trust in Railway, or do you still plan to keep your production applications there?**
I'm not trying to spread panic or suggest that every Railway project has been compromised.
I'm sharing this because I'm running a real application on the platform, and I know other independent developers and small teams are in a similar position.
I'd like to hear how others are approaching the security risks and what precautions they're taking.
Hopefully, Railway publishes a detailed technical postmortem once the investigation is complete.
https://redd.it/1xc6bbp
@r_devops
Ansible-like tool in Go with HCL configs and plan/apply: would you use it? (feedback wanted, no code yet)
Hi all,
I'm thinking about building an open-source tool for configuring servers
(installing packages, managing config files, restarting services),
something in the space of Ansible, but written in Go. Nothing is built
yet, I want to check whether the idea makes sense before I invest time.
What bugs me about Ansible:
\- YAML pretends to be a programming language: `when:`, `loop:`, Jinja
inside strings, and a long list of variable precedence rules
\- it needs Python on every machine and can be slow
\- you can't see clearly what will change before you run it
What I'd like to build:
\- one Go binary, no Python, easy to install
\- configs in HCL (the same language as Terraform) instead of YAML + Jinja
\- `plan` shows what would change (with a diff), then `apply` does it,
just like Terraform
\- resources reference each other and form a dependency graph, instead of
running strictly top to bottom
\- idempotent: running it twice changes nothing the second time
Two ways to reach your servers:
1. Over SSH. You run the tool on your laptop, it connects to the servers,
and nothing needs to be installed on them. This is the first thing
I'd build.
2. Optionally, a small agent: the same binary, running on the server as
a service and listening on a port. Useful when SSH is locked down,
blocked by a firewall, or you just don't want to expose it. The
resource code would be identical in both modes, only the way of
delivering commands differs. This would come later, and the
communication would be encrypted and authenticated (mutual TLS).
Here is how an nginx setup could look:
package "nginx" {
state = "present"
}
file "/etc/nginx/nginx.conf" {
source = "files/nginx.conf"
mode = "0644"
require = [package.nginx\] # install the package first
notify = [service.nginx\] # reload nginx if this file changes
}
service "nginx" {
enabled = true
state = "running"
require = [package.nginx\]
}
Run `plan` on a fresh server and it shows: package will be installed,
file will be created, service will be enabled and started. Run `apply`
and it happens. Run `apply` again and it says "0 changes". Edit the
config file and `plan` shows the diff and a pending service reload.
What I'm NOT trying to do: replace Ansible for everyone. Ansible's huge
module ecosystem is its real strength, and I can't compete with that. The
target is small fleets, homelabs and people who like simple tools.
I'd love honest feedback, including "I wouldn't use this":
1. What annoys you most about Ansible (or whatever you use today)?
2. Would you try something like this, or is the lack of modules a
dealbreaker?
3. Is the optional agent (a small service on the server) useful to you,
or would you rather stick to SSH only?
4. Which resources would you need first besides file, package, service
and command?
5. HCL or something else? Would you prefer another format?
Not selling anything: it's open source and there's nothing to download
yet. Thanks for reading!
https://redd.it/1xbvkcu
@r_devops
ADO as documentation?
Is ADO bullshit? Because I feel like the way our it department is using it is bullshit. I work on a data analytics team who has to use data created by IT.
Where's the documentation for this script? "It's in ADO". Who's the customer? Why did they request this change? "It's in the ADO".(No it isn't, it's in someone's email inbox somewhere).
What's the overarching structure/purpose for this screen? "We don't document that anymore, it's in [I shit you not\] the individual ADO tickets that built the screen". You can NOT be serious. No one wants to read every individual ticket turning features off and back on to figure out what the current state is. There's no universe in which this is remotely coherent.
Also, the vast majority of the screens were built before the ADO went live, so we have old school documentation that was last updated in 2019 establishing the purpose and basic design of 80% of screens but then the current state is old documentation plus any ticket affecting that screen layered on top that old documentation?
I feel like this is a scam and totally incoherent. They program in Java, we work exclusively in SQL/PowerBI (so we can't use or read any of their code anyway, but they've refused to send it to us even if we could). We had an item where my staff tried to recreate a screen-calculated variable and it was a nightmare trying to combine the pre-2019 documentation with new tickets because the calculation could be affected by a ticket that appeared to have nothing to do with that particular calculation at all. And again, we can see absolutely zero code, just user stories. But I don't want to see their code, my staff can't read or execute it. We aren't devs. We just need to know what the system is doing so that we can do accurate analysis of the resulting.
But it's "all in the ADO".
https://redd.it/1xc0ouu
@r_devops
LiteLLM Compromised again - 1.107.0. PyPI Installs a Credential Stealer and Backdoor
The full article: https://safedep.io/litellm-pypi-compromise-credential-stealer-backdoor/
https://redd.it/1xbq9of
@r_devops
How do you simulate realistic user traffic in load testing?
We are looking to overhaul our load testing, since our current script only slams 10k users in 60 seconds and gives us garbage data. It exhausts connection pools and creates artificial bottlenecks that would not show under any kind of normal use case.
Real users ramp up slowly, and idle on pages, and don't jump from one to another without a second thought, they do not fire off database transactions with zero pause.
We should be reflecting real behavior, not inducing cold start failures.
Has anyone modeled this at scale? Should we make the step curves with warm up times to let caches and other systems warm up before hitting the expected load? We want to reflect real user behaviour, not just trigger cold start failures.
https://redd.it/1xblst1
@r_devops
I was thinking on competition about chess AI
What if we hosted a public tournament for chess bots to compete against each other and see which one comes out on top?
https://redd.it/1xb068c
@r_devops
Interviewing for a Software Support Engineer role (Linux, Python, Bash, CI/CD) — what technical questions should I prepare for?
Hey everyone,
I have a 30-minute interview coming up for a Software Support Engineer position at a defense contractor, and I'm trying to prepare as thoroughly as possible.
The recruiter mentioned there won't be a formal technical assessment, but I still want to be ready for any technical questions or troubleshooting scenarios.
For background, I currently work in cloud/data center infrastructure operations, where I use Linux, Python, and Bash for diagnostics, troubleshooting, and automation. I also have experience developing software applications, working with REST APIs, automated testing, and CI/CD.
Here's the relevant job description:
Responsibilities:
Support lab tooling and version control through CI/CD pipelines.
Create and deploy test scripts.
Support and troubleshoot software deployments on custom servers.
Support configuration management processes.
Develop unit tests and tailor rule sets.
Required qualifications:
Bachelor's degree in CS, Computer Engineering, or a related technical field.
2+ years of relevant experience.
Experience with Python and Bash.
Ability to independently and collaboratively solve complex technical problems.
Preferred qualifications:
Knowledge of CI/CD pipelines.
Experience with GitLab.
Experience with Linux and NFTables.
I'm particularly interested in moving toward DevOps/platform engineering, so this seems like a great opportunity.
For anyone working in DevOps, Linux administration, software integration, or similar roles:
1. What technical questions would you expect in an interview for this position?
2. What Linux troubleshooting scenarios should I be prepared to walk through?
3. How deeply should I study GitLab CI/CD, unit testing, and NFTables?
4. Does this sound more like DevOps, software integration, or traditional software support?
5. If you were interviewing a candidate for this role, what would you want them to demonstrate?
I'm not looking for shortcuts or leaked interview questions. I genuinely want to understand the technical expectations and make sure I can speak confidently about the work.
Appreciate any advice!
https://redd.it/1xaxlif
@r_devops