Sumit ⚙️ DevOps @devopswithsumit

How are you managing secrets in production RIGHT NOW? → Kubernetes Secrets (it's base64 😬) → AWS / GCP Secrets Manager → HashiCorp Vault → .env files we don't talk about 😅 Drop your setup + what you wish you'd done differently 👇 #DevOps #Security #Kubernetes #India

Friday 5pm. "It's just a config change." 90 minutes of downtime later, I learned that "just a config change" is the most dangerous phrase in production operations. Full story, the mistake I almost missed, and the 4 rules we now enforce. ━━━━━━━━━━━━━━━━━━━━━━ The Change One nginx timeout line. timeout 30s → timeout 60s. Been sitting in the ticket queue for 2 weeks. Everyone wanted it done before the weekend. Deploy at 5:12pm Friday. ━━━━━━━━━━━━━━━━━━━━━━ 5:17pm — Error rate climbing. 2%, now 8%. We waited, assuming post-deploy noise. 5:23pm — Error rate 34%. Not noise. ━━━━━━━━━━━━━━━━━━━━━━ The Root Cause We Almost Missed The timeout line was correct. But in the same nginx config template — 3 lines below — was a variable substitution referencing $UPSTREAM_KEEPALIVE_TIMEOUT. This variable existed on staging nginx. It had never been set on production nginx. When nginx reloaded with the new config, it couldn't resolve the variable. It set keepalive timeout to 0. Every connection to upstream APIs closed immediately after one request. Under Friday evening traffic load, the connection pool exhausted in 4 minutes. ━━━━━━━━━━━━━━━━━━━━━━ 5:31pm — Rollback initiated. 5:38pm — nginx reload complete. 5:41pm — Error rate back to baseline. 29 minutes of 34% errors. On Friday evening. During peak traffic. ━━━━━━━━━━━━━━━━━━━━━━ The 4 Rules We Now Enforce 1. No production deploys after 3pm on Fridays. Hard stop. The pipeline refuses without an emergency override and manager approval. 2. nginx config changes go through nginx -t AND a variable substitution validation script — checks every $VARIABLE in the config exists in the target environment's env vars. 3. "Config-only changes" go through the same canary process as code changes. "It's just config" is a lie that takes down prod. 4. A config diff step in the pipeline: before deploying, show the exact diff between current prod config and incoming. Human eyes on every change. ━━━━━━━━━━━━━━━━━━━━━━ The rule we live by now: if you wouldn't do it at 2am, don't do it at 5pm on a Friday. The only difference is your availability to fix it. What's your worst "just a small change" deployment? 👇 #DevOps #WarStory #SRE #CICD #nginx

I've reviewed 200+ DevOps portfolios and resumes in India over 3 years. Here's the honest truth about what gets people hired — and why most engineers are optimising for the wrong things. ━━━━━━━━━━━━━━━━━━━━━━ The myth: certifications get you hired. The reality: certifications get you pass the ATS filter. Humans don't hire certifications. They hire evidence. I've seen engineers with CKA + CKS + AWS SAA fail round one because they couldn't explain what happens when a readiness probe fails. I've seen engineers with zero certifications get hired at ₹35L because they had a GitHub repo with real infra code and a postmortem from a real incident. ━━━━━━━━━━━━━━━━━━━━━━ What a strong DevOps portfolio actually looks like: 1. A GitHub repo with real infrastructure code — not tutorial code, not "followed along with YouTube" code. Real code that solves a real problem. Even a small one. Example: "I set up a self-hosted Plausible analytics instance on a ₹800/month Hetzner VPS, managed all infra with Terraform, deployed via ArgoCD, and set up Prometheus uptime alerts." That one project covers: Terraform, GitOps, Kubernetes, Observability. All provably real. 2. At least one public postmortem — "The disk filled up because I forgot log rotation" is a perfectly valid postmortem if written properly: what happened, how detected, root cause, what changed. 3. One thread or post explaining something you learned publicly — writing forces clarity. Hiring managers at product companies read threads. They're looking for engineers who can communicate. ━━━━━━━━━━━━━━━━━━━━━━ 3 things to remove from your resume today: ❌ "Proficient in DevOps methodologies" ❌ "Experience with cloud platforms" ❌ "Knowledge of containerization technologies" Replace with: ✅ "Cut deploy time from 45 min to 8 min by migrating 28 services to EKS" ✅ "Reduced infra cost by 32% using Compute Optimizer + scheduled scaling" ✅ "Set up Prometheus + Grafana + Loki — wrote 23 alert rules, MTTR dropped from 47 to 11 min" Numbers. Tools. Outcomes. That's it. ━━━━━━━━━━━━━━━━━━━━━━ The uncomfortable truth for Tier-3 college engineers: your college matters less than your GitHub after your first job. What matters is what you've shipped and what you've broken. Drop one project you're proud of but haven't published on GitHub yet. Start there today. DM me your GitHub — I'll give you honest feedback. 🙏 #DevOps #Career #India #Resume #PersonalBranding #DevOps #Career #India #Resume

@IndianTechGuide UPI has genuinely become one of India’s biggest tech success stories 🇮🇳 The convenience of scanning and paying instantly - even while traveling abroad - still feels amazing.

Honestly feels great seeing UPI become usable across so many countries 🇮🇳 A few years ago, nobody would’ve imagined India’s payment infrastructure becoming this globally recognized. Super convenient for travelers too 😄

Indian Tech & Infra @IndianTechGuide

4 days ago

🚨 List of countries where Indians can use UPI to make payments. France UAE Singapore Qatar Nepal Sri Lanka Mauritius Bhutan Cambodia

77 227 3K 77K 172

One PostgreSQL query that shows exactly what's slow: SELECT query, calls, mean_exec_time, total_exec_time FROM pg_stat_statements ORDER BY mean_exec_time DESC LIMIT 10; Slow queries hiding in plain sight — revealed in 10 seconds. Run this now. Drop what you find 👇 #PostgreSQL #DBA #DevOps #SQL

Tools I use vs my resume 😭 ACTUALLY USE: k9s, lazydocker, jq, tmux, Lens, httpie MY RESUME SAYS: "enterprise-grade container orchestration solutions" 💀💀💀 What's YOUR version of this? 👇 #DevOps #Tools #Kubernetes #Relatable

Made a GCS bucket public in prod 😱 Didn't realise for 6 hours. Customer data? No. Build logs? Yes. Still the worst 6 hours of my career. Now: Cloud Asset Inventory alerts fire in seconds for any public bucket. Your worst cloud security mistake? 👇 #GCP #Security #DevOps #CloudSecurity

A guy forgot his Bitcoin wallet password for 11 years. Not a random wallet either… It had 5 BTC inside — worth nearly $400,000 today. The crazy part? He originally changed the password in college and never remembered it again. 😭 Over the years he tried everything: • password recovery tools • old notebooks • brute-force attempts • trillions of combinations Nothing worked. Recently, as a last attempt, he uploaded old files from his college laptop into Claude AI. Claude started connecting the dots: – found an old wallet backup – identified a hidden file – spotted an issue in the recovery process – helped piece together the missing info And somehow… the wallet opened. 🔓 After recovering the Bitcoin, he said he’s going to name his child "@DarioAmodei" Yes, after — the CEO behind @claudeai This story is wild because Claude didn’t “hack” Bitcoin. It just did something humans struggle with: finding patterns in years of forgotten digital chaos. We’re entering a phase where AI won’t just answer questions. It’ll recover lost data, debug systems, investigate incidents, and probably save people millions. That’s both exciting and slightly terrifying.

🍜 @cprkrn

4 weeks ago

HOLY FUCKING SHIT OMG CLAUDE JUST CRACKED THIS SHIT, THANK YOU @AnthropicAI THANK YOU @DarioAmodei NAMING MY KID AFTER YOU 😍 blockchain.com/explorer/addre…

3K 3K 39K 16.9M 9K

@NaveenS16 Kubernetes operators feel simple… until you actually build one. The reconcile loop gives freedom, but also enough rope to create some truly terrifying state-management bugs. A lot of operator engineering is basically controlled chaos.

We're building the reliability layer for AI agents. Open source. Apache 2.0. Serious contributors welcome.

@freeCodeCamp @irvingpictures So many production issues start quietly. No major outage. No dramatic alert. Just small signals everyone ignores until things suddenly get painful 😅 This is why Bash and Python automation skills are still so valuable in DevOps.

@awscloud The “scale to zero” + faster autoscaling combo is huge for AI workloads 👀 Agentic systems can be extremely bursty, so paying for idle capacity never made much sense. Interesting direction for teams building retrieval + search-heavy AI apps on AWS.

Cut our AWS bill by 40% in ONE week 💸 Dev/test instances running 24/7 on weekends. Nobody using them. Lambda + EventBridge: stop at 8pm IST, start at 9am. ₹2L/month saved. Zero impact on dev work. Your company is doing this right now. Go check 👇 #AWS #DevOps #CloudCost #FinOps

@awscloud @RDarrylR @ricardoceci Love seeing community builders get recognized like this 👏 The impact of sharing knowledge, mentoring others, and consistently contributing often reaches far beyond code itself. Big congratulations to all the new AWS Heroes 🚀

@Njuchi_ This is such an important point. A lot of engineers jump straight to Terraform, GitOps, or AI tools without ever experiencing the pain those tools were designed to solve. That context changes how you design infrastructure completely 👏

@TheTechMaharaj KubeCon Mumbai is going to be massive for the cloud-native community 🔥 The networking, real-world scaling stories, and hallway conversations alone are worth attending. Excited to see builders, SREs, platform engineers, and DevOps folks all in one place 🙌

My suggestion: before buying anything on @letsblinkit & @blinkitcares , always check whether the item is returnable or not. If it isn’t, better avoid it and purchase from other platforms. Otherwise, you may regret it later and realize you could’ve gotten better quality at a reasonable price elsewhere (@Meesho_Official)with a return policy.

DevOps salaries in India 2025 🇮🇳 (no BS): 0–2 yrs: ₹6–12L 2–5 yrs: ₹15–35L 5+ yrs: ₹40L+ Remote/product: 2–3x Are you being underpaid? DM me your stack + experience. I'll tell you honestly 💪 #DevOps #Career #India #Salary

@autobase_tech Not nearly enough. A lot of teams “have HA” on paper, but the first real failover test happens during an actual outage. That’s when things get very educational.

VACUUM ANALYZE your PostgreSQL. Right now. Dead tuples pile up on write-heavy tables. Slows EVERYTHING. Auto-vacuum isn't enough. Manual VACUUM weekly = night and day query speed difference 🔥 RT if this fixed something for you 👇 #PostgreSQL #DBA #DevOps #Linux