标签：#DevOps

Nginx Load Balancing with DNS-Based Service Discovery on Incus Hari ini saya buat satu practical lab untuk memahami Nginx Load Balancing , DNS-based Service Discovery , dan operational logging dalam persekitaran self-hosted menggunakan Incus. Lab ini bermula dengan architecture yang simple: Client │ ▼ Nginx LB │ ├──► web01 └──► web02 Kemudian saya tambah satu DNS server supaya backend tidak perlu bergantung sepenuhnya kepada hard-coded IP address. 1. Architecture Final architecture: DNS dns / dnsmasq 10.107.109.18 ▲ │ DNS lookup: web.incus │ │ Nginx LB 10.107.109.69 │ Load Balancing ┌──────────┼──────────┐ ▼ ▼ ▼ web01 web02 web03 .100 .253 .xxx Ada dua jenis communication flow dalam architecture ini. DNS resolution Nginx LB ──────► DNS │ └── web.incus ↓ .100, .253, .xxx DNS hanya digunakan untuk mengetahui IP address backend. HTTP traffic Client │ ▼ Nginx LB │ ├────► web01 ├────► web02 └────► web03 DNS tidak membawa HTTP traffic . DNS hanya menjawab: Where is web.incus ? Nginx kemudian menggunakan IP yang diperoleh daripada DNS untuk melakukan load balancing. 2. Static / Hard-Coded Upstream Cara paling mudah untuk configure Nginx Load Balancer ialah dengan meletakkan IP backend secara terus. Contoh: upstream backend { server 10.107 .109.100 ; server 10.107 .109.253 ; } Architecture: Nginx LB │ ├──► 10.107.109.100 │ └──► 10.107.109.253 Kelebihan Simple Mudah difahami Predictable Sesuai untuk environment kecil Tidak memerlukan DNS service discovery Kekurangan Kalau tambah web03 : web01 web02 web03 Nginx configuration perlu diubah: upstream backend { server 10.107 .109.100 ; server 10.107 .109.253 ; server 10.107 .109.xxx ; } Kemudian configuration perlu divalidasi dan biasanya Nginx perlu di-reload. 3. DNS-Based Service Discovery Pendekatan kedua ialah menggunakan hostname sebagai service identity. Contohnya: web.incus DNS: web.incus ├── 10.107.109.100 ├── 10.107.109.253 └── 10.107.109.xxx Nginx tidak perlu mengetahui backend IP secara hard-coded. Contoh: resolver 10.

After development comes deployment - whether on-premise or on a cloud based environment. And then we face a simple question: how much resource should I assign to this system? What is the ideal numbers? If we get this wrong, we often need to go back time and again to fine tune - either to ensure our application is capable of handling the targeted load, or to avoid paying for resources we are not using. This article explains the approach step by step. So that we spend just enough time upfront to avoid spending exponentially more time and money at later stages. Who Is This For? This article is written primarily for developers. But if you are a manager or a CTO, there are sections written specifically for you. Feel free to jump straight there. 👉 If you are a Manager or Project Manager 👉 If you are a CTO or Architect For everyone else - the full article is worth reading top to bottom at least once. But if you are revisiting a specific topic, jump to whatever is relevant. Table of Contents Local is the Starting Point When Should You Start Thinking About Right Sizing? How Long Will This Actually Take? Start With What You Have - Your Local Setup Setting Up Your Load Generation - The Hammer The Cost of Testing - This Is Not Free Sizing Your Pod More Resources Per Pod or More Pods? Scaling - Easy to Set Up, Hard to Get Right Periodic Right-Sizing - You Are Not Done Yet Local is the Starting Point Local system is always where we start. To try things out, to check if things work. But 99% of what we test locally is the sunny day scenario. Does the MVP work? Does the happy path hold? Even if you're diligent enough to test negative scenarios, you're almost certainly not testing production-level load on your laptop. Which means you have no idea what resources your app actually needs when it matters . This is where the problem starts. On local, we routinely kill the heavy IDE, close browser tabs, shut down background processes -without ever stopping to ask: how much memory and CPU d

I wrote this for the All Things Agentic Hackathon. Every incident-response demo you have seen ends the same way: something breaks, the agent fixes it, everyone applauds. I want to show you the opposite. Here is my agent, at 95% confidence, having correctly diagnosed a bad deployment, deciding not to roll it back. That refusal is the whole project. The question underneath At 3am an alert fires. An engineer wakes up, reads several hundred log lines, correlates them against recent deploys, and rolls something back. Most of it is mechanical. It is an obvious target for automation. But "automate it with an LLM" does not dissolve the problem, it relocates it. The new question is: how much would you let an agent change in production without asking you first? Give it too little and it is a chatbot that writes summaries. Give it too much and one confidently wrong diagnosis takes down your service at 3am with nobody watching. I named the project Sonjomon — Bengali for restraint. The autonomy ladder An agent should not have one blanket permission level. How far it may act alone is a function of two things: how confident it is, and how much damage the proposed action does if that confidence turns out to be wrong. tier = f(confidence, blast_radius) OBSERVE record findings, take no action SUGGEST recommend to a human, do not execute APPROVE stage the action, execute on explicit approval ACT execute now, then verify independently A restart is medium risk — reversible in seconds. A rollback is high risk — it shifts production traffic, and a needless rollback during a real outage extends it. Deleting data is critical, and no confidence level unlocks it. Six conditions can only ever push the tier down, never up: the blast-radius ceiling, thin evidence, a similar action that just failed, a third attempt at the same fix, a stale incident, and a global dry-run switch. Nothing pushes it up. A wrong action is far more expensive than a missed one. Three things the model does not control It

Most nginx misconfigurations announce themselves. You typo a directive, nginx -t fails, you fix it. That feedback loop is fast and it works. The dangerous ones are different. The config is valid. nginx -t passes. The server starts, serves traffic, logs nothing unusual. And the thing you configured is quietly not happening. I maintain gixy-ng , a static analyzer for nginx configs. A growing share of its checks exist for exactly this category, because it turns out static analysis is the only practical way to catch a failure that produces no signal at runtime. Here are four worth knowing about. 1. OCSP stapling that staples nothing server { listen 443 ssl ; server_name example.com ; ssl_certificate /etc/ssl/example.com.pem ; ssl_certificate_key /etc/ssl/example.com.key ; ssl_stapling on ; ssl_stapling_verify on ; } Looks right. It does nothing. OCSP stapling means nginx fetches the certificate's revocation status from the CA itself and attaches it to the handshake, so the client does not have to. To do that, nginx has to make an outbound request to a hostname. nginx does not use the system resolver for runtime lookups. It has its own, and it only exists if you configure it. No resolver in scope means the hostname never resolves, the fetch never happens, and stapling is silently skipped. Your config test passes. Your clients go do their own OCSP lookups, which is the exact thing you turned stapling on to avoid. resolver 127.0 .0.1 valid=300s ipv6=off ; resolver_timeout 5s ; Use a local resolver or your cloud provider's internal DNS. Pointing this at 8.8.8.8 sends every internal lookup off your network in cleartext, which is its own problem. Check it with: echo | openssl s_client -connect example.com:443 \ -servername example.com -status 2>/dev/null \ | grep -A 17 'OCSP response' Working stapling prints OCSP Response Status: successful . Broken stapling prints no response sent . Run it twice, since the first handshake after a reload usually goes out unstapled while the f

This article is Part of a series on running vLLM on AKS and walks through creating an AKS cluster with a GPU node pool, deploying vLLM onto it, and wiring up Prometheus and Grafana for visibility. Companion pieces: Choosing the right GPU | Why your autoscaler flaps | Source Setup Summary Cloud: Azure GPU node: Standard_NV36ads_A10_v5 (1× A10, 24 GB) Image / model: vllm/vllm-openai:latest serving Qwen/Qwen2.5-7B-Instruct-AWQ Observability: kube-prometheus-stack (Prometheus + Grafana), KEDA, NVIDIA DCGM exporter All commands below are bash. The steps are ordered and each one depends on the previous. Dependency chain The build order follows one chain: model → VRAM requirement → GPU SKU → region availability → quota. Step 0 — Prerequisites (one-time, survives resource group deletion) GPU quota. Request through Portal → Quotas → Compute → This article: Requested Standard NVADSA10v5 Family vCPUs = 108 in westus (108 = 3 nodes × 36 vCPUs, matching the autoscaler's max-count 3 set in step 3). Quota is granted per-subscription and survives resource group deletion, so this step happens once, not on every rebuild. A quota is Azure's per-subscription limit on how much of a resource (here, GPU vCPUs in a specific VM family) you're allowed to provision at once. New subscriptions start at 0 for GPU families since it's expensive and can be abused. You need it because without an approval, az aks nodepool add for a GPU will fail outright. The request goes through manual Azure approval, so it has to happen before you plan to build. * Prerequisites * Existing Azure Subscription: Local tooling: Helm 3+ kubectl Bash Bash Variables to set for use through the setup RG = <resource-group-name> CLUSTER = <cluster-name> LOCATION = <preferred-location> Step 1 — Create Resource group az group create -n $RG -l $LOCATION Step 2 — Create AKS cluster, on a CPU system pool az aks create -g $RG -n $CLUSTER \ --node-count 1 --node-vm-size Standard_D2s_v5 \ --generate-ssh-keys The GPU does not go on thi

An AI agent telling you "done" is not evidence. When I started delegating work to Codex, I took those reports at face value — until I checked the code and found the change missing, the wrong file edited, or no commit at all. So I stopped trusting language and started making the shell write the facts to disk. Why this design works When you hand a task to Codex, it comes back with "Completed." At first that satisfied me. But when I actually checked the code, the critical change wasn't there, or a different file had been touched, or git commit had never run. The output "I did it" and the fact "it was actually done" are two different things. This is true of Claude Code too. Whether tool results were read correctly, whether errors were swallowed — even with code I wrote myself, running a self-audit right after declaring completion turns up something every single time. Delegating implementation to an AI amplifies that problem by one more notch. The fix is simple: make it write state to a file, not to language. Even if the AI says "completed," it isn't complete unless State: completed exists in the status file. If the handoff file doesn't contain the real output of git status --short , you don't know what changed. If the four sections you specified in the task file (Summary, Files Changed, Validation, Remaining Risks) aren't there, you can't verify it. Files don't lie. An AI under pressure will insist "I did it," but the output of cat status-file can't be forged. Pushing state management down into the filesystem is what makes it possible for a human to cross-check it in a shell . That's the essence of this design. The other important piece is separation of concerns . orchestrate-codex-worker.sh takes three arguments up front. bash scripts/orchestrate-codex-worker.sh <task-file> <handoff-file> <status-file> Each of these three files has a clear role. task-file : The work order for Codex. It contains only "what to do." handoff-file : The handoff note after Codex finishes. Wr

If you have ever built a health check, you have probably written something close to this: const res = await fetch ( url , { method : ' GET ' , signal : AbortSignal . timeout ( 10000 ) }); const isUp = res . status === 200 ; I ran a version of that for a while. It is wrong in at least five ways, and every one of them bit me while building an outage tracker for Indian services. This is a write-up of what actually breaks, because most monitoring tutorials stop at the snippet above. 1. The server answers, the service is dead The single biggest gap. 200 OK tells you a server returned a response. It tells you nothing about whether the thing a user came to do still works. A bank homepage can render in 400ms while UPI payments from that same bank are failing at the switch. Different systems, different teams, different failure modes. Your check is green and the feature is on fire. You cannot fully solve this from outside. What you can do is stop treating a 200 as proof of health, and stop displaying it as one. 2. 403 is not down Plenty of sites block automated requests deliberately. Bot protection, WAF rules, rate limits, geo rules. In India this is common on high-value government and travel portals. IRCTC is the obvious example. A naive checker marks these down permanently. Users learn to ignore your tool inside a week. 403 means the server is alive and refusing your specific request. That is different information from 500 , and treating them the same throws away the distinction that matters most: Code Server state What it tells a user 200 Alive, responded Little. The feature may still be broken. 401 / 403 Alive, refusing this request Usually nothing about the outage. Often your check being blocked. 404 Alive The path is wrong, not the service 429 Alive, rate limiting you You are the problem, back off 500 / 502 / 503 Broken, overloaded, or in maintenance Genuine signal 504 Something upstream did not answer Genuine signal, usually a dependency Timeout / DNS failure Unknown A

Sometimes, a change that looks simple on the surface is not actually that simple. Imagine that you need to replace the source table feeding a refined or trusted table in a data pipeline. At first, it might look like a one-line change: update the table name, deploy the code, and move on. But in a real data platform, there is usually much more behind that change. There are dependencies, scheduling rules, upstream and downstream processes, resource consumption, concurrency, data lineage, and, sometimes, assumptions that were not immediately obvious when the pipeline was first created. I recently had to look into exactly this kind of situation in an Apache Airflow project, and one of the questions that came up was: Should this DAG be scheduled using a cron expression, or should it be triggered based on an Asset? The answer, as usual in software engineering, is: it depends. And understanding why it depends is much more important than simply knowing how to configure either option. Cron: the familiar way of scheduling a DAG Let's start with the simplest and most familiar option: a time-based schedule. With Airflow, we can define a DAG to run according to a cron expression: with DAG ( dag_id = " my_pipeline " , schedule = " 0 13 * * 0 " , catchup = False , ): ... In this example, the DAG is scheduled to run every Sunday at 1 PM. In a real environment, we might have different schedules for different environments. For example: Environment Schedule Development Saturday at 1 PM Homologation Sunday at 1 PM Production Monday–Friday at 1 PM The important characteristic here is that the schedule is based on time . If the DAG is configured to run at 1 PM every Sunday, Airflow will try to run it at that time, regardless of whether the data it depends on has actually changed. This is not necessarily a bad thing. In fact, sometimes this is exactly what we want. But there is another approach. When data becomes part of the schedule Modern data pipelines often have dependencies that are b

A little over two months ago, I registered a new domain for personal use. The idea was simple. I wanted a permanent, professional email address based on my last name, something like first@lastname.me . I registered the domain for ten years because I wasn’t building a disposable project, launching a marketing funnel, or testing some short-lived startup idea. I wanted an email identity I could keep for the long haul. I configured the domain properly. It has valid DNS. SPF is enabled. DMARC is enabled. It isn’t parked for sale. It isn’t sending spam. It isn’t distributing malware. It isn’t impersonating a bank, crypto exchange, social network, government agency, or anyone else. Then I checked it with IPQualityScore, also known as IPQS. The result was absurd: Phishing: true Suspicious: true Risk score: 95 Spamming: false Malware: false SPF enabled: true DMARC enabled: true DNS valid: true Parked domain: false Hosted content: false Category: N/A Domain rank: 0 Risky TLD: true In other words, IPQS acknowledged that the domain had valid DNS and email authentication, found no spam, found no malware, found no hosted content, assigned it no content category, and still labeled it as phishing with a risk score of 95 out of 100. I submitted a correction request about a month ago. I received no explanation. No evidence. No request for verification. No ticket update. No human response. As of August 29, 2026, the status is still unchanged. That isn’t a harmless technical oddity. IPQualityScore sells reputation and fraud-risk data that businesses can use to block users, reject signups, review transactions, investigate security alerts, and decide whether a domain, email address, IP address, phone number, or device should be trusted. If you’re going to sell suspicion as a service, you need to be accountable when your suspicion is wrong. IPQS, in my case, has been neither accurate nor accountable. A score of 95 is not a gentle warning IPQualityScore’s documentation describes its URL ri

Line 267 of a 427-line Terraform plan: # aws_rds_cluster.reporting must be replaced - /+ resource "aws_rds_cluster" "reporting" { ~ arn = "arn:aws:rds:us-east-1:842910557412:cluster:reporting" - > ( known after apply ) ~ cluster_resource_id = "cluster-D85642F9611A" - > ( known after apply ) ~ engine_version = "14.9" - > "15.4" ~ id = "reporting" - > ( known after apply ) ~ storage_encrypted = false - > true # forces replacement # (29 unchanged attributes hidden) } The merge request says "bump reporting Postgres to 15.4." The plan does exactly that. It also destroys the reporting database and creates an empty one in its place. Underneath the known-after-apply churn, two attributes are changing. One is the version bump, the thing your MR is about. The other is storage_encrypted flipping from false to true , and it isn't yours. Someone on another team that shares this repo merged it earlier in the week. You're just the one deploying. You review other people's Terraform MRs and have a feel for what each stack normally does; most weeks someone else shepherds the deploy. Today it's you. Your change goes out next, so you're carrying everything merged since the last deploy, including work you never reviewed and had no reason to know about. Nobody was negligent. The queue simply had someone else's change in it. It's a good change, by the way. You want encrypted storage. But there's no in-place path from unencrypted to encrypted on an RDS cluster. Terraform's only move is destroy and create. That's what -/+ means, and the comment at the end of the line says it in plain English: forces replacement . And the version bump alone would have failed. Going from 14 to 15 is a major version upgrade, and Aurora refuses those unless the config sets allow_major_version_upgrade = true . This one doesn't. That MR by itself would have died at apply, loudly, with an error naming the exact problem. A replacement doesn't upgrade anything. It creates a new cluster at 15.4 from scratch, so the f

The problem I run Hermes, an open-source agent with tools, memory, and cron built in. Before running my own managed service SaaS, I used various competitors to deploy on VPS. This method has a lot of downsides because you are often SSH'ing in and managing secrets directly in a .env file, which can leave you exposed if your box is compromised. It is also very cumbersome, especially for agencies to manage these "alway-on agents" for clients on VPS. Luckily, these are just hosting problems, so I built SEAOTTER to fix it for myself, and then realized other people probably have the same problem. * What it does * SEAOTTER is a managed control plane for Hermes Agent: Per-agent isolation - each agent runs in its own namespace with a gVisor sandbox, so one client's agent can't see or touch another's. Operate without SSH — pause, restart, restore, and read logs through an API instead of a terminal. MCP-native — talk to a hosted agent from Claude, Cursor, or Codex. Secrets handled for you — backed by Google Secret Manager instead of a .env file you have to remember exists. The rough idea POST /api/v1/agents or on "Create Agent", and it provisions a namespace, installs Hermes via Helm, brings up the sandbox, wires DNS/TLS, and gives you a reachable dashboard, typically in under five minutes. Who it's actually for Agencies running one isolated agent per client without spinning up a VPS per client Hermes power users who want lifecycle control (pause/restart/restore) without maintaining SSH access Hobbyists who want a standing assistant without becoming an ops person Try it There's a 14-day free trial on the Hobby plan. Worth being upfront: it currently asks for a card at checkout, which I know is friction — I'm working on a no-card way to try it. In the meantime, the docs walk through the API and dashboard in detail if you want to look before you sign up. What I'd love feedback on If you're currently self-hosting Hermes on a VPS: what would actually get you to switch, or keep you

I'm fairly new to SRE/DevOps, and one of the topics I recently spent time studying properly was EKS cluster upgrades . My first instinct, like most people starting out, was: "it's just a version bump, click upgrade in the console, done." That's basically what most beginner blog posts say too. But the more I read and the more I dug into real-world postmortems and discussions, the more I realized — the actual Kubernetes control plane upgrade is the easy part. Almost everything that can go wrong seems to happen around it, not because of it. Sharing what I learned here, mainly for my own notes, but hoping it's useful for anyone else early in their journey too. Learning #1: There's No "Undo" Button This was the first thing that surprised me. I assumed upgrades work like most software — if something breaks, you roll back. But with EKS, you cannot downgrade the control plane version once you upgrade it. So the plan can't be "upgrade, and if it breaks, revert." It has to be "test enough beforehand that breaking isn't really an option," and if something does go wrong, the fix is always moving forward, not backward. That single fact changes how you're supposed to approach the whole thing — testing has to happen before the button is clicked, not after. Learning #2: APIs Get Deprecated, and It's Usually Not Your Own Code That Breaks Kubernetes removes old API versions on a schedule. I already knew this conceptually, but what I didn't realize is that the risk usually isn't your own YAML files — it's the Helm charts and third-party tools you installed a while back and forgot about , which might still be using an older API version internally. There are tools built exactly for catching this before it becomes a problem: pluto detect-helm -owide pluto detect-files -d ./manifests kubent (kube-no-trouble) does something similar. I hadn't heard of either tool before researching this, and it made me realize how much of "being good at Kubernetes" is really just knowing which small tools e

Cisco ACE ilə işləyən administratorun qarşısında qəribə vəziyyət dayanır: cihaz zəngin funksiyalara malikdir, trafik yolunun tam ortasındadır, amma özü artıq keçmiş nəsil platformadır. Buna görə konfiqurasiyaya yalnız “request hansı serverə getsin?” sualı ilə baxmaq kifayət etmir. Tətbiqin sağlamlığı, session davranışı, SSL sərhədi və cihaz sıradan çıxanda baş verəcək hadisələr eyni xəritədə görünməlidir. Problem də budur. ACE 4710 ayrıca appliance kimi, ACE modulları isə şəbəkə avadanlığının daxilində application delivery funksiyası verirdi. Cisco bu iki məhsulu data center üçün load balancing və application delivery həlli kimi təsvir edir. Bu sinif cihaz client ilə backend arasında reverse proxy və ya Layer 4 load balancer rolunda dayanır; client virtual IP-yə qoşulur, ACE uyğun server farm-ı tapır, işlək real server seçir və bağlantını ora ötürür. Kağız üzərində sadədir. Production-da isə hər oxun öz state-i və nasazlıq ssenarisi var. Trafik ACE-dən necə keçir? Konfiqurasiyanı oxumağın rahat yolu ayrı-ayrı komandaları əzbərləmək deyil, obyektlər arasındakı yolu izləməkdir. Virtual IP xidmətin xarici ünvanıdır. Class map trafiki tanıyır, policy map həmin trafikə load balancing davranışı bağlayır, server farm backend hovuzunu saxlayır, real server isə konkret tətbiq instansiyasıdır. Health probe real serverin rotasiyada qalıb-qalmayacağına qərar verir. Diaqram — orijinal məqalədə Bu axında class map və policy map giriş trafikinin hansı xidmətə aid olduğunu müəyyən edir. Server farm seçildikdən sonra predictor işlək real serverlər arasından birini seçir. Cavab client-ə ACE üzərindən qayıdırsa, cihaz connection state-i saxlayır; asimmetrik routing yaranarsa paketlərin bir hissəsi bu state-dən yan keçə və bağlantı qırıla bilər. Deməli, routing dizaynı load balancer konfiqurasiyasından ayrı məsələ deyil. Predictor serverin həqiqi yükünü həmişə bilmir ACE-də round-robin və least connections davranışları fərqli məqsədlərə xidmət edir. Weighted round-robin standart predic

Remember when spotting a phishing email was as easy as scanning for broken English, a generic "Dear Customer" greeting, and a weird sender address that looked like a random string of numbers and letters? For years, cybersecurity awareness training focused heavily on those exact red flags. We taught teams to look for misspellings, awkward phrasing, and mismatched URLs. We built a collective intuition around digital bad hygiene. That playbook is officially obsolete. Generative artificial intelligence and large language models (LLMs) have completely rewritten the rules of social engineering. Bad grammar is gone, hyper-personalization has been automated at scale, and threat actors are no longer just typing—they’re cloning voices, automating OSINT, and orchestrating multi-channel attacks that look breathtakingly real. The Great Equalizer: How LLMs Murdered the Obvious Clue In the pre-AI era, threat actors faced a frustrating bottleneck. High-volume attacks meant blasting out cheap, poorly worded emails, while high-value spear-phishing campaigns required hours of manual research into a specific executive's writing style and background. AI completely eliminated that friction. While a human analyst might take over half a day to craft a hyper-realistic targeted lure, an LLM can generate dozens of contextually flawless variants in seconds. This shift has introduced several dangerous characteristics to modern social engineering: Native-Language Fluency: Language barriers have vanished. Scammers can use LLMs to generate native, localized content in English, French, Japanese, or any other language without a single syntactic slip-up. Automated OSINT: Attackers use automated scripts to scrape LinkedIn profiles, corporate websites, and social footprints, weaving real colleagues, ongoing projects, and corporate milestones directly into the lure. Behavioral A/B Testing: Cybercriminals treat phishing like digital growth hacking, using AI to churn out multiple narrative variations (e.g

A GitHub Copilot spending limit is a monthly budget, set in billing settings, that caps metered AI credit consumption for an enterprise, an organization, a cost center, or a single user. Creating one takes about two minutes. Knowing what it stops takes longer, and the gap between those two things is where most surprise Copilot invoices live. Two facts account for nearly all of them. On enterprise, organization and cost center budgets, the setting that actually blocks usage is off by default, so a budget in its default state is an alert rather than a limit. And no budget of any kind caps seat cost, because seats are license-based rather than metered. A spending limit governs what happens after the included credit pool runs out, and nothing before it. How to set a GitHub Copilot spending limit Budgets live in the billing settings of the account that pays. Enterprise owners and billing managers can set every budget control, including enterprise, cost center and user-level budgets. Organization owners can set a budget for their own organization, and that budget can only restrict usage further below whatever an enterprise admin has already set. It cannot raise the ceiling. The mechanics are the same at every level. Choose the budget type, which determines the metered product being measured. Choose the scope, which determines whose usage counts against it. Enter a monthly amount. Then, if the option appears, enable Stop usage when budget limit is reached and switch on threshold alerts at 75, 90 and 100 percent. That single checkbox is the whole exercise. Skip it and you have built a notification. What a GitHub Copilot spending limit actually caps GitHub splits its products into license-based and metered. For license-based products, which include Copilot seats, setting a budget does not prevent usage above the amount. It only alerts. For metered products, which include Copilot AI credits, a budget can prevent usage once the threshold is reached. The consequence is worth st

Un déploiement devrait être la chose la plus ennuyeuse de ta semaine. S'il est excitant, c'est mauvais signe. Au début de ma carrière, les mises en production étaient des événements. On retenait son souffle, on croisait les doigts, quelqu'un exécutait de mémoire une séquence d'étapes manuelles, et on regardait les journaux avec une boule au ventre. C'était palpitant. C'était aussi terrifiant, et le côté palpitant était précisément le problème : chaque déploiement était un pari, parce que chaque déploiement était un peu différent du précédent. Un bon déploiement est répétable. La même chose, de la même façon, à chaque fois — automatisée, pas récitée par un humain fatigué à la fin d'une longue journée. Quand le processus est un script plutôt qu'une cérémonie, l'ennui remplace l'angoisse. Tu ne pries plus. Tu appuies sur un bouton, et le résultat est prévisible parce qu'il a déjà été prévisible cent fois. L'automatisation fait ici plus que gagner du temps. Elle supprime toute une catégorie d'erreurs : l'étape oubliée, le mauvais paramètre, le « je croyais que tu l'avais fait ». La machine ne se fatigue pas, ne saute pas de ligne, ne se laisse pas distraire à mi-chemin. Elle rend le déploiement fiable au point d'en être ennuyeux — et l'ennui, en production, est un luxe. Alors, si tes mises en production font encore monter le rythme cardiaque, ce n'est pas de la prudence. C'est un signal. Rends-les répétables, rends-les automatiques, rends-les ennuyeuses. Garde le frisson pour ta vie ; ton système de production, lui, mérite l'ennui. – Serguey Shinder

On ne gère pas ce qu'on ne mesure pas. C'est l'une des premières leçons de l'exploitation, et pourtant je l'ai apprise à l'envers, en pilotant à l'aveugle bien trop longtemps. Sans mesure, tu ne sais pas si un système va bien. Tu le supposes. Il tourne, personne ne se plaint, donc tout va bien — jusqu'au jour où quelque chose se dégrade lentement, sous le radar, et où tu ne l'apprends que lorsque c'est déjà une panne. La lente fuite de mémoire, le disque qui se remplit, la latence qui grimpe d'une milliseconde par semaine : rien de tout cela ne crie. Ça glisse. La mesure transforme les suppositions en faits. Un tableau de bord, quelques alertes bien choisies, et soudain tu vois le problème arriver au lieu de le subir. Tu n'attends plus que l'utilisateur t'apprenne que ton système est cassé ; tu le sais avant lui. Mais il y a un piège que j'ai appris à éviter : mesurer trop. Cent métriques que personne ne regarde ne valent pas mieux que zéro. Le bruit noie le signal, et les alertes qui se déclenchent sans raison finissent par être ignorées — jusqu'à celle qui comptait vraiment. Bien mesurer, ce n'est pas tout mesurer. C'est choisir les quelques signaux qui prédisent réellement un problème. Alors, avant de bâtir la prochaine chose, demande-toi comment tu sauras si elle va mal. Si la réponse est « quelqu'un finira par le remarquer », tu ne la gères pas encore. Tu espères. Et l'espoir n'est pas une stratégie d'exploitation. – Serguey Shinder

Alle paar Jahre verkündet jemand, dass die IT-Jobs verschwinden. Die Cloud ersetzt die Administratoren. Die Automatisierung ersetzt die Operatoren. Jetzt ersetzt die KI, was übrig ist. Und alle paar Jahre verschwinden die Jobs nicht, sie wandern. Sie wandern den Stack hinauf. Als physische Server der Cloud wichen, brauchten wir keine Infrastruktur-Leute weniger, wir brauchten Menschen, die Infrastruktur auf einer höheren Ebene verstehen, die Systeme entwerfen, absichern und kostenbewusst betreiben, statt Hardware in Racks zu schrauben. Als manuelle Deployments der Automatisierung wichen, brauchten wir Menschen, die die Automatisierung selbst bauen konnten. Jede Welle entfernte nicht die Arbeit. Sie hob den Boden und verschob den wertvollen Teil nach oben. Die KI ist die nächste Welle, und ich erwarte, dass sie sich genauso verhält. Sie wird viel Routine aufsaugen: das Skripten, die ersten Config-Entwürfe, die Standard-Fehlersuche. Was sie nicht aufsaugt, ist Urteilsvermögen: zu wissen, was zu bauen ist, zu entscheiden, was das Risiko wert ist, zu verstehen, wie die Teile einer echten Organisation zusammenpassen, und geradezustehen, wenn etwas schiefgeht. Wer in jedem Übergang strauchelt, sind die, die sich über die Aufgabe definieren, die automatisiert wurde. Wer gedeiht, definiert sich über das Problem, das er löst, und lässt die Werkzeuge dafür sich darunter ändern. Verteidige also nicht das eine, das du heute tust. Werde gut in der Schicht darüber. Der IT-Job hat jedes Werkzeug überlebt, das ihn beenden sollte, indem er den Stack hinaufwanderte. Diesen wird er genauso überleben. – Serguey Shinder

Microsoft is placing greater emphasis on controlling disruption in Azure Kubernetes Service (AKS) Node Auto-Provisioning (NAP), publishing new guidance to help platform teams balance the efficiency benefits of automated node consolidation with application availability. By Craig Risi

If you run ClickHouse in production, you're probably on 26.3 LTS. And now 26.8 LTS has been announced, which means the LTS-to-LTS upgrade conversation starts again. Here's the thing most release posts skip: this is not a one-release hop. Going from 26.3 LTS to 26.8 LTS means crossing 26.4, 26.5, 26.6 and 26.7 as well. Every breaking change in those four releases applies to you, and some of the ones most likely to ruin your day aren't in 26.8 at all. So instead of writing another "here are the 26.8 features" post, I wanted to write the thing I'd actually want before scheduling this upgrade: what breaks, what silently changes, what order to do things in, and what you get for the trouble. A note on release timing As of writing (27 August 2026), 26.8 has been announced but is not fully released yet. The release branch is cut and versioned (v26.8.1.1-lts), but the tag and Docker images have not been published yet, and the upstream changelog still marks the 26.8 section as in progress. By the time you read this, the tag has probably landed. Check for yourself: curl -s https://raw.githubusercontent.com/ClickHouse/ClickHouse/master/utils/list-versions/version_date.tsv \ | awk -F '\t' '$1 ~ /^v26\.8\./ {print "26.8 is released - newest: " $1 " (" $2 ")"; f=1; exit} END {if (!f) print "26.8 not released yet"}' version_date.tsv is the list ClickHouse maintains of every released version and its date, so this is the most direct answer available - no auth, no rate limit, nothing to download. As of writing it prints 26.8 not released yet . Worth knowing: the Docker image will lag whatever that command tells you. The Docker Official Images repo trails the GitHub tags by a few patch versions - clickhouse:lts currently resolves to 26.3.20.7 even though 26.3.24.4 has already shipped. So don't treat a missing image as evidence the release hasn't happened. Either way, the timing works in your favour. Historically ClickHouse LTS releases pick up several patch releases quickly - 26.7 had