标签：#ki

Production lock analysis: vì sao pg_stat_activity một mình không đủ, và join với pg_locks mới ra root cause Lock contention trong Postgres hiếm khi báo bằng error — nó báo bằng wait_event_type = 'Lock' ở pg_stat_activity và bằng latency tăng từ phía application. Khi một incident xảy ra ("API treo, không ai biết tại sao"), thứ team cần trong 60 giây đầu là một bức tranh: PID nào đang đợi, đợi lock loại gì trên object nào, bị block bởi PID nào, PID block đó đang chạy query gì và đã giữ transaction bao lâu . pg_stat_activity một mình chỉ trả lời được nửa câu hỏi ("ai đang đợi"); pg_locks một mình chỉ trả lời nửa còn lại ("ai giữ gì"). Phải join hai view này — và bám theo pg_blocking_pids() — để dựng được blocking tree. Không có dashboard cho luồng dữ liệu này là lý do điển hình một production freeze kéo dài 30 phút thay vì 3 phút: incident commander phải mò ad-hoc bằng psql , gõ sai query, miss idle in transaction đang giữ AccessExclusiveLock của một migration nửa đời trước. Cơ chế hoạt động pg_locks là một view phơi nội dung trực tiếp của shared lock manager trong shared memory. Mỗi dòng là một lock request (đã granted hoặc đang chờ) thuộc một backend. Theo Postgres docs phần "System Views → pg_locks", các column then chốt: locktype ( relation , transactionid , tuple , virtualxid , advisory ...), relation (OID — join pg_class ), transactionid , virtualtransaction , pid (backend PID), mode ( AccessShareLock , RowExclusiveLock , ShareUpdateExclusiveLock , AccessExclusiveLock ...), granted (bool), fastpath (lock đi qua fast-path tránh shared lock manager), và waitstart (timestamp bắt đầu chờ — bổ sung sau v14, hữu ích để đo lock wait time mà không cần snapshot diff). pg_stat_activity là view phơi trạng thái runtime của mỗi backend: pid , usename , datname , application_name , client_addr , backend_start , xact_start , query_start , state ( active , idle , idle in transaction , idle in transaction (aborted) ), wait_event_type , wait_event , backend_xid , backend_xmin , qu

Deadlock detection: vì sao Postgres chỉ dò cycle sau deadlock_timeout , và vì sao retry mù làm vỡ thêm Deadlock xảy ra khi hai (hoặc nhiều) transaction đang giữ lock mà bên kia cần, tạo thành một vòng tròn đợi không thoát ra được nếu không có trọng tài. Postgres là trọng tài đó: nó dựng wait-for graph từ trạng thái lock hiện tại, phát hiện cycle, và chủ động abort một transaction để cắt vòng. Dev gặp class lỗi này không phải vì viết code "sai" mà vì hai code path đụng cùng tài nguyên theo thứ tự khác nhau dưới load cao — ở môi trường dev một mình thì không thấy, lên production concurrent đủ thì ERROR: deadlock detected (SQLSTATE 40P01 ) ném ra, một transaction bị rollback, request user trả 500. Hậu quả nặng nhất không phải bản thân deadlock, mà là retry layer viết kiểu "cứ lỗi thì retry vô hạn": cùng pattern lock sai thứ tự, cùng tạo lại cycle, deadlock counter tăng tuyến tính với traffic, log Postgres ngập, và database CPU bị burn vào việc chạy graph algorithm thay vì việc thật. Cơ chế hoạt động Postgres không dò deadlock liên tục. Mỗi lần một backend phải đợi lock, nó đăng ký với lock manager và đi ngủ; chỉ khi ngủ vượt deadlock_timeout (mặc định 1 giây, GUC ở runtime-config-locks ), backend đó mới được đánh thức để chạy thuật toán dò cycle. Đây là quyết định thiết kế nêu trong Postgres docs về Lock Management: phần lớn lock chờ trong workload bình thường chỉ kéo dài vài ms (chờ một UPDATE ngắn commit) — chạy graph algorithm trên mọi lần lock-wait sẽ tốn CPU vô ích. Lazy detection đánh đổi: nếu có cycle thật, mọi transaction trong cycle treo ít nhất deadlock_timeout trước khi được giải phóng. Khi timer kích hoạt, backend đó dựng wait-for graph từ lock manager: với mỗi process đang đợi lock, vẽ một cạnh hướng từ "process đợi" sang "process giữ lock đó". Postgres còn phải xử lý lock soft-conflict (hai process cùng đợi, không ai giữ — nếu được sắp xếp lại thứ tự thì cả hai đều chạy được mà không cần kill ai); thuật toán thật trong deadlock.c (source tree Postgres) th

DDL lock: vì sao một ALTER TABLE 50ms vẫn đủ làm sập cả API trong giờ cao điểm DDL trong Postgres không phải "vài lệnh schema chạy nhanh". Hầu hết các form của ALTER TABLE , toàn bộ DROP TABLE , TRUNCATE , REINDEX , CLUSTER , VACUUM FULL , và cả REFRESH MATERIALIZED VIEW (không có CONCURRENTLY ) đều yêu cầu ACCESS EXCLUSIVE — lock mode mạnh nhất, xung đột với mọi mode khác kể cả ACCESS SHARE mà một SELECT thuần đọc cũng cần. Một DDL chạy 50ms vẫn có thể chôn cả service vài phút vì hai thứ Postgres làm theo design: lock được giữ tới hết transaction (không nhả sớm), và lock queue là FIFO — bên đến sau dù mode tương thích vẫn phải đứng sau bên đang đợi. "Migration chạy lúc giờ cao điểm, toàn bộ API treo 5 phút" gần như luôn là class incident này. Cơ chế hoạt động Mỗi lệnh DDL acquire một relation-level lock trên các object nó chạm, theo bảng mode cố định trong Postgres docs mục Explicit Locking . Có ba nhóm cần thuộc: ACCESS EXCLUSIVE — xung đột với mọi mode. Cấp bởi: phần lớn ALTER TABLE (kể cả những form không rewrite data), DROP TABLE , TRUNCATE , REINDEX (non-concurrently), CLUSTER , VACUUM FULL , REFRESH MATERIALIZED VIEW (non-concurrently), LOCK TABLE không kèm mode. Đây là lock "đông cứng" object. SHARE — xung đột với mọi mode ghi ( ROW EXCLUSIVE , SHARE UPDATE EXCLUSIVE , SHARE ROW EXCLUSIVE , EXCLUSIVE , ACCESS EXCLUSIVE ). Cấp bởi: CREATE INDEX (non-concurrently). Cho đọc đi qua, nhưng chặn mọi INSERT / UPDATE / DELETE — trên bảng OLTP nóng đó là downtime ghi. SHARE UPDATE EXCLUSIVE — xung đột với chính nó và các mode mạnh hơn, nhưng không xung đột với ROW EXCLUSIVE . Cấp bởi: CREATE INDEX CONCURRENTLY , REINDEX CONCURRENTLY , VACUUM (không FULL), ANALYZE , ALTER TABLE ... VALIDATE CONSTRAINT , ALTER TABLE ... SET STATISTICS , ALTER INDEX ... RENAME . Đây là mode "online maintenance": cho cả đọc lẫn ghi đi qua, chỉ tự khoá lẫn nhau. Quan trọng: lock giữ tới hết transaction , không có cách nhả sớm. BEGIN; ALTER TABLE ...; <30 phút làm việc khác>; COMMIT; giữ A

FOR UPDATE: pessimistic row lock để chặn lost update, và cái giá deadlock khi không lock theo thứ tự SELECT ... FOR UPDATE là cách rõ ràng nhất để nói với Postgres "tao sẽ sửa row này, đừng cho ai khác đụng vào cho tới khi tao commit". Nó là một row-level lock thật sự — khác SELECT thường (chỉ chụp snapshot MVCC, không ngăn ai update song song). Lý do dev gặp nó trong việc thật là class bug lost update : hai transaction cùng đọc một row, cùng tính giá trị mới dựa trên giá trị đọc được, rồi cùng UPDATE — bản ghi cuối đè bản trước, một nửa thay đổi biến mất không log lỗi gì. FOR UPDATE ép hai bên xếp hàng tại bước đọc, một bên đợi bên kia commit rồi tự đọc lại bản mới. Đổi lại, nếu nhiều code path khoá nhiều row theo thứ tự khác nhau, Postgres sẽ bắn ERROR: deadlock detected và một bên transaction bay theo. Cơ chế hoạt động Khi một transaction chạy SELECT ... FOR UPDATE , Postgres không ghi row lock vào lock table chính (như cách nó làm với relation-level lock). Thay vào đó, nó ghi xid của transaction hiện tại vào xmax của chính tuple đó trên heap, kèm cờ infomask đánh dấu "đây là lock chứ chưa phải delete". Hệ quả: ôm row lock cho hàng triệu row gần như không tốn shared memory. Khi một transaction khác chạm cùng row (qua UPDATE , DELETE , hay một SELECT ... FOR UPDATE nữa), nó đọc xmax , thấy transaction kia còn sống, và đăng ký một heavyweight lock kiểu transactionid trong pg_locks để đợi xid đó kết thúc. Đó là cơ chế "đợi xid" lộ ra qua wait_event = 'transactionid' ở pg_stat_activity . Khi bên giữ COMMIT hoặc ROLLBACK , bên đợi được đánh thức, đọc lại tuple (visibility check theo isolation level), rồi mới chạy tiếp. -- Session A BEGIN ; SELECT id , balance FROM accounts WHERE id = 42 FOR UPDATE ; -- giữ row lock trên id=42, chưa commit -- Session B (terminal khác) BEGIN ; UPDATE accounts SET balance = balance - 100 WHERE id = 42 ; -- treo, đợi xid của Session A FOR UPDATE có bốn biến thể, mạnh dần ngược lại: FOR KEY SHARE (yếu nhất, chỉ chặn thay đổi key — đây là l

Lock trong Postgres: hai tầng heavyweight/lightweight, vì sao một ALTER đứng yên kéo theo cả service Postgres chia khoá thành hai tầng chạy song song. Heavyweight lock (Postgres docs gọi là "regular locks" hay table-level/row-level lock) giữ vai trò bảo vệ object logic — relation, tuple, transactionid — sống suốt một transaction, hiện ra trong view pg_locks và là loại mà application code "thấy" được. Lightweight lock (LWLock) bảo vệ các vùng nhớ chia sẻ bên trong server — buffer descriptor, WAL insert position, hash table của chính lock manager — chỉ ôm trong thời gian rất ngắn, không hiện trong pg_locks nhưng lộ ra qua wait_event_type = 'LWLock' ở pg_stat_activity . Hai tầng có hai cơ chế chờ khác nhau, hai loại sự cố production khác nhau, và nếu chỉ "biết row lock" thì sẽ bỏ qua hai class incident kinh điển: một ALTER TABLE đứng đợi ACCESS EXCLUSIVE làm đông cứng mọi SELECT đến sau nó, và LockManager partition LWLock thành điểm nghẽn khi đụng nhiều partition trong một query. Cơ chế hoạt động Heavyweight lock được quản lý bằng một hash table trong shared memory, chia thành nhiều partition để giảm contention (số partition là NUM_LOCK_PARTITIONS , set tại compile time). Mỗi entry ứng với một (locktag, mode) ; locktag là tuple (locktype, database, relation, ...) nhận diện object cần khoá. Tám lock mode trên relation chia thành ma trận xung đột cố định: ACCESS SHARE (cấp bởi SELECT ) xung đột duy nhất với ACCESS EXCLUSIVE (cấp bởi DROP TABLE , TRUNCATE , hầu hết ALTER TABLE , REINDEX , CLUSTER ); ROW EXCLUSIVE (cấp bởi INSERT / UPDATE / DELETE ) xung đột với SHARE trở lên; SHARE UPDATE EXCLUSIVE (cấp bởi VACUUM không-FULL, ANALYZE , CREATE INDEX CONCURRENTLY ) tự xung đột với chính nó — đó là lý do chỉ một autovacuum hoặc một CREATE INDEX CONCURRENTLY chạy mỗi lần trên cùng bảng. Ma trận này nằm trong docs Postgres mục Explicit Locking và là kiến thức nền cho mọi quyết định migration online. -- xem mode đang giữ trên một bảng SELECT pid , mode , granted , locktype , re

Grow a backyard’s worth of greens and vegetables in your house with a vertical hydroponic garden. Here are a few that might be worth the investment.

Cursor has spent the past week in headlines after confirming a partnership with SpaceX that could eventually lead to a $60 billion acquisition . The deal, for now, centres on training more capable coding models using SpaceX’s compute infrastructure. Alongside that push on model performance, however, Cursor is now addressing a separate issue: the reliability of the code those models produce. Cursor has partnered with Chainguard , which provides verified open-source packages, to route dependencies through its curated repositories, aiming to reduce the risk of compromised components entering AI-built applications. The announcement lands as AI coding tools push more software into production with less human review, raising questions about how much of that code can be trusted. Supply chain risks in the agentic era The partnership addresses a problem developers know all too well. Modern applications depend heavily on open-source libraries and container images, most of which are pulled from public registries such as npm, PyPI, and Docker Hub. Those registries operate on openness, with limited checks in place. Developers — and now AI agents — often install dependencies without knowing who built them or whether they have been tampered with. Recent incidents have underlined the risk . In March, projects such as Trivy, LiteLLM, Telnyx, and Axios were compromised, with attackers using poisoned packages to steal credentials and spread malware. For teams using AI-generated code, the exposure increases. Agents can select and install dependencies automatically, making trust decisions at a pace that outstrips manual review. As Chainguard co-founder and CEO Dan Lorenc put it, generating code is becoming routine — checking its integrity is where the pressure now sits. “AI agents are making dependency decisions at a scale and speed no security team can manually review,” he wrote in a blog post . “As organizations adopt agentic development, the biggest blocker is no longer how fast code

"It works on my test input" is the most dangerous sentence in on-device AI development. I typed that sentence - or some version of it - a dozen times while building Redacto, our on-device PII redaction app running Gemma 4 E2B on a Samsung Galaxy S25 Ultra. The model would redact a patient name from a clinical note, I would nod, and I would move on. Then I would hand the phone to a teammate, they would type a police report, and the model would redact the suspect description instead of the victim name. The problem is not the model. The problem is that manual spot-checking is not validation. You are testing a single input against your own expectations, with all the confirmation bias that entails. When you have five domain modes (HIPAA, Financial, Tactical, Journalism, Field Service), three difficulty levels, and two candidate models, you need something systematic. You need a benchmark suite. This post covers how I built one - from dataset curation to scoring methodology to on-device infrastructure - for a hackathon app running entirely on a phone. No cloud. No API calls. No data leaving the device. Why Not Use an Existing Framework? The LLM evaluation space has mature tools. EleutherAI's lm-eval-harness is the community standard for evaluating language models against academic benchmarks like MMLU, HellaSwag, and ARC. Stanford's HELM (Holistic Evaluation of Language Models) provides a multi-metric evaluation framework with standardized scenarios. Google's BIG-bench offers hundreds of tasks for probing specific capabilities. These frameworks are excellent for what they do. They are also completely wrong for this problem, for three reasons. First, they assume server-side inference. lm-eval-harness expects to call a model through an API or load it in PyTorch on a GPU server. Redacto's model runs on a Qualcomm Hexagon NPU inside a phone. There is no Python runtime, no HuggingFace tokenizer at evaluation time, no way to hook into the framework's inference loop. Second, their

Understanding the 550 SPF Check Failed Error The "550 SPF Check Failed" error indicates that a receiving mail server rejected an incoming email. This rejection occurs because the sender's domain failed its Sender Policy Framework (SPF) validation. SPF is an email authentication protocol defined in RFC 7208 . SPF helps prevent email spoofing. It allows domain owners to specify which mail servers are authorized to send email on behalf of their domain. Receiving mail servers perform an SPF check by querying the sender's DNS for an SPF TXT record. If the sending server's IP address is not listed in the domain's SPF record, the SPF check fails. The receiving server then rejects the email based on its configured policy, often resulting in a 550 error. This error protects recipients from unauthorized emails and enhances email security. Initial Diagnosis: Identifying the Root Cause Diagnosing an SPF failure requires examining the bounce message and the domain's DNS records. The bounce message often provides specific details about the SPF failure. Look for phrases like "SPF validation failed," "unauthorized sender," or "IP address not permitted." Common reasons for a 550 SPF Check Failed error include: Missing SPF Record: No SPF TXT record exists for the sending domain. Incorrect SPF Syntax: The SPF record contains errors, making it unreadable or invalid. Incomplete SPF Record: The SPF record does not list all legitimate sending IP addresses or hostnames. DNS Lookup Limit Exceeded: The SPF record requires more than 10 DNS lookups, violating RFC 7208. DMARC Policy Enforcement: A DMARC (Domain-based Message Authentication, Reporting, and Conformance) policy ( RFC 7489 ) with p=reject or p=quarantine is in place, enforcing strict SPF failure handling. To begin diagnosis, use our SPF checker to verify your domain's SPF record and its validity. This tool quickly identifies syntax errors and lookup issues. Step-by-Step Troubleshooting and Resolution Resolving SPF failures involves

Understanding DMARC and the 'p=none' Policy DMARC (Domain-based Message Authentication, Reporting, and Conformance), defined in RFC 7489, is an email authentication protocol. It builds upon SPF (Sender Policy Framework, RFC 7208) and DKIM (DomainKeys Identified Mail, RFC 6376) to provide domain owners with greater control. DMARC instructs recipient mail servers on how to handle emails that fail authentication and provides reporting on these failures. The p=none policy is often adopted as a preliminary step in DMARC implementation. It instructs recipient servers to take no specific action on emails failing DMARC alignment. Its primary function is to enable the collection of aggregate and forensic reports without impacting email deliverability. Many organizations view p=none as a safe, non-disruptive way to begin their DMARC journey. This initial perception, however, overlooks critical security implications. While it offers visibility, p=none provides no actual enforcement against malicious email. The Critical Security Vulnerability of p=none The fundamental flaw of DMARC p=none lies in its complete lack of enforcement. When a DMARC record is set to p=none , recipient mail servers will not block, quarantine, or reject messages that fail DMARC authentication. This includes emails that spoof your domain directly. Threat actors exploit this vulnerability to conduct phishing, business email compromise (BEC), and brand impersonation attacks. They can send emails appearing to originate from your legitimate domain, knowing that p=none offers no protective barrier. The recipient mail server simply delivers the fraudulent message. This policy effectively leaves your domain unprotected against direct domain spoofing. Despite having a DMARC record, your organization remains susceptible to advanced phishing techniques. The security posture of your email ecosystem is compromised. The Illusion of Insight: Data Without Action DMARC p=none does provide valuable data through its repor

Introduction to DMARC and the p=quarantine Policy DMARC (Domain-based Message Authentication, Reporting, and Conformance), defined in RFC 7489 , is an email authentication protocol. It builds upon SPF and DKIM to provide domain owners with the ability to protect their domain from unauthorized use. DMARC enables senders to specify how receiving mail servers should handle unauthenticated emails originating from their domain. It also provides a mechanism for receiving servers to report back to the domain owner about authentication results. DMARC policies dictate the action receiving mail servers should take when an email fails DMARC authentication. The three primary policies are: p=none : Monitor mode. Receiving servers take no action on failed messages but send reports. This is the initial deployment phase. p=quarantine : Receiving servers should treat failed messages as suspicious. They are typically placed in the recipient's spam folder or flagged for further review. p=reject : Receiving servers should outright reject messages that fail DMARC authentication. This is the strongest enforcement policy. Implementing p=quarantine is a critical step towards full domain protection. It allows domain owners to mitigate spoofing and phishing attempts without immediately blocking legitimate, but misconfigured, email streams. This policy provides a balance between security enforcement and minimizing potential deliverability disruptions. Prerequisites for DMARC p=quarantine Implementation Before deploying a p=quarantine policy, proper configuration of SPF and DKIM is mandatory. DMARC relies on these underlying authentication mechanisms and their alignment with the sending domain. SPF (Sender Policy Framework) SPF, specified in RFC 7208 , allows domain owners to publish a list of authorized sending IP addresses in their DNS. Receiving mail servers check the SPF record to verify if an incoming email originated from an authorized server. An SPF record is a TXT record at the root of

A cat astrologer, spec-driven and running on Amazon Bedrock A companion to A Builder in Paris: Do Devs Dream of Électrique Chats? Last month I wrote about the idea. Six rainy days in Paris, a closed laptop, and a hackathon I did not mean to enter, and somewhere between the Musée de l'Orangerie and a lot of walking, an idea arrived. Cats are inscrutable. The people who love them are obsessed with understanding them anyway. Astrology is an old framework for making the unknowable feel readable, and maybe, just maybe, it helps us understand them a little. Her name is Madame Minou , a French cat astrologer who reads your cat's stars from a café terrace. That first article was the idea . This one is the build. Vibe-coded, but on rails Was it vibe-coded? You know it! AI wrote the lines, and I said "no, not like that" more times than I can count. But it was vibe-coding on rails, and the rails were Kiro. Before a single line of app code, I wrote the requirements in EARS notation, a design doc, and a build-ordered task list, all living in .kiro/specs . Decide what "done" means before letting anyone, human or model, start building. The specs are what kept the vibes on track. Then the steering files. .kiro/steering held the enduring rules of the project: product principles, security guardrails, technical direction, and UI law. These were the thing that kept a long, multi-session build from drifting. When a new session opened, the steering files were already the shared context. "The café blue" was one token, not five guesses. Security was not optional. The garbled café sign was a deliberate easter egg, not a bug to fix. From there, the loop: Kiro implemented one approved block at a time, ran each task's PASS/FAIL QA gate on itself before moving to the next, and only stopped for my review on the two things that actually mattered. I directed and approved. Kiro proposed and built. Spec first, block by block, human in the loop. The facts are sacred Here is the part that looks like a

สวัสดีครับเพื่อนๆ! 👋 วันนี้จะมาเล่าเรื่องน่าตื่นเต้นให้ฟังนะเพื่อนๆ สำหรับใครที่เป็นสาย SvelteKit เตรียมตัวอัปเดตความรู้ใหม่กันได้เลย เพราะตอนนี้เขามีของเล่นใหม่ที่กำลังอยู่ในช่วงทดลองใช้งาน แต่บอกเลยว่าว้าวมาก! เราไปดูกันดีกว่าว่ามันคืออะไร... 📡 Remote function คืออะไร เป็น function ตัวใหม่ ✨ (ที่คาดว่าจะเป็น new way to implement สำหรับ Sveltekit 3.0) เอาไว้ใช้สื่อสารพูดคุยกันระหว่างฝั่ง client และ server ของ Sveltekit นั่นเอง 💬 ความเจ๋งคือเราสามารถเรียกใช้มันจากมุมไหนของ Sveltekit ก็ได้ 🌍 ไม่จำเป็นต้องจำกัดแค่ฝั่ง server หรือ client แต่จุดสำคัญคือ การทำงานของมันจะเกิดขึ้นที่ฝั่ง server เสมอ 👍 นั่นหมายความว่ามันสามารถทะลุทะลวงไปดึงข้อมูลหรือโมดูลที่เป็น server-only ได้สบายๆ เช่น ตัวแปร environment ที่เราประกาศไว้ หรือพวกฐานข้อมูลต่างๆ ก็ดึงมาได้ชิลๆ เลย 😎 เวลาจะใช้งาน เราจะต้องใช้ท่าการ await แบบใหม่ของ Sveltekit ⏳ ที่ช่วยให้คุณโหลดหรือดึงข้อมูลแบบ promise มาใช้ใน component ของคุณได้ทันที 🚀 ⚠️ หมายเหตุ: ตอนนี้ทั้ง await และ remote function ยังอยู่ในช่วงทดลองใช้งาน 🧪 (experimental) นั่นแปลว่า syntax บางอย่างอาจจะมีการปรับเปลี่ยนหรือบินหายไปบ้างในอนาคต 🥲 แต่แกนหลัก (core functional) ของมันก็จะยังทำงานได้ตามที่เราคาดหวังแน่นอน ถ้าใครคันไม้คันมืออยากลองของใหม่ตอนนี้ สามารถไปเปิดโหมด experimental ได้ที่ไฟล์ svelte.config.js(.ts) ตามโค้ดด้านล่างนี้เลย 👇 svelte.config.js(.ts) /** @type {import('@sveltejs/kit').Config} */ const config = { kit : { experimental : { remoteFunctions : true } }, compilerOptions : { experimental : { async : true } } }; export default config ; 🏃‍♂️ Let get started!! เราสามารถเริ่มใช้ remote function ได้ง่ายๆ ผ่านการสร้างไฟล์นามสกุล .remote.js หรือ .remote.ts 📝 ซึ่งตอนนี้มี function ให้เราหยิบมาเล่นทั้งหมด 4 ตัวด้วยกันคือ: query (ที่เราจะมาพูดถึงกันในบทความนี้) form command prerender หลักการทำงานเบื้องหลังคือ เวลาที่เรา import ตัว remote function ไปใช้ในฝั่ง client มันจะถูกแอบแปลงร่างเป็นโค้ดที่หุ้มด้วย fetch ในช่วง build time 🏗️ นั่นหมายความว่าระบบจะใจดีสร้างเส้น HTTP endpoint ให้เราแบบอัตโนมัติ ✨ ด้วยเหตุนี้เราเลยเอาไฟล์ .remote.js หรือ .remote.

“It is a direct attack on the rule of law,” says one European Parliament member of the new findings from Citizen Lab.

Digital Signatures: Format, Certificate, and Validation Policy Are Not the Same Thing The right move when a digital signature fails validation is don't look at the cryptography first . I know that sounds backwards. If the algorithm is RSA-2048 and the certificate chain is intact, why would validation fail? Because a signature can be cryptographically perfect and still get rejected by the validator. The problem isn't the hash or the private key — it's the format, the wrong certificate, or the validation policy the system is applying. My thesis is simple: most errors that look cryptographic in digital signatures are actually upper-layer errors — incompatible format, a certificate that doesn't meet the required profile, or a validation policy that the issuer and receiver never aligned on. And confusing those three layers has a real cost: debugging time wasted in the wrong place. The Real Mess: Three Layers People Keep Mixing Up When a digital signature fails validation, the typical mental sequence is: "Is the algorithm correct? Does the private key match the public one? Did the certificate expire?" Those are reasonable questions, but they're all in the same layer. The problem is there are three distinct layers, and each one can fail independently. Layer 1 — Signature Format The format defines how the signature is packaged together with the signed data. CMS/PKCS#7 is not the same as XAdES, PAdES, or JAdES. Each has variants: BASELINE-B , BASELINE-T , BASELINE-LT , BASELINE-LTA . Choosing CAdES-BASELINE-B when the receiver expects XAdES-BASELINE-LT produces a rejection that has nothing to do with the cryptographic algorithm. The public documentation for DSS (Digital Signature Service) from the European Commission describes these variants in detail. DSS is the reference library for eIDAS-compliant signatures, and its documentation is one of the most complete and verifiable resources publicly available. Layer 2 — Certificate The certificate is the signer's identity, but it

The homelab started flat. One /24, everything on it. My workstation, the NAS, the Proxmox host, and — over time — a growing list of workloads sharing the same broadcast domain because that was the path of least resistance. For a while, that was fine. A homelab running one workload doesn't need segmentation any more than a house needs an office door. Then I stood up an Akash provider. An Akash provider is, in shape, a Kubernetes cluster that accepts inbound tenant workloads from the internet — real deployments, paying for compute, containers I didn't write landing in namespaces on my hardware. The provider itself is documented at github.com/jjozzietech/akash-provider-ops-public — this piece is about the network underneath it. The containerisation posture itself is fine. I trust the isolation model. But trust isn't a network design. And the network at that moment had the tenant workload cluster sitting on the same subnet as my workstation, my NAS, and my Proxmox management interface. That was the moment I stopped thinking of the rack as a home network with extra boxes, and started thinking of it as a small data centre. This piece is the network design that came out of that shift. I'll cover the layout, the rules that hold it together, and the Nexus and Proxmox configs that anchor it — with the specifics of my own deployment sanitised. It's not a step-by-step replication guide. It's the design pattern, with enough of the shape to be useful and enough restraint to not double as a recon document for my own rack. // the original design The flat layout looked like this: home lan — 192.168.1.0/24 opnsense (perimeter) cisco nexus (dumb L2 switching) proxmox host workload VMs (all on the same subnet) What it got right: zero routing complexity, everything reachable from everywhere, fast to stand up. If you're running one project on a homelab, this is the correct design. Don't over-engineer it. What stopped working, as soon as the second project landed on the rack, was that the

Engineering teams building with AI agents have largely solved the single-agent problem. The harder challenge arrives when capabilities get split across multiple independently deployed agents — each owned by a different team, each running on its own release cadence. Keeping a coherent conversation alive across those boundaries turns out to be one of the messier architectural questions in production agent systems today, and one that Tessl's own work on context engineering and skill sprawl has been circling from a different angle. Microsoft's Industry Solutions Engineering ( ISE ) team, which embeds with clients on complex technical engagements, has published a detailed account of how they tackled that context problem in a recent engagement. Working with Agent2Agent (A2A ) — an open agent communication protocol originally developed by Google and now maintained by a cross-vendor technical steering committee at the Linux Foundation — they needed coordinator agents to hand off conversational history to domain agents that held no shared infrastructure and no persistent memory. Where the Model Context Protocol ( MCP ) standardises how agents connect to tools and data, A2A operates at a different level : it defines how agents communicate with each other as peers, passing tasks and messages across service boundaries. Shared storage creates dependencies agents shouldn't have Microsoft says it evaluated three core approaches before settling on the one that worked best. The first option entailed domain agents reading from a shared storage layer, using a common identifier to retrieve conversation history. The appeal with this is minimal message size and a single source of truth, but it requires every domain agent to have credentials and connectivity to storage owned by another team — a dependency that becomes unwieldy fast when agents cross organisational lines. A second option makes each domain agent stateful, maintaining its own record of the conversation. However, the operatio

Fourth of July weekend is the last great grill and griddle sale of the summer, including $250 off my favorite pellet smoker.

You have seen this loop before. An agent starts a “simple” task, say scrape listings, refactor a repo, research a market, or whatever. It fails, it retries, it re-reads context, it apologizes and tries all over again. Twenty minutes in and the dashboard shows six figures of tokens and zero useful outputs or deliverables. The model did not misbehave on purpose. The orchestrator never had a hard budget gate with an ROI in mind. Skillware v0.4.0 ships a new skill for exactly that gap: monitoring/token_limiter . It lets you monitor and limit any agent’s token budget in real time — Gemini, Claude, OpenAI, DeepSeek, Ollama, custom Python loops, you name it. Same skill, same JSON, any runtime. What Skillware is in a nutshell Skillware is an open registry of installable agent capabilities . Each skill is a bundle: skill.py — deterministic Python ( execute() returns JSON) instructions.md — when the model should call the tool manifest.yaml — schema, constitution, issuer Tests and docs — shipped in the wheel You load by ID, adapt for your provider, call execute() on tool use. The model decides when , the skill decides how , predictably, every time. That split matters for budget control. You do not want the LLM guessing whether it is “allowed” to spend more tokens. You want a small, auditable function that answers: continue, warn, or stop. Meet the Token Limiter This skill is a budget gate , not a kill switch wired into OpenAI or Anthropic. After each model turn, your host loop passes cumulative usage. The skill returns one of three actions: Action Meaning CONTINUE Under the soft threshold — keep going WARN Approaching the limit (default 80%) — tighten scope FORCE_TERMINATE Hard ceiling hit — stop the loop Important nuance: the skill does not cancel API sessions or kill processes. It returns a structured decision. Your orchestrator must act on it. That is by design — Skillware skills stay portable and provider-neutral. No skill-specific API keys. No network calls. Pure Python m