Topic
Infrastructure
24 posts · Newest first
-
A monitoring tool caused most of the monthly requests
Out of 839k monthly requests to one platform, 512k came from a sitemap status check running every few minutes.
-
A full infrastructure audit is underway
An infrastructure audit has started, with more writing planned on the findings.
-
Upgrading Bitcoin Core from 27.0 to 32.1
What changed between Bitcoin Core 27.0 and 32.1: AssumeUTXO on mainnet, a 1GB default dbcache, and parallel block input fetching for faster initial sync.
-
Non-profits and the Internet
Signal and Let's Encrypt stand out for their contributions to the Internet.
-
GitHub is throwing 500s
GitHub returns server errors, so the evening turns into a forced break.
-
Approve hit counts look unreal
Approve.so October traffic jumps well past the prior daily pace, and the spike looks worth debugging.
-
Groq powering McLaren F1
Groq showing up as the compute behind McLaren's Formula 1 effort.
-
Cloudflare KV Instant pricing and tradeoffs
Notes on KV Instant’s fast reads versus steep per-MB storage and per-operation Class A costs.
-
Has the Rust server moved into production?
A question for 37signals after their Basecamp outage, noting reported Rust rewrite gains of over 50x.
-
LoomUp server response timings
A breakdown of LoomUp's sub-millisecond stages, with most of the remaining 80 ms blamed on the DEL-BLR trip.
-
Approve hits a million requests
Approve.so crosses one million September hits across assets, push, sockets, and CRUD, with AI next.
-
Linux machine booting in under 10ms
A reaction to boxd's claim of a whole Linux machine that boots in under 10 milliseconds.
-
When the usage meter costs more than usage
A Cloudflare cost comparison shows how chatty Durable Object writes can dwarf the underlying usage.
-
Cutting API response time by over a second
WaitUntil accounting, fewer D1 round trips, and Smart Placement reclaim more than a second per request.
-
Project L cuts monitoring check frequency
SSL, DNS, Lighthouse, and related checks across 23 projects are paced down to ease server load.
-
Cloudflare 525s and Caddy peer retransmits
Project A returns 525s while project L stays fast, and some Caddy–Cloudflare peers show heavy retransmits.
-
A 4.94 GB database copy in 70 seconds
A schema-change workflow hits a snag while copying and verifying an existing database.
-
Saving 100 TB of RAM across a fleet
A quick appreciation of saving 100 TB of RAM across a fleet.
-
Questioning three million requests in a day
Investigating whether excessive crawling or unoptimized network calls explain a surprising request count.
-
43 requests per second and a 341 ms p95
Checking CPU usage, request throughput, and latency as a backend project grows.
-
Three million daily requests on my Rust server
Comparing a custom Rust server’s growing request volume with Cloudflare Workers pricing.
-
Bots generating their own bots
Questions about nested bots, account access inside containers, and parent-account monitoring.
-
Rethinking the Rust and SQLite service architecture
Weighing the current service against a split between API workers and the database layer.
-
One backend, two very different project outcomes
Trying to understand why one project works and another fails on the same backend service.