Skip to main content

notes from the workshop.

Field notes on uptime monitoring, status pages, and the unglamorous work of keeping production boring.

2026-11-15

Caddy ran out of memory, and the heap was not the problem

Caddy went from 94 MiB to its 1 GiB cap in three minutes. The access log blamed a scanner. The memory was in goroutine stacks, and the bug was in a library.

  • caddy
  • go
  • debugging
  • postmortem
  • ops

2026-11-15

ClickHouse async inserts were not batching, and an HTTP/2 bug blocked the fix

Three inserts a second made nine new ClickHouse parts a second. A fixed async insert window fixed it, after an HTTP/2 bug and a ClickHouse error were gone.

  • clickhouse
  • rust
  • http2
  • performance
  • postmortem

2026-11-15

systemd quietly removed my container's swap limit

I capped Caddy with no swap and checked it in the kernel. Then a libc upgrade re-executed systemd, and systemd set the swap limit back to unlimited.

  • docker
  • systemd
  • linux
  • postmortem
  • ops

2026-11-13

A 4.5 GB browser tab, an htmx leak and a measurement that lied

A dashboard tab reached 4.5 GB. My first measurement blamed the page and was wrong. The real leak was a newline per poll, far too small to explain the tab.

  • htmx
  • javascript
  • debugging
  • performance
  • rust

2026-11-10

Two rejected series stalled all my metrics

Three alerts fired while production was fine. A rate limiter put client addresses in metric labels, and a retry on 429 turned two rejected series into a stall.

  • prometheus
  • observability
  • postmortem
  • caddy
  • ops

2026-11-06

Opsgenie end of life: 7 alternatives before April 2027

Opsgenie shuts off on 5 April 2027 and Atlassian deletes what you leave. Seven ways out, what each costs five people, and how to move without missing a page.

  • on-call
  • alternatives
  • incident-response

2026-11-04

Atlassian Statuspage pricing in 2026: what each plan costs

Statuspage costs $0 to $1,499 a month for a public page. Every plan and limit, where the bill jumps, and what the list price leaves out.

  • status-page
  • alternatives
  • monitoring

2026-11-01

Certbot renewal failed: how to diagnose and fix it

Trace failed Certbot renewals through logs, timers, HTTP and DNS challenges, saved plugin settings, and the certificate your public server actually serves.

  • ssl
  • tls
  • certificates
  • debugging
  • monitoring

2026-11-01

How to fix NET::ERR_CERT_COMMON_NAME_INVALID on your website

Find the hostname your certificate misses, check what the TLS endpoint serves, and fix SAN, wildcard, proxy and CDN mistakes without disabling verification.

  • ssl
  • tls
  • certificates
  • debugging
  • monitoring

2026-11-01

How to read a PEM certificate: names, issuer and expiry

Read PEM certificates locally: inspect SAN names, issuer and dates, compare SHA-256 fingerprints, and check what your server actually serves.

  • ssl
  • tls
  • certificates
  • openssl
  • debugging

2026-11-01

Unable to verify the first certificate in Node.js: find the chain problem

Trace Node.js certificate-chain errors, inspect the intermediates your server sends, and configure private CA trust while keeping TLS verification enabled.

  • ssl
  • tls
  • nodejs
  • certificates
  • debugging

2026-10-30

Best free status page tools in 2026

Ten free status page tools compared from their own pricing pages: free monitors, check interval, custom domain and first paid price, each source dated.

  • status-page
  • alternatives
  • monitoring

2026-10-30

Best uptime monitoring for a small SaaS in 2026

Eight uptime monitors with status pages, compared for a small SaaS: free tiers, check intervals, custom domains and first paid prices, sourced and dated.

  • monitoring
  • status-page
  • alternatives

2026-10-30

9 best Better Stack alternatives in 2026, free and paid

Nine Better Stack alternatives sorted by the part they replace: uptime checks, status pages, on-call or logs. Prices from each pricing page, dated.

  • monitoring
  • alternatives
  • status-page

2026-10-30

8 best UptimeRobot alternatives in 2026, free and paid

Eight UptimeRobot alternatives compared on faster free checks, status page subscribers and your own domain. Prices from each pricing page, dated.

  • monitoring
  • alternatives
  • status-page

2026-10-15

How to announce scheduled maintenance on your status page

Announce planned work before it starts, show it while it runs, and keep it in your status page history afterwards. Lead times and example messages.

  • status-page
  • incidents
  • reliability
  • monitoring

2026-10-12

How to get alerted when someone changes your DNS records

A changed DNS record can still resolve, so uptime checks stay green. Watch the exact record set and get an alert when a record is added, removed or changed.

  • dns
  • monitoring
  • security
  • email

2026-10-06

Google says on-call needs eight engineers. You have three.

Google's SRE book says a 24/7 on-call rotation needs eight engineers. What a team of two or three can change instead: the pages, the hours and the ladder.

  • on-call
  • alerting
  • incident-response
  • reliability
  • monitoring

2026-10-04

WordPress missed schedule: why it happens and how to fix it

A post marked "Missed schedule" means WP-Cron did not run on time. Here are the four usual causes, and how to hear about it the next time it stops.

  • wordpress
  • wp-cron
  • cron
  • heartbeat
  • monitoring

2026-09-24

Best open-source status pages you can self-host (2026)

Eleven open-source status pages compared: which ones monitor for you, which let customers subscribe, what each takes to run, and which ones are abandoned.

  • open-source
  • self-hosted
  • status-page
  • incidents

2026-09-24

What is an SSL certificate chain? Leaf, intermediate, root

How a server certificate links to a trusted root through intermediates, why a missing intermediate breaks some clients and not others, and how to fix it.

  • ssl
  • tls
  • certificates
  • security
  • networking

2026-09-22

Uncle Bob's uml-viewer on Rust: 35 dependency cycles down to 3

I fed my Rust codebase to Uncle Bob's uml-viewer and let an AI agent fix what the red arrows showed. Six commits: 35 cycles to 3, 167 violations to 109.

  • ai-agents
  • rust
  • architecture
  • clean-architecture

2026-09-05

How to debug a redirect loop, one hop at a time

Trace ERR_TOO_MANY_REDIRECTS with curl and response headers. Find conflicting HTTPS, hostname, and login rules, then monitor the path after fixing it.

  • http
  • redirects
  • debugging
  • monitoring
  • cloudflare

2026-09-05

How to monitor SSL certificate expiry when auto-renewal fails

Check the certificate your server serves, catch failed renewals before expiry, and set up TLS alerts alongside HTTPS monitoring. Includes OpenSSL checks.

  • ssl
  • tls
  • certificates
  • monitoring
  • reliability

2026-09-05

Uptime Kuma API: what exists, and what to use instead

Uptime Kuma ships no REST API for managing monitors. Here is every HTTP route it exposes, what the Socket.IO interface really is, and where the wrappers stop.

  • uptime kuma
  • api
  • monitoring as code
  • open-source

2026-09-05

Why DNS returns different IP addresses on different resolvers

Compare DNS answers from Cloudflare and Google, read TTLs, and query authoritative nameservers to separate normal CDN routing from a broken DNS change.

  • dns
  • debugging
  • monitoring
  • cdn

2026-09-02

ICMP vs TCP vs UDP: the difference, explained for developers

What ICMP, TCP and UDP each prove about a host, why a closed port answers differently on each, and where ping and port checks mislead. With RFC links.

  • networking
  • icmp
  • tcp
  • udp
  • ping
  • monitoring

2026-08-26

Your cron job can fail for months and nothing will tell you

A scheduled job that stops running produces no error and no alert. GitLab's nightly backup failed for months before anyone noticed. Why, with sources.

  • cron
  • heartbeat
  • monitoring
  • backups
  • reliability

2026-08-18

How to monitor an MCP server

An MCP server can return 200 on every request and still be useless to an agent. Probe the handshake instead, with one HTTP POST and one assertion.

  • mcp
  • monitoring
  • agents
  • uptime

2026-08-06

11 best Statuspage alternatives in 2026, free and paid

Eleven real Atlassian Statuspage alternatives compared honestly: which include the monitoring it leaves out, and which stop charging you per subscriber.

  • status-page
  • alternatives
  • monitoring

2026-08-06

What is an uptime SLA? Percentages, credits and fine print

An uptime SLA is a contract promise with a penalty, not a measurement. What each percentage allows, what a credit really pays, and the clauses that decide it.

  • sla
  • uptime
  • reliability
  • downtime

2026-08-05

Scheduling in Rust: one heap, not a timer per task

Every check my uptime monitor runs is scheduled from one BinaryHeap on one task. No timers, no locks, no drift. Five things that fell out of building it.

  • rust
  • tokio
  • scheduling
  • performance

2026-08-02

How much downtime is 99.9% uptime? 43 minutes a month

99.9% uptime allows 43 minutes of downtime a month and 8 hours 46 minutes a year. What uses that time, and why three nines is the normal SaaS promise.

  • sla
  • uptime
  • slo
  • downtime

2026-08-02

How much downtime is 99.95% uptime? 21.6 minutes a month

99.95% uptime in minutes and hours: the downtime it allows per day, week, month and year, and why it is the internal target behind a 99.9% SLA.

  • sla
  • uptime
  • slo
  • downtime

2026-08-02

How much downtime is 99.99% uptime? 4.3 minutes a month

99.99% uptime allows 4 minutes 19 seconds of downtime a month and 52 minutes 34 seconds a year. Four nines is a design decision, not a question of effort.

  • sla
  • uptime
  • slo
  • downtime

2026-08-01

Your login page returns 200. Nobody can sign in.

A 200 proves the page was sent. It proves nothing about the form. Four ways a login breaks while every check stays green, and how to tell them apart.

  • monitoring
  • login
  • trust
  • uptime

2026-08-01

Your E2E login test never runs in production

Your end to end login test runs in CI, against staging, at merge time. The faults that lock real customers out cannot happen there.

  • qa
  • e2e
  • synthetic-monitoring
  • testing
  • playwright

2026-07-25

How I stop one bad probe from waking you at 3 a.m.

One monitoring location with a bad network day is the classic false alert. Here is how repeat checks plus a region vote keep that alert away from your phone.

  • monitoring
  • multi-region
  • false-alerts
  • alerting

2026-07-24

Your domain can expire while your uptime monitor stays green

An expired domain does not go dark. It serves a parking page that returns 200 OK, so uptime monitors report it healthy while the site is gone.

  • domain expiry
  • monitoring
  • uptime
  • dns

2026-07-23

How I mapped my codebase for humans and AI agents

I turned 146,000 lines of Rust into three files: for me, for the next AI agent, and an interactive map. The method, the prompts, and the AI's wrong counts.

  • ai-agents
  • llm
  • codebase
  • documentation

2026-07-22

Do I need an uptime monitor? Count what downtime costs you

An uptime monitor is cheap. One outage you find out about from a customer is not. How to price your own downtime, and what to watch besides the homepage.

  • uptime
  • monitoring
  • downtime
  • reliability

2026-07-20

How to write incident status updates that build trust

A guide to writing status page updates during an outage: the four stages, how often to post, and what to say at each one so customers keep trusting you.

  • incidents
  • status-page
  • on-call
  • postmortem
  • reliability

2026-07-18

Why I chose Rust over Go for an uptime monitor

Go is the usual pick for a service like this. I chose Rust for one reason: a monitor sells clean timing, and its own runtime must not add random delay.

  • rust
  • go
  • monitoring
  • devops

2026-07-17

Is 98% uptime good? It allows 7.3 days of downtime a year

98% uptime means 28.8 minutes down a day. The downtime table from 90% to 99.999%, what each level costs you, and when 98% is enough.

  • sla
  • uptime
  • reliability
  • downtime

2026-07-15

Why a status page shows 100% uptime after an outage

A status page can show 100% uptime after a real outage when its bar comes from published incidents, not checks. How to tell, and a quick test.

  • status-page
  • monitoring
  • trust
  • uptime
  • postmortem
  • sre
  • devops

2026-07-14

8 best Pingdom alternatives in 2026, free and paid

Eight real Pingdom alternatives compared honestly: which replaces uptime checks, which replaces synthetics and RUM, and which ends usage-based pricing.

  • monitoring
  • status-page
  • alternatives

2026-07-13

Error budgets, explained: SLOs, burn rate, when to stop shipping

What an error budget is, the simple formula, how burn rate becomes an alert, and the rule that makes it work. With a free calculator.

  • sre
  • slo
  • reliability
  • monitoring
  • on-call

2026-07-09

ClickHouse disk full: system tables ate 12 GB (the fix)

Disk at 100%, Postgres down with 'no space left on device'. Real data: 20 MB; ClickHouse had logged 12 GB about itself. The config that fixes it, for good.

  • clickhouse
  • postgres
  • postmortem
  • observability
  • performance
  • ops

2026-07-08

Postgres vs ClickHouse is the wrong question. I use both.

My uptime monitor runs Postgres and ClickHouse side by side. Which data goes where comes down to one question, and four database tricks fall out of it.

  • postgres
  • clickhouse
  • databases
  • rust
  • monitoring

2026-07-07

Why I dropped reqwest for hyper in my Rust uptime prober

I swapped reqwest for raw hyper to probe uptime in Rust. Four things it taught me, each one a detail a normal HTTP client hides from you.

  • rust
  • hyper
  • monitoring
  • devops

2026-07-04

Email bombing through uptime monitoring pages

Attackers flood an inbox to bury one real alert, like bank fraud. How subscription bombing works, and how to build a monitor that can't be abused.

  • security
  • email
  • monitoring
  • abuse

2026-06-30

Building an uptime monitor in Rust: one binary, two databases

The Rust build behind Uptimepage: a custom hyper client doing ~130K checks a second on one core, a single-heap scheduler, and ClickHouse rollups.

  • rust
  • clickhouse
  • monitoring
  • devops

2026-06-20

Best open-source, self-hosted uptime monitors (2026)

A fair look at the open-source, self-hostable tools for watching sites and APIs in 2026: what each is good at, where it stops, and how to pick one.

  • open-source
  • self-hosted
  • monitoring
  • status-page

2026-06-18

The night Emma stopped opening dashboards

A 2 a.m. alert, and an on-call engineer who never opened a tab. A short story about asking your monitoring questions instead of clicking through it.

  • mcp
  • ai
  • monitoring
  • on-call

2026-06-18

The mystery of the "down" website

"The site is down!" But what does "down" really mean? A detective story through the seven network layers, and the one quiet failure no alarm caught.

  • monitoring
  • networking
  • osi
  • tls
  • dns

2026-06-16

Monitoring as code: put your uptime checks in a pull request

Click-created monitors rot: nobody recalls why a threshold is set, and the reasoning leaves with its author. Terraform fixes that, and bites back in places.

  • terraform
  • infrastructure-as-code
  • monitoring
  • devops

2026-06-03

Your monitors can talk to an AI, with your permission

Uptimepage now speaks MCP, so an LLM can answer "what's broken and since when?" in plain language, plus what we did to stop it from wrecking things.

  • mcp
  • ai
  • monitoring
  • security
  • api

2026-05-20

Why your uptime monitor should be boring

A monitor that surprises you is doing the wrong job. What changes when you treat the watchdog like a smoke detector: cheap, dumb, unmissable when it matters.

  • monitoring
  • incidents
  • status-pages
  • on-call