Essays

43 entries

· 3 min read

Y2K, reconsidered as a success

The disaster that did not happen is remembered as a panic that was wrong. The record says otherwise, and the difference matters for every deadline since.

#history#remediation#non-events

· 3 min read

On the colour of the bikeshed, sixty-nine years on

Parkinson's law of triviality is from 1957; the email that renamed it for software is from 1999. Both are quoted weekly and obeyed nowhere. On attention, consequence, and a budget for reviews.

#history#reviews#attention

· 2 min read

The bus factor has a literature

Knowledge concentration can be measured from the repository itself, and researchers have been doing it for a decade. On the number nobody computes until the resignation.

#knowledge#teams#risk

· 1 min read

On the monthly availability report

GitHub's July report lists eight incidents, and seven of them were changes and expiries rather than code. A note on the institution of publishing your own ledger.

#reliability#incidents#institutions

· 1 min read

Ariane 5, thirty years on

The most expensive integer overflow in history is now three decades old. Its lesson was never about the cast.

#failures#reuse#history

· 1 min read

On finally reading the mainframe

AI made forty-year-old COBOL cheap to read at last. Comprehension and migration remain different problems, and the industry keeps buying one as the other.

#modernization#mainframe#ai

· 3 min read

The formats that outlive us

CSV never got better and never went away; Parquet is a teenager and load-bearing. An essay on file formats as the longest contracts in computing.

#formats#interfaces#longevity

· 3 min read

Ten hours without CI

GitHub Actions failed for ten hours and forty-two minutes on August 6. Read as an instrument, the outage measured every pipeline that runs through it.

#delivery#incidents#ci

· 4 min read

What the mainframe's survival teaches

Six decades after System/360, the machine every strategy deck kills is still clearing the payments. A long essay on why, and what its persistence actually indicts.

#mainframe#legacy#institutions

· 2 min read

On the engineering memoir

A quiet genre is doing documentation's best work: the practitioner explaining, in public, how their own system actually works. Notes toward taking it seriously.

#writing#documentation#learning

· 5 min read

The homelab argument

The strongest case for running servers in a closet has nothing to do with the servers. A long essay on small systems as professional education.

#operations#learning#self-hosted

· 2 min read

In praise of the quiet quarter

No incidents, no migrations, no launches. A short argument that the uneventful quarter is an engineering artifact, and that someone built it.

#operations#leadership

· 3 min read

Confidence is not a permission system

Automated systems have always failed by choosing the wrong rule with conviction. An essay on authority design, from Therac-25 to the agent on your estate.

#automation#agents#safety

· 3 min read

Best practice has an expiry date

Microservices were mandatory in 2016 and apologized for by 2023. An essay on why engineering wisdom decays, and how to date a practice before adopting it.

#practices#microservices#judgment

· 2 min read

Engineering ladders and load paths

Career ladders are load-bearing structures. What an organization promotes for is what its systems will eventually be made of.

#leadership#careers#incentives

· 2 min read

The audit that found the architecture

Sometimes the truest system diagram is produced by outsiders with a checklist. Why audits see what architects stopped seeing, and how to borrow the trick.

#architecture#audits#risk

· 2 min read

The meeting that replaced the design doc

Organizations drift from written design to verbal consensus one calendar invite at a time. What the written form was doing, and how to notice it going.

#leadership#design-docs#writing

· 2 min read

What a service boundary is for

Before arguing about microservices it helps to ask what a boundary buys you. The answer was written down in 1972 and has not changed.

#architecture#boundaries#microservices

· 2 min read

Enterprise AI is an integration problem

The model is the easy part. Enterprise AI succeeds or fails on the same unglamorous surfaces as every prior integration wave.

#enterprise-ai#integration#architecture

· 2 min read

The half-life of internal frameworks

Internal frameworks decay on a schedule set by staffing, not technology. On estimating the half-life before writing the framework.

#modernization#frameworks#maintenance

· 2 min read

Staff engineers and standing systems

The senior IC role was formalized around new construction. The harder and rarer craft is stewardship of systems that must not fall down.

#leadership#staff-engineers#careers

· 2 min read

The monolith was fine

Twice a decade the industry rediscovers that one well-made deployable can be the correct architecture. A history of the rediscovery cycle.

#architecture#monolith#microservices

· 2 min read

The rewrite that worked

Everyone can name a rewrite that failed. The instructive ones are the rewrites that shipped, because they share an unglamorous shape.

#modernization#rewrites#strangler-fig

· 2 min read

The platform team's second act

Every platform team is founded to pave roads and judged, three years later, as a toll booth. The second act decides which one it becomes.

#platform-teams#leadership#devops

· 3 min read

The upgrade nobody rushes

Postgres 18 shipped in September to the usual absence of panic. On release discipline as an institution, and what the calmest corner of infrastructure teaches.

#postgres#upgrades#operations

· 4 min read

Chesterton's fence has a maintenance schedule

The famous fence argues for caution before removal. It says nothing about leaving fences unexamined forever. An essay on institutionalizing the second half.

#legacy#decision-records#modernization

· 2 min read

On-call is an architecture signal

The pager is the most honest reviewer a system has. Reading on-call load as architectural feedback instead of an HR problem.

#architecture#on-call#operations

· 2 min read

Architecture review as an institution

Review boards drift toward theater or gatekeeping unless designed against both. Notes on making the review a working institution rather than a ritual.

#architecture#governance#process

· 1 min read

On reading other people's postmortems

The industry's best literature is filed under incident reports. A short note on reading them for structure rather than schadenfreude.

#incidents#learning

· 2 min read

DevOps, fifteen years on

The word conquered the industry and dissolved into it. An audit of what the movement actually changed, what it renamed, and what it quietly gave up.

#delivery#devops#history

· 2 min read

Feature-freeze archaeology

A change freeze is an accidental experiment. What the exemption list, filed in order of arrival, reveals about an organization's real architecture.

#change-control#organizations

· 4 min read

Two years after the license turn

HashiCorp relicensed in August 2023, Redis followed, and the forks are now old enough to evaluate. An essay on licenses as architectural dependencies.

#licensing#open-source#forks

· 2 min read

Release trains and their discontents

The release train solved a 2005 problem and still runs on schedule in organizations whose problem changed a decade ago. When the train helps, and when it is theater.

#delivery#release-management#process

· 2 min read

In praise of boring deploys

The most sophisticated thing a delivery pipeline can be is uninteresting. On boredom as an engineering achievement with a bill of materials.

#delivery#deploys#reliability

· 3 min read

The cost of a second datacenter

Multi-site resilience is bought in step functions, not increments. On the honest arithmetic of the second location, cloud or concrete.

#modernization#infrastructure#resilience

· 4 min read

The glue code is a system, and it needs an owner

Every enterprise runs on integration code nobody drew. An essay on middleware's three eras, and why the connections deserve the architecture review the boxes get.

#integration#middleware#architecture

· 2 min read

Conway's law is a budget, not a curse

The 1968 observation gets quoted as fate. Read closely, it is a constraint you can spend on purpose, and organizations do, mostly by accident.

#architecture#organizations#conway

· 2 min read

The deprecation nobody owns

Systems accumulate deprecated things that never finish dying. On why removal is the least-funded phase of the lifecycle, and what owning it looks like.

#modernization#deprecation#maintenance

· 1 min read

The interface is the asset

Amazon S3 turns nineteen this month. Its implementation has been rebuilt continuously; its API has barely moved. On what that asymmetry teaches.

#interfaces#apis#longevity

· 2 min read

Why migrations stall at eighty percent

The first eighty percent of a migration moves on rails. The last twenty is a different project with different economics, and it is rarely funded as one.

#modernization#migrations#planning