Skip to content

~/cold-path

cat README.md

Notes from the cold path

Long-form writing on databases, queues and the parts of production nobody demos. One post a month, with the code that goes with it — in Go, Rust and TypeScript.

posts
12
authors
4
reading
2h 10m
since
Aug 2025
latest14 min read
insert into jobs (queue, idem_key, payload)values ($1, $2, $3)on conflict (queue, idem_key) do nothing;

Exactly-once is a lie: an idempotent job queue on Postgres

We replaced a broker with one table, a unique index and SKIP LOCKED. What we gave up, what we got back, and the numbers after six months.

Deniz Aksoy @deniz

git log --since=2026-01-01

Recent posts

All 12 posts
/

7 posts

  1. Reading EXPLAIN (ANALYZE, BUFFERS) without guessing

    Shared hits, reads and the one line that tells you the planner was wrong.

    #postgres #observability11 min

  2. Our p99 was a DNS problem

    Incident review: 41 minutes of 2-second tails, a resolver with a 5-second timeout and ndots:5.

    #incidents #networking8 min

  3. Backpressure, explained with a kitchen sink

    Bounded queues, load shedding and why an unbounded buffer is just a slower outage.

    #queues #distributed9 min

  4. The cost of a retry: budgets, jitter and the thundering herd

    A 10% retry budget saved us more than any timeout tuning ever did.

    #reliability #distributed12 min

  5. Zero-downtime column renames in four deploys

    Expand, dual-write, backfill, contract — with the exact migrations we ran.

    #postgres #migrations7 min

  6. Structured logs you can actually grep

    One event per line, stable keys, and why we stopped logging stack traces at INFO.

    #observability6 min

  7. Feature flags are a distributed system

    Stale caches, partial rollouts and the flag that turned itself back on.

    #reliability #distributed10 min

finger @authors

Who writes here

  • Deniz Aksoy@deniz· 3 posts

    Staff engineer, payments infrastructure

    Runs the queue that moves money between ledgers. Has opinions about retries and keeps a list of every incident since 2019.

  • Mira Lindqvist@mira· 4 posts

    Database reliability engineer

    Reads query plans for fun. Spent four years keeping a 9 TB Postgres cluster boring.

  • Tomas Reiher@tomas· 3 posts

    SRE, network edge

    Owns the load balancers and, therefore, most of the blame. Writes the incident reviews on this blog.

  • Selin Varol@selin· 2 posts

    Platform engineer, developer tooling

    Builds the deploy pipeline and the flags service. Believes every log line should answer one question.

subscribe --monthly

One post a month. No tracking pixels.

The full post arrives as plain text with the code inline. 3,180 engineers read it this way.

One email a month. Unsubscribe from any issue.

rss.xmlatom.xmlfeed.json