8:57

The laptop opens.

The fan is already spinning.

It never stopped. Not since yesterday.

A Tuesday, honestly.

One developer, a few projects, one ordinary day. Told twice.

TerminaldozeFileEditView

Tue9:02

web · pnpm devapi · poetry run…worker · go rundocker composekafka.shzsh (??)billingdev env

Last login: Mon Nov  2 18:52:11 on ttys004
$

$ docker compose up postgres
[+] Running 1/0
 ✔ Container acme-postgres-1  Created
Error response from daemon: Ports are not available:
exposing port TCP 0.0.0.0:5432 -> 0.0.0.0:0: listen tcp
0.0.0.0:5432: bind: address already in use

$ lsof -i :5432
COMMAND   PID USER  NAME
postgres 4121 you   localhost:postgresql (LISTEN)
$ ps -o lstart= -p 4121
Tue Oct 27 10:14:03 2026
$ # last week. the client project.

$ go run ./services/orders
failed to connect to `user=orders database=orders`:
  127.0.0.1:5432: FATAL: role "orders" does not
  exist (SQLSTATE 28000)
exit status 1

$ cd ~/code/billing && git pull -q
$ make dev
make: *** No rule to make target 'dev'.  Stop.
$ cat .env.example | wc -l
       4

$ git log --oneline -4 -- services/
a41c9e0 add notifications service (TypeScript)
7be20d1 move worker to Python
3c0aa18 billing runs on the dev cluster now
19ff3a2 bump node to 22

$ kubectl port-forward svc/billing-api 8080:80
Forwarding from 127.0.0.1:8080 -> 80
$ kubectl port-forward svc/postgres 5432:5432
Unable to listen on port 5432: Listeners failed
to create with the following errors: [unable to
create listener: Error listen tcp4 127.0.0.1:5432:
bind: address already in use]
$ kubectl port-forward svc/postgres 5433:5432
Forwarding from 127.0.0.1:5433 -> 5432

$ # save. sync. rebuild. somewhere else.
syncing 214 files → dev-you … done
rebuilding billing-api (remote) …… done
waiting for debugger on 127.0.0.1:9229 …
timed out
$ # logs are in the other tool. in the other tab.

$ kubectl get pods
Unable to connect to the server: dial tcp
10.0.12.4:443: i/o timeout
$ kubectl get pods
Unable to connect to the server: dial tcp
10.0.12.4:443: i/o timeout

$ docker ps -q | wc -l
      23
$ ps -eo etime,comm | grep node
 1-02:14:09 node
 1-01:58:40 node

README.md

## Getting started (2023)~~1. make dev2. open localhost:3000~~3. if that fails, ask Priya

Assistant

orders can't reach the database. fix it?

Make sure Postgres is running on localhost:5432 and DATABASE_URL is set in your .env file.

it is. it's a different project's postgres

Apologies for the confusion! Try port 5433.

that's the other other one

Homelarder1billing⌘K Search or do

dozestarts a service when something connects, and puts it to sleep when nothing does · 3 stacks · 1 up

Open folder… ⌘O

StackRunningLast

larder~/code/larder · 14 declared

5 awake · 7 waking · 2 asleep212 MB · 6 connections

13 awake · 1 dozing536 MB · 21 connections

stoppedlast time: everything

starting…up 5hup 9h ago

billing~/code/billing · 5 declared

stoppedlast time: everything

2 awake · 2 waking · 1 asleep70 MB · 1 connection

5 asleepno connections

up 7 months agostarting…up 4h

client-site~/work/client-site · 3 declared

stoppedlast time: everything

up 7 days ago

Drop a project folder anywhere in this window, or press ⌘O, to add it.

Machinenamesnetwork daemonDoctor

larder~/code/larder · up 1mTilesListMapGridUp

490 MB489 MBresidentstorefront 61 · recommendations 58 · fulfilment 52 · +10 more

0.01coresnotifications 0.9% · the rest quiet

12awake1 dozing · 1 crashing

1920connectionstasks ✓ 9 done · 1 not run

Who calls whom left to right, the way a request travelsNobody calls these they work from eventsShared by the stack one line each would be 17 lines; the chips on each card say who uses what

storefrontts61 MB

no shared services✓

gatewaygo46 MBport taken

redis✓

identitygo42 MB

postgreskafka✓

fulfilmentpy52 MB

kafkaaws✓

recommendationspy58 MB

postgreskafkaredis✓

ordersgo47 MBcrashing

postgreskafka✓

cataloggo47 MB

postgresaws✓

paymentsgo47 MB

postgreskafkaredis✓

notificationsts45 MB

kafkamailpit✓

mailpit24 MB

brought in by youused by 1 · notifications

redis5 MB

valkey 9used by 3

kafka24 MB

kafka 4used by 6

postgres13 MB

postgres 18used by 5

aws25 MB

awsused by 2

billing~/code/billing · up 6m~/code/billing · up 1hTilesListMapGridUp

155 MBresidentportal 58 · invoices 44 · notifications 41 · postgres 12

0.01coresthe rest quiet

4awake1 asleep

3connectionstasks ✓ 3 done

Who calls whom left to right, the way a request travelsShared by the stack the chips on each card say who uses what

portalts58 MB

no shared services✓

invoicests44 MB

postgres✓

notificationsts41 MB

mailpit✓

postgres12 MB

postgres 18used by 1 · invoices

mailpitasleep

brought in by youused by 1 · notifications

gatewaygo run ./apps/gateway · port taken

OverviewLogs

Port 8000 is taken · gateway cannot listen while python3, pid 5387 has it

Stop pid 5387 and start gatewaydoze sends it a polite TERM. It is not one of doze's; doze looks again that it still holds the port first.

Stop python3

Use another portChange port = 8000 in doze.hcl; gateway's name keeps working on whatever you pick.

Edit doze.hcl

doze never stops a process it did not start without asking, which is why this is a question and not a thing that happened.

ordersgo run ./services/orders · crashing · try 2 of 5 · next in 8 sStop trying sTry now r

OverviewLogs

DebugInfo+Warn+ErrorsThis runAll 2 runsRelativeClock

run 2 · started 11:31:04 · exit 1 after 0.6 s · 3 lines

-2sINForders starting · connecting to postgres as orders

-2sERRfailed to connect to `user=orders database=postgres`: password authentication failed for user "orders" (SQLSTATE 28P01)

-2sdoze: orders exited 1 after 0.6 s · try 2 of 5 · next in 8 s

dozeorders crashed in lardertry 2 of 5 · password authentication failed for user "orders"

Show

Priyacan you look at billing?

# platform

status Dev environments degraded. Investigating. No ETA.

is there a workaround? billing is down for us

on-call Sorry, we're on a production incident first. Will update.

same here, can't run anything

Reconnecting to dev-you…

Assistant

orders can't reach the database. fix it?

read doze status --json · doze logs orders

orders is crashing (try 2 of 5): password authentication failed. .env.local overrides the DATABASE_URL doze gives it. Remove that line?

invoices.ts — billing

40  const total = lines.reduce(sum, 0);
41  const tax = taxFor(customer, total);
42  return { total, tax, due: total + tax };
43}

Paused on breakpoint · invoices.ts:42

↺ 8:57

3000?3000?3001 = the other web30018000 or 8001?8001?5432 (theirs!)5432!5433 = ours??5433?6379 shared?6379?5173 vite somewhere5173?8080 ???8080?

Fanoff

Tuesday8:57

Laptopcool

Act oneMorning

9:02

Coffee. Lid up.

The tabs come back, one at a time.

web · pnpm dev api · poetry run… worker · go run docker compose kafka.sh zsh (??)

Six of them. You know what five are doing.

Nobody knows what zsh (??) is doing. You leave it alone.

9:20

You start the database. The first error of the day:

Address already in use.

Whose address? You ask the machine.

It's last week's client project. Its Postgres is still up, holding 5432, with the wrong data in it.

You open the README to see how this one is meant to start.

It says Getting started (2023). It was right, once.

You start writing ports down where you can see them.

3000? 3001? 8000 or 8001? Your head is full, so the bezel gets the rest.

Act twoNoon

11:30

Orders can't reach its database. You ask the AI.

It answers before you've finished asking. Quick. Confident.

Wrong.

“Apologies for the confusion! Try 5433.”

That's the other other one.

It read every file. It can't see your machine.

Act threeAfternoon

14:00

A message. can you look at billing?

A repo you haven't opened since spring. You wrote half of it.

make dev. No rule to make target.

There's a TypeScript service you've never seen. And a commit that says billing runs on the dev cluster now.

Nothing is wrong, exactly. It just isn't familiar any more.

How we got here

15:30

Last year local development got too hard, so it moved. Every team got a dev environment in the cloud, run by the platform team.

Which means tunnels. One port-forward per thing you need, and 5432 is still taken on this side.

Every save is a sync and a rebuild, somewhere else.

The debugger waits for the tunnel. The tunnel waits for the debugger. Timed out.

The logs are in another tool, in another tab.

Then #platform: Dev environments degraded. Investigating. No ETA.

Nobody on your team can run anything.

The platform team is on a production incident first, as they should be. Half of them are asleep in another timezone.

You sit, and wait.

You needed a platform team just to write code.

And on your desk, the most powerful computer you've ever owned sits there, waiting for a cluster to come back.

Local development didn't get fixed. It got moved.

Act fourEvening

18:40

Done. Before you go, you look.

23 containers still up. Two node processes since yesterday.

You close the lid.

None of it stops. Nothing stops on its own, so nothing stops.

The bag is warm by the time you're home.

You didn't design this.
It grew.

Rewind.

Same Tuesday. One app.

Act fiveThe same day, with doze

9:02

Lid up. The fan is off. It was off all night.

You open doze. Every project on this Mac is in one list, each with a switch. Last week's client project is stopped. It holds nothing.

You flip Larder on. It starts what ran last time, each service after the ones it needs.

Larder is our made-up grocery example: nine apps, four modules and a mail catcher. Or type doze up in a terminal; the app shows the same stack.