Brace — the lab

BRACE crash-test lab for postgres migrations
brace docs connecting
  1. webstatic
  2. apigo engine
  3. dbpostgres 16
no twin mounted warm · warming

writes / sec

0
baseline · window 60s
chamber nominal
impact 0.0s writes frozen

live traffic — right now · not the run below

checkout00 died
login00 died
add to cart00 died
casualties0writes killed

migration console

runs against a throwaway twin
run a multi-step migration (a sequence — optional)

Steps run in order on one twin, so a later step sees the schema the earlier ones produced. You get a verdict per step and the worst-of overall. When this is filled it takes precedence over the single statement above; hostile mode doesn't apply to a sequence.

rehearse your own schema (synthetic data — optional)

One CREATE TABLE only. Brace seeds a bounded sample (up to a few million rows) and projects the result onto the size you give. Your migration above runs against this table instead of orders.

Ctrl↵ to run

crash report

awaiting run

No run on record.

write freeze
writes died on the twin
verdict

Run a migration below. This card fills in the moment the twin stops moving.

bring your own data · self-host

This rig runs one synthetic dataset. Your data never touches it.

This public lab crashes every migration into the same shared, synthetic 10M-row orders table — so there is nowhere to upload your own schema here, on purpose. One lab cannot, and should not, clone a stranger's production database. To rehearse against your indexes, row widths and traffic, run Brace inside your own infrastructure, where the dump never leaves it.

Full walkthrough in the docs Overview on the landing page →

  1. 01
    Import once

    One import.yml stands up web · api · postgres on your own Zerops project.

  2. 02
    Restore your dump as the template

    Every throwaway twin is cloned from it. Turn synthetic seeding off with BRACE_SEED=off.

  3. 03
    Describe your write traffic

    BRACE_LOAD_LABELS sets the mix, so casualties are measured against your real workload.