The Far Archive · columnar warehouse

A signal from the dead sector

First lesson: Вход в Дальний архив

0/27
lessons done
0· +10/lesson

Interactive course · BigQuery

You already write SQL.
BigQuery works differently.

Not another "SQL from scratch" course but a walk through the differences: arrays and structs right in a cell, windows with QUALIFY, day shards, partitions instead of indexes, and a bill for the bytes you read. Queries run inside the lesson — no Google Cloud account, no billing.

Start freeThe first chapter is free — 4 lessons
  • queries inside the lesson
  • no Google Cloud account
  • an exam and a certificate at the end
googlesql · bigquery Run
-- a resident’s tags live in the cell itself
SELECT name, ARRAY_LENGTH(tags)
FROM fa_users
Артём     3
Кир       1
Лада      0
the bill for a querySELECT * → 2 columns
Chapter 5500 ₽ → 0,5 ₽

The same query, a different bill

One line of DDL cuts the scan a thousandfold. You will learn to see it before you run.

8
chapters
27
lessons
live
queries inside the lesson
no GCP
no account, no billing

The offer

What is different from Postgres here

Six habits you will have to change. Each gets a lesson with live queries — not a retelling of the docs, but a walk-through on the archive data.

List 39 columns to hide one
SELECT * EXCEPT(col)
And REPLACE — swap one column for an expression
A separate table for a list of tags
An ARRAY right in the cell
Plus STRUCT: a row is no longer flat
A subquery to keep the best row per group
QUALIFY on top of a window function
Filtering by a window without wrapping it
An index will speed this up
No indexes — partitions and clustering
And a bill for the bytes you read
LIMIT 10 means cheap
LIMIT does not lower the bill
You pay for columns, not for output rows
A table per day is painful
`logs_*` and _TABLE_SUFFIX
One query across every shard

Three scenes

What this actually looks like

The three differences that break the habit hardest — drawn rather than described: a row is no longer flat, tables come one per day, and a filter now costs money.

Chapter 3

A cell holds a list

An array and a struct live right inside the row. No link table and no JOIN for three tags — but you now need UNNEST, and it can silently drop rows.

Chapter 6

One table per day

Journals are sharded by date right in the table name. A wildcard reads them in one query, and the _TABLE_SUFFIX filter decides how many days you pay for.

Chapter 5

There are no indexes

Partitions and clustering take their place. A well-written WHERE does not speed the query up — it physically cuts out data you then never pay for.

How it works

A course where the queries actually run

The query — right on the page

Every live cell executes against a real BigQuery-dialect engine on the archive data. No Google Cloud account, no billing, nothing to install.

The delta, not a retelling

No lesson re-explains what you already know from Postgres. Each answers one question: what works differently here, and what it will cost you.

The bill as part of the syntax

In BigQuery the price of a query is a property of it, alongside the result. The course teaches reading that price before you run, not from the invoice.

The Far Archive · the last Earth colony

A signal that was not noise

A weak signal came out of the dead sector — and turned out to be a warehouse: orders, payments, terminal logs and telemetry of a colony that kept its journal to the last day. The dialect is foreign, the station’s power counter is finite, and QUERY, your companion, does not know this language either and learns it alongside you.

  • 02four time types instead of one
  • 03the row that stopped being flat
  • 05a 500 ₽ bill for one forgotten line
  • 06seven hundred tables in one query
  • 08the last session: power for one query

The route

Three steps of the transfer, then the exam

The program is built like a transfer: first the different syntax and its price, then the depth of the dialect, then scale — and the case closes with queries.

I

The transfer

The new archive's syntax, money and error language

II

Depth

Arrays, structs, windows — the reason you came

III

Scale

Money, partitions, shards and writing data

IV

The exam

The case closes with queries, not with retelling

The program

Eight chapters of the Far Archive

I

The transfer

2 chapters · 9 lessons · 1 h 38 min

The new archive's syntax, money and error language

Chapter 1free

Другой диалект

You will learn to address tables, read the price of a query before running it, and speak the language of BigQuery errors.

SELECT * EXCEPTpay per byteSAFE mode
4 lessons · 42 min
Chapter 2 PRO

Типы, строки и время

You will sort out a type system with no SERIAL, money in NUMERIC, re2 regexes, and four time types instead of one.

NUMERICre2DATE / DATETIME / TIMESTAMP
5 lessons · 56 min
II

Depth

2 chapters · 9 lessons · 1 h 52 min

Arrays, structs, windows — the reason you came

Chapter 3 PRO

Массивы и структуры

You will stop assuming a row is flat: lists and records live in the cell, and UNNEST expands them without losing rows.

ARRAYSTRUCTUNNEST
5 lessons · 1 h 4 minartefact: a GA4 event breakdown
Chapter 4 PRO

Агрегаты и окна по-новому

You will write aggregates and windows idiomatically: COUNTIF over CASE, QUALIFY over a subquery, an honest window frame.

COUNTIFQUALIFYROLLUP / PIVOT
4 lessons · 48 min
III

Scale

3 chapters · 8 lessons · 1 h 32 min

Money, partitions, shards and writing data

Chapter 5 PRO

Деньги и масштаб

You will design a table so the same query reads a thousand times less — and verify that before you run it.

PARTITION BYCLUSTER BYdry run
3 lessons · 30 minartefact: a cheap-query checklist
Chapter 6 PRO

Шарды, wildcard и GA4

You will read seven hundred daily tables in one query, build an end-to-end GA4 report, and pull data out of schemaless JSON.

wildcard_TABLE_SUFFIXJSON
3 lessons · 38 minartefact: a daily station dashboard
Chapter 7 PRO

Меняем данные и автоматизируем

You will write to the warehouse without OLTP reflexes: batches instead of row-by-row edits, MERGE instead of ON CONFLICT, scripts instead of plpgsql.

MERGEstagingscripting
2 lessons · 24 min
IV

The exam

1 chapter · 1 lesson · 25 min

The case closes with queries, not with retelling

Chapter 8 PRO

Экзамен Дальнего архива

You will gather every delta into one migrant cheat sheet and close the case with eight tasks that cannot be copied.

the delta mapHARD / EXPERTcertificate
1 lesson · 25 minartefact: an archivist's licence

What you keep

More than chapters you have read

Idioms, not recipes

QUALIFY instead of a subquery, UNNEST without losing rows, MERGE instead of ON CONFLICT — what separates a migrant from someone at home in the dialect.

The delta map

The final lesson gathers every difference into one table: Postgres habit on the left, BigQuery idiom on the right. People keep it open for months.

The exam and the certificate

Eight HARD and EXPERT tasks, auto-graded on live data: copying will not pass. Six of eight to clear it — and a personal certificate.

The archive is open. The query is yours.

The first chapter is free and needs nothing but a browser: no Google Cloud account, no billing, no install. Open the lesson and run queries against the archive right on the page.

The average lesson takes 12 minutes