firedrill v0.4.0

firedrill 0.4.0

---

What's Changed

• firedrill 0.4.0: drill a live database, s3:// URLs, extension images, plain-SQL error fix by @MaXiMo000 in #5

Full Changelog: v0.3.0...v0.4.0

firedrill v0.3.0

firedrill 0.3.0: plain SQL dumps

---

What's Changed

• Plain SQL dumps (.sql, .sql.gz) with truncation caught (0.3.0) by @MaXiMo000 in #4

Full Changelog: v0.2.1...v0.3.0

firedrill v0.2.1

v0.2.1

---

What's Changed

• A materialized view can exist and answer nothing by @MaXiMo000 in #1
• A fetch nobody checked reported as ok by @MaXiMo000 in #2
• Phase 1 fixes: 0.2.1 by @MaXiMo000 in #3

New Contributors

• @MaXiMo000 made their first contribution in #1

Full Changelog: v0.2.0...v0.2.1

firedrill v0.2.0

A backup you have not restored is not a backup.

firedrill restores a PostgreSQL backup into a disposable, version-matched
container and proves the result is usable — schema, row counts, your own smoke
queries, sequences and collation — inside a stated recovery-time budget. It
fails your build the day a backup stops being restorable, not the day you need
it.

The rule the whole tool is built on: a verification that could not run never
reports as passing. A rung that could not run reports NOT RUN, never a tick.

• Directory and tar dumps (-Fd, -Ft). Directory format is what large
  databases are actually dumped in, because it is the one pg_restore can
  parallelise — so the tool aimed at hours-long restores could not previously
  read the format those people use. All three restorable formats carry a
  PGDMP header, so the major version still comes out of the artefact itself
  with no PostgreSQL client on the host.

• PostgreSQL 13 through 18, tested end to end. 13 and 14 were entirely
  broken before: datcollversion arrived in 15, and selecting a column that
  does not exist failed the whole snapshot query, so --write-reference
  wrote nothing and every structure check returned unreadable.

• Structure checks that cover more than tables. Views, materialised views,
  routines with identity arguments, triggers, RLS policies, whether row
  security is enabled, enum/domain/composite types, and NOT VALID
  constraints. Measured before this: a database that lost its view, function,
  trigger, RLS policy and enum type restored green with exit 0.

• Point-in-time recovery — firedrill pitr --base DIR --wal DIR --target TS
  — with the boundary assertion that makes it worth anything: the row written
  before the target must exist and the row written after must not. Either half
  alone is satisfiable by a restore that is simply wrong.

• A recovery target could inject postgresql.conf, including
  archive_command — command execution inside the container. Targets are now
  constrained to a timestamp's one legitimate shape.

• Sequences wired by a column DEFAULT rather than OWNED BY were never
  checked at all. pagila links all thirteen of its sequences that way, so the
  flagship SEQUENCE_BEHIND check examined nothing and reported "0 sequences".

• The step that reads the backup could not read the backup: pg_dump -Fd
  writes mode 700 and -Fc is commonly 600, owned by whoever ran it, and the
  container's postgres uid is a different one.

• The version was declared twice and the two drifted, so the wheel was 0.1.1
  and --version said 0.1.0.

pip install firedrill          # or firedrill[s3]

- uses: MaXiMo000/firedrill@v0
  with:
    config: firedrill.yml
    rto: 45m
    history: firedrill-history.json

Also ghcr.io/maximo000/firedrill:0.2.0, with build provenance attestation.

Docs: https://maximo000.github.io/firedrill/

---

Full Changelog: v0.1.1...v0.2.0

v0.1.1

Full Changelog: v0.1.0...v0.1.1

v0.1.0

Full Changelog: https://github.com/MaXiMo000/firedrill/commits/v0.1.0