Backparq

Table-level PostgreSQL Backup to Parquet on S3.

Backparq exports PostgreSQL tables as Parquet files to S3, enabling fast restores, columnar analytics, and long-term retention at a fraction of the cost of keeping data in your database.

Why Backparq?

Problem	How Backparq Solves It
Database growing too large	Archive old data to S3, optionally delete from PostgreSQL
Need table-level backups	Full or incremental snapshots per table, not whole database
Want to query historical data	Parquet format works with DuckDB, Athena, Spark, Pandas
Slow restores from pg_dump	Restore specific tables and date ranges in minutes
No visibility into backups	SHA256 checksums, verification commands, progress bars

Backparq vs WAL-G vs pg_dump

	pg_dump	WAL-G	Backparq
Backup Scope	Full database	Full database (WAL + base)	Per-table, selectable
Backup Format	SQL / custom binary	WAL segments + base backup	Parquet (columnar)
Incremental	No	Yes (WAL streaming)	Yes (by date ranges)
Restore Granularity	Entire DB or single table	Point-in-time (whole DB)	Per-table + date range
Query Backups Directly	No	No	Yes (DuckDB/Athena/Spark)
Storage Efficiency	Low (uncompressed SQL)	Medium	High (columnar + zstd)
S3 Native	Via pipe/script	Yes	Yes
Use Case	Migrations, full dumps	Continuous DR, PITR	Table archival, analytics-ready backups

When to Use Each

• pg_dump: One-off migrations, development snapshots, schema exports
• WAL-G: Continuous disaster recovery, point-in-time recovery to any second
• Backparq: Table-level backups, cold data archival, analytics on historical data

Best practice: Use WAL-G for continuous DR + Backparq for table-level archival and analytics.

Features

• Two Modes: Backup (full snapshots) or Offload (archive + delete old data)
• Table Selection: Choose specific tables, not the entire database
• Date-Range Restore: Restore only the data you need, not everything
• Columnar Format: Query backups directly with DuckDB, Athena, or Spark
• Parallel Processing: Concurrent table and chunk processing
• Streaming Export: Constant memory usage regardless of table size
• Safety First: SHA256 checksums verified before any deletion
• Encryption: S3 SSE-S3, SSE-KMS, or client-side Parquet encryption

Installation

# Recommended: use uv
curl -LsSf https://astral.sh/uv/install.sh | sh
uv pip install backparq

# Or pip
pip install backparq

# With DuckDB for querying archives
uv pip install backparq[query]

Quick Start

# 1. Generate config
backparq init

# 2. Test connections
backparq test --config backparq.yaml

# 3. Run backup (preview)
backparq archive --config backparq.yaml --dry-run

# 4. Run backup
backparq archive --config backparq.yaml --stats

Usage

Backup Tables

Create full snapshots of specific tables:

backparq archive --config backup.yaml --stats

# backup.yaml
archive:
  mode: backup           # Full snapshot, no deletion
  tables:
    - public.users
    - public.orders
    - public.transactions

Archive + Delete Old Data

Move data older than 90 days to S3 and reclaim database space:

backparq archive --config offload.yaml --stats

# offload.yaml
archive:
  mode: offload
  cutoff: "-90d"         # Archive data older than 90 days
  perform_delete: true   # Delete after verified S3 upload
  tables:
    - public.events
    - public.audit_logs

Restore

# Restore specific date range
backparq restore --config restore.yaml \
  --start 2024-01-01 --end 2024-03-01

# Restore from a specific backup snapshot
backparq restore --config restore.yaml \
  --backup-id 2024-01-15_120000

# Update existing rows with archived data
backparq restore ... --conflict-mode upsert

Query Archives Directly

# Query with DuckDB (requires backparq[query])
backparq query --config backparq.yaml \
  --sql "SELECT COUNT(*) FROM public_orders WHERE created_at > '2024-01-01'"

Verify & Maintain

backparq check --config backparq.yaml   # List archives
backparq verify --config backparq.yaml  # Verify checksums
backparq prune --config backparq.yaml   # Delete old backups per retention

Configuration

database:
  host: localhost
  name: production_db
  user: postgres
  password: "${PG_PASSWORD}"

s3:
  bucket: my-backups
  prefix: postgres
  region: us-east-1
  sse: "AES256"  # Server-side encryption

archive:
  mode: backup   # or "offload"
  tables:
    - table: public.orders
      primary_key: order_id

See examples/reference.yaml for all options.

Example Configs

File	Use Case
examples/backup.yaml	Full table snapshots for DR
examples/offload.yaml	Archive old data + delete from DB
examples/restore.yaml	Restore with date ranges
examples/reference.yaml	All configuration options

Security

• Credentials: IAM roles, environment variables, or AWS credentials file
• Encryption at rest: S3 SSE-S3, SSE-KMS, or client-side Parquet encryption
• Encryption in transit: HTTPS to S3
• Checksums: SHA256 verification before any deletion

See SECURITY.md for vulnerability reporting.

Contributing

uv venv && source .venv/bin/activate
uv pip install -e ".[dev]"
uv run pytest

See CONTRIBUTING.md.

License

MIT