Prerequisites

What you need installed before building or running dbsync.

Toolchain

Tool	Version
GHC	9.14.1
cabal-install	3.16.1.0
PostgreSQL	≥ 16 (18 recommended)

These GHC and cabal versions are the known-good pair CI builds against; other combinations may fail to solve against the pinned index-state. Install both via ghcup; the platform pages cover the per-OS specifics.

System libraries

Library	Required?	Used by
snappy	yes	Compression for the on-disk ledger snapshots.
pkg-config	yes	Build-time discovery of snappy and libpq.
libpq	yes	Bundled with PostgreSQL on most distros.
liburing	Linux only	The asynchronous block-I/O backend used by the LSM-tree dedup stores and the LedgerDB.

On macOS (and Windows, and any non-Linux target), liburing isn't available. dbsync detects this at build time and selects a synchronous fallback automatically — no manual flag required. See the macOS page for the details.

PostgreSQL

PostgreSQL 16 or newer is required — dbsync uses features in the COPY path and the LOGGED/UNLOGGED machinery that landed in 16. PostgreSQL 18 is recommended, and is the version CI and the installation guides below target.

You don't need a separate database account — the connection string is read from the profile JSON, and a local installation with peer authentication for the current OS user works for development. For a production deployment, give dbsync a dedicated user with CREATEDB the first time, then narrow the grants afterwards.

Read the tuning config first

scripts/postgres-tuning.conf in the repository documents the server-side settings that materially affect bulk-sync performance (wal_level = minimal, max_wal_size, shared_buffers, maintenance_work_mem, parallelism). Worth the few minutes before a multi-day mainnet sync.

Cardano node

A running cardano-node reachable over its Unix socket. Any n2c-supporting node works — see Cardano node setup for how to get one, including pre-built binaries, Docker, Nix, and source builds via the official Cardano docs.

Disk

Rough figures for a mainnet sync. The everything row is measured; the smaller profiles are approximations derived from the everything-profile per-table breakdown. Actual usage depends on the profile and on PostgreSQL's autovacuum behaviour:

Profile	Approximate pg_database_size at mainnet tip
minimal	~30 GB
utxo-only	~160 GB
spo	~185 GB
dapp	~265 GB
everything-no-ledger	~475 GB
everything	~480 GB

Plus ~220 GB for the cardano-node database (mostly the ImmutableDB chain history at mainnet tip). Allow comfortable headroom over these figures — the bulk-load phase needs working space for index builds and the LOGGED-flip table rewrite.

The on-disk ledger state (if ledger.enabled = true) lives under whatever directory you pass to --ledger-state-dir and runs ~11 GB at mainnet tip with the LSM backend.

Big single contributor: tx_cbor

The cbor extractor stores raw transaction CBOR bytes and accounts for ~218 GB on its own — roughly half of an everything-profile database. If your consumers don't need to re-serialise or replay transactions, disabling cbor in a custom profile derived from one of the bigger presets cuts the database nearly in half. See Preset configs for the per-extractor breakdown.