Archival Node Setup Guide

Archival Node Setup Guide

Published
AuthorDevansh

An archival node is a full node that never forgets. Where a normal full node keeps recent state and prunes the rest, an archival node keeps every historical state back to the genesis block, so it can answer questions like what a balance was at a block from three years ago. That makes it essential for block explorers, indexers, and analytics, and overkill for almost everything else. This guide explains when you actually need one, how to plan the disk, which client to pick, and how to run the setup without paying for storage you will never use.

If you only need to know the difference between node types before deciding, read our full node vs archival node guide first. If you run nodes across several chains, our multi-chain archival setup post covers the per-chain differences. This post is the single-node setup: one chain, one machine, done properly.

Do you actually need an archival node?

Most teams that think they need an archival node do not. A wallet, a staking setup, a dapp frontend, and most bots only need current state, which a full node provides at a fraction of the disk and sync time. You need an archive node when your software asks for historical state: a block explorer showing old balances, an indexer rebuilding token transfers from genesis, an auditor replaying a transaction from two years ago, or a data pipeline that backfills history. If your queries only touch recent blocks, a full node is the right answer, and the money saved on disks buys better monitoring instead.

There is also a middle option worth knowing about. Some clients now support partial history, keeping a window of recent history instead of everything or nothing. That window is enough for many analytics jobs that only look back a few months. Check what your client offers on the day, because history handling changes with client releases, and a partial-history node can save terabytes if your use case allows it.

Disk planning: the number that decides everything

Disk size is the first decision because it decides the machine, the budget, and the sync time. The numbers depend heavily on the client you choose, and they grow every month, so treat any figure as a snapshot, not a constant. For Ethereum mainnet, ethereum.org's run-a-node page, measured in September 2026, lists a full archive at about 2.7 TB for Erigon and about 2 TB for Geth in its newer path-based archive mode, while keeping full trie history in Geth takes around 6.5 TB. A Geth full node, by comparison, needs about 1.8 TB. Other clients sit in the same broad range, and third-party guides from 2026 put legacy Geth archive storage far higher, at 12 TB or more, because the older hash-based mode stored far more data.

The lesson is not the exact number but the spread: client choice can change your disk bill several times over. Plan the disk from the client you will actually run, add at least 50 percent headroom for a year of growth, and use NVMe SSDs, not spinning disks. Archive syncs write constantly for days, and a hard drive will fall behind the chain tip and never catch up. Buy the disk once, buy it big, and check the client's hardware docs on the day you build, because a figure from six months ago is already wrong.

RAM and CPU matter too, just less

Archive nodes are disk-bound more than CPU-bound, but do not starve them. 16 GB of RAM is a floor for Ethereum archive work and 32 GB is the comfortable choice, with at least 4 fast CPU cores. More RAM helps the client cache state during sync, which shortens the initial build. The network link should be unmetered or generously capped, because the initial sync downloads terabytes and the node keeps serving peers afterwards. A slow or capped connection turns a two-week sync into a two-month one.

Pick the client before you pick the disk

For Ethereum, the practical choice in 2026 is between Erigon and Geth, with Reth and Nethermind as alternatives depending on your tooling. Erigon is the default for new archive deployments because its storage layout deduplicates intermediate state and keeps the disk footprint small. Geth is the most documented client and the one most teams already know, but its full archive mode is far larger unless you use the newer path-based mode, which trades away historical proofs for older blocks. Reth is newer and fast, and worth evaluating if your team is comfortable with a younger codebase. Whatever you choose, confirm the archive size and sync behaviour in that client's own docs on the day, and do not copy a disk plan from a guide written for a different client.

For other chains, the same rule applies with different numbers. Some chains have small archive footprints, others are larger than Ethereum. The client's documentation is the only honest source for the figure, and a chain's archive size from a year ago is not a plan. If the docs do not publish an archive size, ask in the chain's operator channel before buying hardware.

Setup steps, in order

Start with the server from our Ubuntu hardening approach: a supported LTS release, a non-root user, key-only SSH, and a firewall with only the ports you need open. Mount the NVMe storage as the data directory and confirm the filesystem has room for the full archive plus headroom before you start the sync. A sync that fills its disk halfway through wastes days, so check the free space first and set a disk alert at 80 percent.

Install the client the way its docs describe, either as a binary run under systemd or as a Docker container with the data directory on a volume. Enable archive mode explicitly: on Geth that means archive garbage-collection mode, on Erigon archive is the default sync behaviour. Then start the sync and leave it alone. An archive sync from genesis takes days to weeks depending on the client, the disk, and the network, and restarting it to check on progress only slows it down. Watch the logs for steady progress and peer counts, and compare your block height against a public explorer to confirm the gap is closing.

Once the node reaches the chain tip, test it the way your application will use it. Query a balance at an old block, replay an old transaction, and confirm the answers match a public explorer. Then enable the RPC methods your application needs and restrict them: an archive node answers expensive historical queries, so put it behind authentication or a firewall rule that only your services can reach. A public archive RPC with no rate limit is an invitation to burn your disk and bandwidth on strangers.

Cost and growth: the part that keeps costing

The disk is the main cost, and it never stops growing. Chain data grows every week, so a disk that fits today will be tight in a year. Budget for a larger disk or a replacement cycle, and monitor growth monthly so you can plan the upgrade before the disk fills. Back up the configuration, the client version, and any keys, but do not try to back up terabytes of chain data the usual way: chain data can be re-synced or restored from a snapshot, while a lost config or key cannot. Our validator backup guide explains what belongs in a backup and what does not, and the same logic applies here.

Pruning is the decision that defines an archive node, and the answer is simple: do not prune. Pruning deletes the history the node exists to keep. If disk pressure forces the question, the honest options are a bigger disk or a second node, not pruning the archive into a full node. Our pruning guide explains when pruning is safe, and an archive node is the case where it is not. For the full monthly picture, including bandwidth and monitoring, see our validator cost guide, and the same buckets apply to archive work.

An archival node is a commitment to disk, time, and patience, and it pays off only when your software truly needs history. Confirm the need, pick the client first, size the NVMe disk from that client's own docs with generous headroom, and test historical queries before you point production at it. If your team needs explorers, indexers, or historical data pipelines without running the hardware itself, see our blockchain node setup and cloud cost optimization pages, or start with a free infra audit and we will size the archive setup against your actual query load.

Disclaimer: This post is educational only, not financial advice. Figures are listing or project data as of 2026-11-08 and may change - always confirm version numbers, requirements, and dates in the official docs before acting on them.

Related Articles

DevOps & Infrastructure Services

Need hands-on help? We build and manage the infrastructure behind the concepts covered in this article.

DevOpsBy Assistant

How can we help?

Select a topic to get started: