BlogDeveloper

How to Find a Reliable Peer for Your Hyperliquid Node

Why Hyperliquid nodes lose their peer and fall behind, how to find peers with free slots on Flowscan, and when a reserved peer is worth paying for.

HydromancerTeam
How to Find a Reliable Peer for Your Hyperliquid Node

Your Hyperliquid node was in sync a minute ago. Now it's falling behind, the logs show it hunting for a new peer, and it happens again the next day around the US market open. If that sounds familiar, the problem is usually not your machine. It's the peer your node gets its blocks from.

Node operators raise this in the Hyperliquid Discord all the time: the node loses its connected peer, some peers can't keep up when markets get busy, and the node gets stuck retrying. This guide explains why that happens and how to find a peer that stays up.

The short answer: a reliable peer is one that has a free slot for you, has the bandwidth to keep up at peak times, and sits close to the validators, ideally in Tokyo. There are three ways to get one:

Whichever you pick, add backup peers so a dropped connection costs your node seconds rather than minutes. This guide assumes you already run a non-validator node. If you don't yet, start with how to set up a Hyperliquid non-validator node.

Why your Hyperliquid node keeps losing its peer

A Hyperliquid non-validator gets new blocks through gossip: peers pass the data along from node to node. It all starts at the validators. Each validator has two slots for non-validators connected straight to it, called sentry nodes, and the data spreads from those sentries to everyone else.

Your node takes its blocks from one parent peer at a time. If that peer drops you, your node stalls for 10 to 30 seconds while it finds and connects to a new one, and that's if it can find one with room. Three things make that more likely than it should be.

Most public peers are full

Every peer has a limited number of slots for nodes downstream of it. On 18 October 2026, Flowscan counted 50 sentries. 40 were full, 8 were offline and 2 had an open slot.

Hyperliquid's official API has a gossipRootIps request that lists recently available peers, and it's where many operators start. It returned 25 IPs that day. By Flowscan's count, 21 of them were full, 3 were offline and 1 had room. A list of peers that exist is not the same as a list of peers that will take you.

Serving peers costs money, and public peers owe you nothing

Every node a peer serves pulls data out of its server. On AWS, acting as the upstream peer for one non-validator is roughly 3 TB of outbound data a month, about $300 at AWS's rate of ~$0.104 per GB. A peer running for free has every reason to cap how many nodes it serves, and no obligation to keep yours connected.

Some peers can't keep up at peak times

When trading spikes, there is more data to pass along. A peer without enough bandwidth falls behind, and so does every node downstream of it. That's why operators see their nodes stall around the US open rather than at quiet times.

Check your own node first

Before you blame the peer, rule out the problems that make peers drop or ignore you:

1. Find a public peer with free slots on Flowscan

Flowscan's peer map is a free page that crawls the Hyperliquid gossip network and shows how data flows from the validators outward. It's the quickest way to see which peers exist, where they are and whether they have room.

The page splits peers into three groups:

Each sentry is labelled Open (it has a free slot), Full or Offline, with the number of peers currently connected to it. Filters narrow the view to Foundation or community validators, to one country, or to Connectable only to hide offline peers. Below the map, a network footprint shows which countries and hosting networks the crawled peers run on.

To pick public peers:

Then measure them from your own server. This script takes the official API's list of recent peers and times a plain TCP connection to each one's gossip port, fastest first. It needs pip install requests, and you can add IPs from Flowscan to the list.

code
import socket
import time

import requests

ips = requests.post("https://api.hyperliquid.xyz/info",
                    json={"type": "gossipRootIps"}, timeout=10).json()

def connect_ms(ip, port=4001, timeout=2):
    """Time a plain TCP connection to the peer's gossip port."""
    start = time.perf_counter()
    try:
        with socket.create_connection((ip, port), timeout=timeout):
            return (time.perf_counter() - start) * 1000
    except OSError:
        return None

results = [(ip, connect_ms(ip)) for ip in ips]
results.sort(key=lambda r: (r[1] is None, r[1] or 0))   # fastest first
for ip, ms in results:
    print(f"{ip:<16} {'no answer' if ms is None else f'{ms:.1f} ms'}")

A fast answer only shows the peer is up and close to you. It doesn't mean the peer has a free slot, so cross-check your shortlist against Flowscan before you add it to your config.

Good for: testing, personal projects and backup peers. Public peers come with no service guarantee, so don't make one the only peer behind a trading system or a paid product.

2. Get a reserved peer from a Foundation-listed provider

A reserved peer is a slot kept for your node. The provider adds your IP to its own reserved_peer_ips list, and in the words of the README, that setting lets a peer "always allow incoming connections from specific IPs." Your node always gets in, however busy the network is.

To make this easier to buy, the Hyperliquid Foundation keeps a list of infrastructure providers that have committed to public peering at a reasonable price. To be listed, a provider needs:

The list currently has 16 providers: 0xArchive, Alchemy, Allium, Altitude, ASXN, DoubleZero, Dwellir, Hydromancer, Hypedexer, Hyperdash, HyperTracker, Luganodes, Proof Group, Quicknode, Spectrum and ZAN. If a listed service falls short, the Foundation asks you to report it at bugbounty@hyperfoundation.org.

Prices and setups differ between providers, so ask each one:

Good for: trading systems, data products and anything where a 30-second stall costs money.

3. Peer directly with a Foundation sentry

The Foundation runs two sentries that outside nodes can connect to, but not everyone qualifies. You need to be a high-volume trader, with more than 0.5% of 14-day weighted maker volume, or a reputable infrastructure provider. Providers that qualify can then offer peering to everyone else, one hop below the sentry.

Good for: large market makers and infrastructure providers. Everyone else gets close to the same position by peering with a provider that sits one hop below a Foundation sentry.

Public peer vs reserved peer vs Foundation sentry

Public peerReserved peerFoundation sentry
CostFreeMonthly fee; the Foundation's
benchmark is under $1,000		Not published
Getting a slotOnly if one is open;
most are full		Kept for your IPOnce the Foundation
approves you
Uptime promiseNoneSet by the provider's
SLA		None stated
Mempool streamingOnly if every upstream
peer has it on		Ask the providerNot stated
Who can use itAnyoneAnyoneHigh-volume traders and
infrastructure providers
Best forTesting and backupsProduction systemsLarge market makers

Set up failover so a dropped peer costs seconds, not minutes

Even a good peer can drop during a network upgrade or an outage, so give your node somewhere else to go. Peers are set in ~/override_gossip_config.json:

json
{
  "root_node_ips": [
    { "Ip": "203.0.113.10" },
    { "Ip": "203.0.113.20" },
    { "Ip": "203.0.113.30" }
  ],
  "try_new_peers": false,
  "chain": "Mainnet",
  "reserved_peer_ips": [],
  "split_client_blocks": true
}

The IPs above are placeholders; replace them with your own peers.

The node reloads root_node_ips, reserved_peer_ips and try_new_peers without a restart, so you can swap a bad peer while the node runs. split_client_blocks takes effect on the next connection, so restart the node or wait for the next reconnect.

Finally, set up alerts so you hear about a stall before your users do. The node can post to Slack; the node setup guide shows how.

Hydromancer peering

Hydromancer is one of the providers on the Foundation's list, and we think it's the best peering service on Hyperliquid. Three things set it apart:

Hydromancer peering costs $500 a month with mempool streaming included, and comes with:

To start a trial, email data@hydromancer.xyz or message the team on Telegram or Discord.

FAQ

How do I find a reliable peer for my Hyperliquid node?

Check Flowscan's peer map (flowscan.xyz/peers) for peers marked Open, prefer ones in Tokyo close to a Foundation sentry, and list two or three in root_node_ips. For production systems, use a reserved peer from a provider on the Hyperliquid Foundation's list, which guarantees your node a slot.

Why does my Hyperliquid node keep losing its peer and falling behind?

Your node takes blocks from one parent peer at a time, so if that peer drops you, the node stalls for 10 to 30 seconds while it finds another. Most public peers are full, serving nodes costs them bandwidth, and some can't keep up when trading spikes. Closed ports 4001 and 4002 also make peers deprioritize your node.

Which ports does a Hyperliquid non-validator node need open?

Ports 4001 and 4002, open to the public. They carry gossip, and Hyperliquid's node README says peers deprioritize nodes that don't expose them. Ports 4000 to 4010 are only needed by validators.

Where can I see which Hyperliquid peers have free slots?

Flowscan's peer map at flowscan.xyz/peers labels each sentry as Open, Full or Offline and shows how many peers each one serves. Hyperliquid's gossipRootIps API request lists recently available peers, but it doesn't say which ones have room.

Do I need a paid peer to run a Hyperliquid node?

No. A non-validator can run on free public peers. But public peers have no uptime guarantee and are often full, so teams running trading systems or paid products usually add a reserved peer from a provider.

How much does a reserved Hyperliquid peer cost?

The Hyperliquid Foundation's benchmark for listed providers is under $1,000 a month, covering compute and egress. Hydromancer peering costs $500 a month with mempool streaming included.

Where should I host my Hyperliquid node?

In Tokyo. Hyperliquid recommends it for the lowest latency, and about 60% of the peers Flowscan crawls are in Japan. For single-digit-millisecond latency, use AWS zone apne1-az1 with a peer in the same zone.

Can I get mempool data through a Hyperliquid peer?

Yes. Set split_client_blocks to true in override_gossip_config.json and your node writes transactions before they're confirmed to ~/hl/data/mempool_txs/. Every peer between your node and the validators must have it on as well.

Sources

[1] Hyperliquid Docs: Foundation non-validating node for infra providers
[2] Hyperliquid node repository and README
[3] Flowscan: Hyperliquid peers, checked 18 October 2026
[4] Hydromancer docs: Peering
[5] Hydromancer: How to set up a Hyperliquid non-validator node

← All posts

Read next