Forem: The Good Shell

Four things that will get your Cosmos validator slashed before you earn a single block reward

The Good Shell — Tue, 07 Apr 2026 16:04:00 +0000

The most dangerous moment in a Cosmos validator setup is not the on-chain registration. It is the ten minutes before it, when your priv_validator_key.json is sitting unprotected on the validator host and you are about to run create-validator for the first time.
Most guides walk you through the steps. Fewer of them tell you the specific things that will get you jailed or slashed if you skip them. These are four of them, from running validators on Cosmos Hub mainnet.

1. NVMe is not optional, it is the difference between signing blocks and missing them

Every guide lists "4TB SSD" as a hardware requirement. What most of them do not emphasize is that SATA SSDs and standard HDDs will cause I/O bottlenecks under load that manifest directly as missed blocks.
The chain data on Cosmos Hub has grown significantly. Under normal operation, the node is continuously reading and writing to disk. During governance-triggered upgrades, that load spikes. If your disk cannot keep up, the node falls behind on block processing and starts missing signatures.
NVMe specifically matters because the throughput difference between NVMe and SATA SSD is not marginal. It is the difference between a node that stays in sync under pressure and one that starts accumulating missed blocks at exactly the moment you can least afford it.
RAM is the second one people underestimate. You need 64GB. The 32GB setups work fine in normal operation. They fail during upgrades, when memory spikes well above the normal operating baseline. Running out of memory at upgrade height is a jailing event.

Never set DAEMON_ALLOW_DOWNLOAD_BINARIES=true in Cosmovisor This feels counterintuitive. Cosmovisor's auto-download feature sounds useful, you stage the upgrade in governance, and Cosmovisor downloads and swaps the binary automatically at the right block height. The problem is what happens when the download fails. If the binary cannot be fetched at upgrade height, the node halts immediately. You are now racing to manually place the binary before the jailing threshold kicks in. On Cosmos Hub, that window is approximately 500 blocks, around 16 minutes at normal block times. The safer pattern is to always pre-place upgrade binaries manually in the Cosmovisor upgrade directory before the governance proposal passes. You monitor the proposal, you compile and verify the binary, you put it in place. Cosmovisor finds it already there and does the swap cleanly. DAEMON_ALLOW_DOWNLOAD_BINARIES=false forces you into this pattern. It removes the failure mode where an auto-download kills your uptime at exactly the worst moment.

3. The migration double-sign window is where most slashing events happen

Double-sign slashing is permanent. It does not unjail. The tombstone is final.
The scenario that causes it most often is not a configuration mistake during initial setup. It is a validator migration: moving from one host to another. The sequence that causes it:
Old node is stopped. New node is started. Old node process was not actually stopped, or was restarted by a systemd restart policy, or a snapshot was used and the old node resumed from a state that did not reflect the stop.
Both nodes are now signing with the same key. Double-sign event. Tombstone.
The protection is simple but must be deliberate. When migrating: stop the old node, wait for a minimum of 10 confirmed blocks with no signing activity from that key, then start the new node. Never start the new node and then stop the old one. Never assume a stop command worked without verifying it.
Setting double_sign_check_height to a non-zero value in config.toml (10 to 20 blocks is standard) adds a second layer. The node will check recent block history before signing and refuse to sign if it detects a potential double-sign situation.

4. The sentry architecture is what keeps your validator IP off the public internet

A validator without sentry nodes has its IP address visible in the P2P network. That is a DDoS target. Taking your validator offline long enough to miss 5% of blocks in a sliding window triggers jailing on Cosmos Hub.
The sentry pattern is straightforward: two or more public-facing full nodes handle all external P2P connections. The validator node only connects to the sentries, never to the broader network. Its IP is never gossiped to peers.
On the validator node, this means pex = false and persistent_peers pointing only to the sentry node IDs. On the sentry nodes, the validator node ID is listed in private_peer_ids so its address is never shared with the network.
Run sentries in at least two different geographic regions and on different providers. A DDoS that takes down one sentry is neutralised if the second is on a separate network.

These four are the ones that cause the most production incidents on Cosmos validators: the hardware under-specification, the auto-download failure mode, the migration double-sign window, and the missing sentry layer. The rest of the setup, Go installation, gaiad build, state sync, TMKMS configuration, on-chain registration, is more mechanical.
If you want the full setup with all the configuration files and commands from start to production, I wrote a detailed guide covering the complete process:
Cosmos Validator Setup: The Ultimate Step-by-Step Guide for 2026
Happy to answer questions in the comments if you are working through any of these.

Bootnode Security: 6 Essential Hardening Layers to Protect Your Web3 Network

The Good Shell — Tue, 31 Mar 2026 08:38:05 +0000

If you run a blockchain network private, permissioned, or public you have at least one bootnode. Almost nobody has hardened it properly.
This is understandable. Bootnodes are infrastructure plumbing. They don't hold keys, they don't sign transactions. The assumption is that if a bootnode goes down, the network just loses peer discovery for a while. That assumption is wrong.
Here's what a compromised bootnode actually enables: eclipse attacks. An attacker who controls your bootnode can feed newly joining nodes a list of attacker-controlled peers. Those nodes then sync from attacker-controlled infrastructure. For a DeFi protocol or validator, this creates conditions for double-spend attacks, transaction censorship, and consensus manipulation.
A January 2026 paper on arXiv demonstrated the first practical end-to-end eclipse attack against post-Merge Ethereum execution layer nodes. This is not theoretical anymore.
This guide covers 6 hardening layers that every production bootnode needs.

The Real Threat Model

Before writing a single firewall rule, understand what you're actually defending against:
DDoS against the discovery port bootnodes run UDP on port 30303 by default. UDP is stateless and easy to flood. A sustained attack takes down peer discovery for your entire network.
Enode key compromise the enode private key is your bootnode's identity. If an attacker steals it, they can impersonate your bootnode indefinitely with a node your network trusts.
Eclipse attacks via discovery poisoning — attackers inject malicious nodes into a target's peer database using passive discovery behavior. A bootnode without rate limiting amplifies this attack.
*Sybil attacks against the discovery table * bootnodes maintain a Kademlia-style table with 17 K-buckets, each holding up to 16 nodes. A Sybil attacker floods the table with controlled node IDs, crowding out legitimate peers. New nodes then get routed exclusively to attacker-controlled infrastructure.

Layer 1 - Host Hardening

Run nothing else on the bootnode host. Minimal attack surface is not optional.

# Disable unnecessary services
systemctl disable --now snapd cups avahi-daemon bluetooth

# SSH hardening /etc/ssh/sshd_config
Port 22222
PermitRootLogin no
PasswordAuthentication no
PubkeyAuthentication yes
AllowUsers bootnode-admin
MaxAuthTries 3
X11Forwarding no
AllowTcpForwarding no

Store the enode key on an encrypted volume:

cryptsetup luksFormat /dev/sdb
cryptsetup luksOpen /dev/sdb bootnode-keys
mkfs.ext4 /dev/mapper/bootnode-keys
mount /dev/mapper/bootnode-keys /mnt/bootnode-keys
chmod 700 /mnt/bootnode-keys

Layer 2 - Network Hardening

This is where most bootnode security implementations fall apart. The default allows connections from any IP on any port. Fine for getting started. Not acceptable in production.

ufw default deny incoming
ufw default allow outgoing
ufw allow from <MANAGEMENT_IP> to any port 22222 proto tcp
ufw allow 30303/udp
ufw allow 30303/tcp
ufw enable

Rate limit UDP with iptables UFW alone doesn't rate-limit UDP:

iptables -A INPUT -p udp --dport 30303 -m hashlimit \
  --hashlimit-name udp-discovery \
  --hashlimit-above 100/second \
  --hashlimit-burst 200 \
  --hashlimit-mode srcip \
  -j DROP

For private/permissioned networks: restrict discovery to known IP ranges. There is no reason your bootnode should accept requests from arbitrary internet IPs.

ufw allow from <NODE_IP_RANGE>/24 to any port 30303
ufw deny 30303

This single change is the most impactful improvement for private networks and almost nobody does it.

Layer 3 - Enode Key Management

Generate the key before starting the node. Never let the client auto-generate it.

# Generate and record the public key
bootnode -genkey /mnt/bootnode-keys/bootnode.key
bootnode -nodekey /mnt/bootnode-keys/bootnode.key -writeaddress

# Secure permissions
chmod 400 /mnt/bootnode-keys/bootnode.key
chown bootnode-service:bootnode-service /mnt/bootnode-keys/bootnode.key

Systemd with sandboxing:

# /etc/systemd/system/bootnode.service
[Service]
User=bootnode-service
ExecStart=/usr/local/bin/bootnode \
  -nodekey /mnt/bootnode-keys/bootnode.key \
  -addr :30303
Restart=always
NoNewPrivileges=true
PrivateTmp=true
ProtectSystem=strict
ReadWritePaths=/mnt/bootnode-keys

Back up the key to offline storage immediately. The offline backup must be tested, not just created.

Layer 4 - Eclipse Attack Prevention

Run at least 3 geographically distributed bootnodes across different cloud providers. An attacker needs to compromise all three simultaneously to control peer discovery.

# Each node points to all bootnodes
geth --bootnodes \
  "enode://<pubkey1>@<ip1>:30303,enode://<pubkey2>@<ip2>:30303,enode://<pubkey3>@<ip3>:30303"

Each bootnode lists the others for faster discovery and resilience.
Enable ENR/Discv5 where supportedit includes cryptographic verification that makes node impersonation significantly harder than legacy enode.

Layer 5 - Monitoring and Alerting

# Prometheus alerting rules
groups:
  - name: bootnode.security
    rules:
      - alert: BootnodeDown
        expr: up{job="bootnode"} == 0
        for: 2m
        labels:
          severity: critical

      - alert: BootnodePeerCountDrop
        expr: p2p_peers < 5
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: "Low peer count possible eclipse or DDoS"

      - alert: BootnodeUDPFlood
        expr: rate(net_p2p_ingress_bytes_total[1m]) > 50000000
        for: 1m
        labels:
          severity: critical
        annotations:
          summary: "Possible DDoS on discovery port"

Layer 6 - Disaster Recovery and Key Rotation

If the bootnode key is compromised, you need a pre-defined rotation procedure. Test it before you need it.

# Generate new key on new instance
bootnode -genkey /mnt/keys/bootnode-new.key
bootnode -nodekey /mnt/keys/bootnode-new.key -writeaddress > new-enode-pubkey.txt

# Push new enode to all network nodes via Ansible
# Bring up new bootnode
systemctl start bootnode-new

# After confirming healthy take down compromised node
systemctl stop bootnode-old

Multi-region deployment is non-negotiable for production:

Region 1 (AWS eu-west-1) elastic IP
Region 2 (Hetzner Helsinki) static IP
Region 3 (GCP us-east1) static IP

Different providers means a cloud-level outage doesn't take down your entire discovery layer.
The Quick Checklist
Before deploying any production bootnode:
Host: dedicated host, SSH on non-standard port, key-only auth, disk encryption for keys, systemd sandboxing.
Network: UFW default deny, UDP rate limiting, SSH restricted to management IP, IP allowlisting for private networks.
Enode key: generated pre-start, encrypted volume, 400 permissions, offline backup tested, rotation runbook documented.
Architecture: minimum 3 bootnodes, cross-region, cross-provider, cross-referencing each other.
Monitoring: Prometheus scraping, alerts on down/peer drop/UDP flood/SSH failures.

Wrapping Up

Bootnode security is the gap between "we have a network" and "we have a network that can't be trivially disrupted." Eclipse attacks against post-Merge Ethereum were demonstrated in peer-reviewed research in January 2026. The technical foundation has existed since 2018.
None of this is exotic. Every protection here is standard Linux and networking practice applied to a blockchain-specific context. One solid day of work. The result is a bootnode that withstands DDoS, resists eclipse attempts, and survives key compromise with a clean rotation procedure.
Questions? Drop them in the comments happy to go deeper on any of these layers.

Originally published at thegoodshell.com