~/notes / node operation
What a Validator Actually Runs
The software stack, the operational requirements and the failure modes. Considerably more mundane than the discussion of consensus suggests.
Running a validator is a systems administration job. The cryptography is handled by software; what you manage is uptime and keys.
The components
A consensus client. Participates in the consensus protocol, produces attestations and proposals.
An execution client. Processes transactions and maintains state. On modern proof-of-stake designs both are required and they communicate over an authenticated connection.
A validator client. Holds the signing keys and produces signatures when the consensus client requests them. Frequently run as a separate process for security.
A slashing protection database. Records what has already been signed, preventing the client from signing anything that would constitute equivocation.
That last component is the most important file on the machine and the one most often lost during a restore.
The operational requirements
Uptime. Downtime costs rewards and incurs a small penalty. It does not destroy stake.
Correct time. Clock drift causes late attestations and lost rewards.
Disk space. State grows continuously. A validator that runs out of disk stops validating.
Bandwidth. Continuous, modest, and it must be reliable rather than fast.
Updates. Client releases must be applied, particularly before network upgrades. A validator running outdated software through a hard fork is left on a chain nobody else is following.
The failure modes, in order of how often they occur
Disk full. The most common by a wide margin.
Clock drift.
Failed update before an upgrade.
Network connectivity.
Double signing from running keys in two places. Rare and catastrophic, and it is the only one that destroys stake.
The rule that prevents the catastrophic one
Keys active in exactly one place, ever.
Failover arrangements that bring a backup online while the primary is running produce double attestations immediately. The penalty for downtime is small; the penalty for equivocation is not. Prefer downtime.
What it does not involve
Trading decisions. Market views. Anything about price.
A validator’s job is to be online and honest. The reward follows from that and from the protocol’s issuance schedule, and it is not affected by anything the operator thinks.
The economics
Rewards are denominated in the asset. The volatility of the asset dominates the yield by a wide margin, which means validator operation is a business whose revenue is set by something entirely outside its control.
For anyone considering it, the question is whether you want to run infrastructure. If the objective is only the yield, delegating or using a venue’s staking service, such as exchanges that let you specify the network, achieves the same exposure without the operational responsibility, at the cost of a commission and a counterparty.
# Corrections and technical nitpicks are welcome. Send them over. They get published with the fix.