The goal of this guide is to help parachain developers ensure that runtime upgrades succeed.
This guide illustrates:
- How to modify parachain runtimes (add/remove pallets)
- How to do a parachain storage migration
Runtime upgrades on a parachain have much stricter requirements and a slightly different flow required as they you must coordinate with the relay chain to facilitate this. Because of this, and the very confined nature of state transition coordination both quickly and succinctly enough for the block inclusion in the relay chain.
Please do the following:
- Read the general runtime upgrade docs.
- Complete the Prepare a local relay chain tutorial, and learn the
polkadot-launchtool for testing.
If your existing Substrate chain has a very large state which you are migrating between different storage formats, it might not be possible to run all of the runtime migrations within one block. There are a handful of strategies you can use to remedy this problem:
- If the amount of storage items to be migrated can feasibly be processed within two or three blocks you can run the migrations using the Scheduler pallet to ensure they get executed regardless of block producer.
- Use versioned storage and only execute migrations when storage values that haven't yet been upgraded are accessed. This can cause variance in transaction fees between users and could potentially result in more complex runtime code. However, if properly metered (weights are properly benchmarked) this approach will ensure minimal downtime for migration.
If you must split your migrations among multiple blocks you can do it either on-chain or off-chain:
- An on-chain multi-block migration will require custom pallet logic to be written which can either queue changes over time or use the Scheduler pallet to migrate chunks of storage at a time.
- Instead of adding migration code to your runtime you can generate the migration manually off-chain and use multiple
system.setStoragecalls to add and remove storage items as necessary via an origin with root permission (for example democracy). If you are limited in the number of transactions you can make, you can batch multiple transactions to occur over time via the scheduler.
After your migration strategy is established, you should test the migration on a non-production testnet to ensure it will work before you continue.
Testing in a confined network will help you prepare for potential failures in a real network with many collators and validators and constraints like bandwidth and latency. The more closely you can simulate a real network for testing, the more sure you can be that your runtime upgrades will succeeds.
When finally ready to upgrade a parachain, the relay chain needs to be informed about the runtime upgrade of your chain before it happens. The Cumulus library provides functionality to help you notify the relay chain about the upcoming upgrade by:
authorize_upgradeto provide the hash of your upgrade and authorize it.
enact_authorized_upgradeto provide the actual code for the upgrade.
With both these functions called, the relay chain will be notified that the new upgrade has been scheduled.