The aim of this guide is to step you through how to launch a parachain outside of a local testing environment.
You will learn how to make sure
protocolId is set uniquely and that runtime weights are correct.
The runtime constraints on a parachain are much stricter than a solochain, as you must coordinate with the relay chain to finalize state transitions. When launching a parachain to production, it is critically important to make sure a chain's runtime is properly configured and tested.
Network collisions can cause major headaches.
All chains should use a unique
protocolId that no other network of any type—whether a test network, relay chain, or parachain—uses.
Having a unique protocol identifier ensures your nodes connect with the correct peer
nodes and not with nodes from other libp2p networks.
You want to isolate them to a distinct peer group with this ID.
Protocol ID collisions will cause many issues for your nodes.
In order to set a unique protocol ID, change make sure you use some nonce or salt value. This is set
(for the parachain node template)
as a CLI item in
/client/network/src/command.rs, and passed to extend the
// raw chain spec file in polkadot repo `/node/service/chain-specs/polkadot.json`
Monitor this issue for updates about a better way to safely configure this constant in the future.
Collator memory profiling should be done to analyze memory leaks, identify where memory consumption is happening, define temporary allocations, and investigate excessive memory fragmentation within applications.
When launching a parachain, it is critical to use the compressed version of the runtime to lower the amount of resource consumption as much as possible for the relay chain.
- It is recommended to launch a parachain with limited functionality and gradually increase it with runtime upgrades. The reason behind that is that during a runtime upgrade both the previous runtime and the new runtime are included in the PoVBlock and therefore if the changes are large enough the block might be rejected by the Relay Chain due to PoVBlock size limits.
- If the runtime is included in the state proof, ensure the PoV block (i.e. the set of extrinsics, including the new runtime, the PoV state proof, potentially the old runtime) fits within the PoVBlock size limit. If the runtime is not included in the state proof, the size limit of the new runtime will be much higher.
You can check the maximum sizes in the Polkadot repo for all relay chains (these are common constants). Make note of:
- The runtime version of the relay chain you are targeting (these may change)
You must have your parachain fit comfortably within these maxima. You can also use the the Polkadot-JS Apps UI connected to a relay node to see these constants: Developers -> ParachainsConfiguration -> ActiveConfiguration
Use runtime benchmarking to ensure that your runtime weights are actually indicative of the resources used by your runtime.
If you need to diverge from benchmarks, make sure that each pallet in your runtime employs the correct weighting system. Default and "guessed at" weights are not to be used in production, as a general rule.
It is recommended to have a block weight limit (block production time) of 0.5 seconds in the beginning due to uncertainties in block execution time. As the execution time of the network stabilizes the weight limit can be increased to 2 seconds.
Especially when launching a parachain, you might need to highly constrict what is enabled for specific classes of users. This can be accomplished with call filters.
If you are approaching limits outline above before launch (like a too-large large runtime) it is highly advisable to prune down functionality as much as practical and incrementally upgrade. In these cases, you can:
- Generate the genesis state of your chain with full runtime functionality (including all the pallets)
- Remove all pallets that you will not need upon parachain launch from your runtime
- Re-build the WASM blob (validation logic) and the runtime of the chain
- Register your parachain with the updated genesis and the WASM blob generated in (3)
- After your parachain is live you can upgrade your runtime on-chain to include the missing pallets (ensure that pallet indices and names match those used to generate the genesis state in step (1) without having to do storage migrations. For more information on on-chain runtime upgrades refer to the next section.
See the parachain runtime upgrade guide for how to go about actually performing these incremental runtime upgrades.
Before you try anything on a production testnet or mainnet, you should launch your chain on a network that simulates the behavior of a real network as closely as possible. 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.
See the Prepare a local relay chain to for a selection of tools for automation of such testing.