@ -13,7 +13,8 @@ There are two types of full node that use different mechanisms to sync up to the
A snap sync'd node holds the most recent 128 block states in memory, so transactions in that range are always quickly accessible. However, snap-sync only starts processing from a relatively recent block (as opposed to genesis for a full node). Between the initial sync block and the 128 most recent blocks, the node stores occasional checkpoints that can be used to rebuild the state on-the-fly. This means transactions can be traced back as far as the block that was used for the initial sync. Tracing a single transaction requires reexecuting all preceding transactions in the same block **and** all preceding blocks until the previous stored snapshot. Snap-sync'd nodes are therefore full nodes, with the only difference being the initial synchronization required a checkpoint block to sync from instead of independently verifying the chain all the way from genesis. Snap sync then only verifies the proof-of-work and ancestor-child block progression and assumes that the state transitions are correct rather than re-executing the transactions in each block to verify the state changes. Snap sync is much faster than block-by-block sync. To start a node with snap sync pass `--syncmode snap` at startup.
A snap sync'd node holds the most recent 128 block states in memory, so transactions in that range are always quickly accessible. However, snap-sync only starts processing from a relatively recent block (as opposed to genesis for a full node). Between the initial sync block and the 128 most recent blocks, the node stores occasional checkpoints that can be used to rebuild the state on-the-fly. This means transactions can be traced back as far as the block that was used for the initial sync. Tracing a single transaction requires reexecuting all preceding transactions in the same block **and** all preceding blocks until the previous stored snapshot. Snap-sync'd nodes are therefore full nodes, with the only difference being the initial synchronization required a checkpoint block to sync from instead of independently verifying the chain all the way from genesis. Snap sync then only verifies the proof-of-work and ancestor-child block progression and assumes that the state transitions are correct rather than re-executing the transactions in each block to verify the state changes. Snap sync is much faster than block-by-block sync. To start a node with snap sync pass `--syncmode snap` at startup.
Snap sync starts by downloading the headers for a chunk of blocks. Once the headers have been verified, the block bodies and receipts for those blocks are downloaded. In parallel, Geth also sync begins state-sync. In state-sync, Geth first downloads the leaves of the state trie for each block without the intermediate nodes along with a range proof. The state trie is then regenerated locally. The state download is the part of the snap-sync that takes the most time to complete and the progress can be monitored using the ETA values in the log messages. However, the blockchain is also progressing at the same time and invalidating some of the regenerated state data. This means it is also necessary to have a 'healing' phase where errors in the state are fixed. It is not possible to monitor the progress of the state heal because the extent of the errors cannot be known until the current state has already been regenerated.
Snap sync starts by downloading the headers for a chunk of blocks. Once the headers have been verified, the block bodies and receipts for those blocks are downloaded. In parallel, Geth also sync begins state-sync. In state-sync, Geth first downloads the leaves of the state trie for each block without the intermediate nodes along with a range proof. The state trie is then regenerated locally. The state download is the part of the snap-sync that takes the most time to complete and the progress can be monitored using the ETA values in the log messages. However, the blockchain is also progressing at the same time and invalidating some of the regenerated state data. This means it is also necessary to have a 'healing' phase where errors in the state are fixed. It is not possible to monitor the progress of the state heal because the extent of the errors cannot be known until the current state has already been regenerated. Geth regularly reports `Syncing, state heal in progress` regularly during state heal - this informs the user that state heal has not finished. It is also possible to confirm this using `eth.syncing` - if this command returns `false` then the node is in sync. If it returns anything other than `false` then syncing is still in progress.
The healing has to outpace the growth of the blockchain, otherwise the node will never catch up to the current state. There are some hardware factors that determine the speed of the state healing (speed of disk read/write and internet connection) and also the total gas used in each block (more gas means more changes to the state that have to be handled).
The healing has to outpace the growth of the blockchain, otherwise the node will never catch up to the current state. There are some hardware factors that determine the speed of the state healing (speed of disk read/write and internet connection) and also the total gas used in each block (more gas means more changes to the state that have to be handled).
To summarize, snap sync progresses in the following sequence:
To summarize, snap sync progresses in the following sequence:
@ -22,8 +23,10 @@ To summarize, snap sync progresses in the following sequence:
- download block bodies and receipts. In parallel, download raw state data and build state trie
- download block bodies and receipts. In parallel, download raw state data and build state trie
- heal state trie to account for newly arriving data
- heal state trie to account for newly arriving data
**Note** Snap sync is the default behaviour, so if the `--syncmode` value is not passed to Geth at startup, Geth will use snap sync. A node that is started using `snap` will switch to block-by-block sync once it has caught up to the head of the chain.
**Note** Snap sync is the default behaviour, so if the `--syncmode` value is not passed to Geth at startup, Geth will use snap sync. A node that is started using `snap` will switch to block-by-block sync once it has caught up to the head of the chain.
### Full
### Full
A full sync generates the current state by executing every block starting from the genesis block. A full sync indendently verifies proof-of-work and block provenance as well as all state transitions by re-executing the transactions in the entire historical sequence of blocks. Only the most recent 128 block states are stored in a full node - older block states are pruned periodically and represented as a series of checkpoints from which any previous state can be regenerated on request. 128 blocks is about 25.6 minutes of history with a block time of 12 seconds.
A full sync generates the current state by executing every block starting from the genesis block. A full sync indendently verifies proof-of-work and block provenance as well as all state transitions by re-executing the transactions in the entire historical sequence of blocks. Only the most recent 128 block states are stored in a full node - older block states are pruned periodically and represented as a series of checkpoints from which any previous state can be regenerated on request. 128 blocks is about 25.6 minutes of history with a block time of 12 seconds.
@ -38,13 +41,13 @@ It is also possible to create a partial/recent archive node where the node was s
## Light nodes
## Light nodes
A light node syncs very quickly and stores the bare minimum of blockchain data. Light nodes only process block headers, not entire blocks. This greatly reduces the computation time, storage and bandwidth required relative to a full node. This means light nodes are suitable for resource-constrained devices and can catch up to the head of the chain much faster when they are new or have been offline for a while. The trade-off is that light nodes rely heavily on data served by altruistic full nodes. A light client can be used to query data from Ethereum and submit transactions, acting as a locally-hosted Ethereum wallet. However, because they don't keep local copies of the Ethereum state, light nodes can't validate blocks in the same way as full nodes - they receive a proof from the full node and verify it against their local header chain. To start a node in light mode, pass `--syncmode light`. Be aware that full nodes serving light data are relative scarce so light nodes can struggle to find peers.
A light node syncs very quickly and stores the bare minimum of blockchain data. Light nodes only process block headers, not entire blocks. This greatly reduces the computation time, storage and bandwidth required relative to a full node. This means light nodes are suitable for resource-constrained devices and can catch up to the head of the chain much faster when they are new or have been offline for a while. The trade-off is that light nodes rely heavily on data served by altruistic full nodes. A light client can be used to query data from Ethereum and submit transactions, acting as a locally-hosted Ethereum wallet. However, because they don't keep local copies of the Ethereum state, light nodes can't validate blocks in the same way as full nodes - they receive a proof from the full node and verify it against their local header chain. To start a node in light mode, pass `--syncmode light`. Be aware that full nodes serving light data are relative scarce so light nodes can struggle to find peers.**Light nodes are not currently working on proof-of-stake Ethereum**.
Read more about light nodes on our [LES page](/docs/interface/les.md).
Read more about light nodes on our [LES page](/docs/interface/les.md).
## Consensus layer syncing
## Consensus layer syncing
Now that Ethereum has switched to proof-of-stake, all consensus logic and block propagation is handled by consensus clients. This means that syncing the blockchain is a process shared between the consensus and execution clients. Blocks are downloaded by the consensus client and verified by the execution client. In order for Geth to sync, it requires a header from its connected consensus client. Geth does not import any data until it is instructed to by the consensus client.
Now that Ethereum has switched to proof-of-stake, all consensus logic and block propagation is handled by consensus clients. This means that syncing the blockchain is a process shared between the consensus and execution clients. Blocks are downloaded by the consensus client and verified by the execution client. In order for Geth to sync, it requires a header from its connected consensus client. Geth does not import any data until it is instructed to by the consensus client.**Geth cannot sync without being connected to a consensus client**. This includes block-by-block syncing from genesis.
Once a header is available to use as a syncing target, Geth retrieves all headers between that target header and the local header chain in reverse chronological order. These headers show that the sequence of blocks is correct because the parenthashes link one block to the next right up to the target block. Eventually, the sync will reach a block held in the local database, at which point the local data and the target data are considered 'linked' and there is a very high chance the node is syncing the correct chain. The block bodies are then downloaded and then the state data. The consensus client can update the target header - as long as the syncing outpaces the growth of the blockchain then the node will eventually get in sync.
Once a header is available to use as a syncing target, Geth retrieves all headers between that target header and the local header chain in reverse chronological order. These headers show that the sequence of blocks is correct because the parenthashes link one block to the next right up to the target block. Eventually, the sync will reach a block held in the local database, at which point the local data and the target data are considered 'linked' and there is a very high chance the node is syncing the correct chain. The block bodies are then downloaded and then the state data. The consensus client can update the target header - as long as the syncing outpaces the growth of the blockchain then the node will eventually get in sync.
Joe
2 years ago