A clone of an already great blocks subgraph, but this one prunes history so you only store the latest block, but can use time travel queries to get block information for N blocks in the past that are automatically pruned.
{
block(id:"latest_block"){
id
number
timestamp
}
}
{
blocks(block: { number: 404100 }){
id
number
timestamp
}
}
yarn install && yarn codegen && yarn build
Configurerd to deploy to Goldsky, but should work with other subgraph providers.
Make sure to edit:
prune
: set this to the number of blocks to keep behind the head block of the chain, I refer to this as N. Graph Node by default storesN * 1.2
blocks of history before pruning.startBlock
: this can be the most recent block to make indexing fast, or the startBlock-N blocks to index the past N blocks.network
: this should work on any EVM network, so change the network to match the network you want to index blocks for.