cerc-io/ipld-eth-server
18
0
Fork 1
Code Issues 10 Pull Requests 1 Actions Packages Projects Releases 37 Wiki Activity
358575335b
ipld-eth-server/documentation/super-node.md

13 KiB
Raw Blame History

Super Node

Vulcanizedb can act as an index for Ethereum data stored on IPFS through the use of the syncAndPublish and syncPublishScreenAndServe commands.

Manual Setup

These commands work in conjunction with a state-diffing full Geth node and IPFS.

IPFS

To start, download and install IPFS

go get github.com/ipfs/go-ipfs

cd $GOPATH/src/github.com/ipfs/go-ipfs

make install

If we want to use Postgres as our backing datastore, we need to use the vulcanize fork of go-ipfs.

Start by adding the fork and switching over to it:

git remote add vulcanize https://github.com/vulcanize/go-ipfs.git

git fetch vulcanize

git checkout -b postgres_update vulcanize/postgres_update

Now install this fork of ipfs, first be sure to remove any previous installation.

make install

Check that is installed properly by running

ipfs

You should see the CLI info/help output.

And now we initialize with the postgresds profile. If ipfs was previously initialized we will need to remove the old profile first. We also need to provide env variables for the postgres connection:

We can either set these manually, e.g.

export IPFS_PGHOST=
export IPFS_PGUSER=
export IPFS_PGDATABASE=
export IPFS_PGPORT=
export IPFS_PGPASSWORD=

And then run the ipfs command

ipfs init --profile=postgresds

Or we can use the pre-made script at GOPATH/src/github.com/ipfs/go-ipfs/misc/utility/ipfs_postgres.sh which has usage:

./ipfs_postgres.sh <IPFS_PGHOST> <IPFS_PGPORT> <IPFS_PGUSER> <IPFS_PGDATABASE>"

and will ask us to enter the password, avoiding storing it to an ENV variable.

Once we have initialized ipfs, that is all we need to do with it- we do not need to run a daemon during the subsequent processes (in fact, we can't).

Geth

For Geth, we currently require a special fork, and we can set this up as follows:

Begin by downloading geth and switching to the vulcanize/rpc_statediffing branch

go get github.com/ethereum/go-ethereum

cd $GOPATH/src/github.com/ethereum/go-ethereum

git remote add vulcanize https://github.com/vulcanize/go-ethereum.git

git fetch vulcanize

git checkout -b statediffing vulcanize/statediffing

Now, install this fork of geth (make sure any old versions have been uninstalled/binaries removed first)

make geth

And run the output binary with statediffing turned on:

cd $GOPATH/src/github.com/ethereum/go-ethereum/build/bin

./geth --statediff --statediff.streamblock --ws --syncmode=full

Note: other CLI options- statediff specific ones included- can be explored with ./geth help

The output from geth should mention that it is Starting statediff service and block synchronization should begin shortly thereafter. Note that until it receives a subscriber, the statediffing process does essentially nothing. Once a subscription is received, this will be indicated in the output.

Also in the output will be the websocket url and ipc paths that we will use to subscribe to the statediffing process. The default ws url is "ws://127.0.0.1:8546" and the default ipcPath- on Darwin systems only- is "Users/user/Library/Ethereum/geth.ipc"

Vulcanizedb

There are two commands to choose from:

syncAndPublish

syncAndPublih performs the functions of the super node- syncing data from Geth, converting them to IPLDs, publishing those IPLDs to IPFS, and creating a local Postgres index to relate their CIDS to useful metadata.

Usage:

./vulcanizedb syncAndPublish --config=<config_file.toml>

The config file for the syncAndPublish command looks very similar to the basic config file

[database]
    name     = "vulcanize_demo"
    hostname = "localhost"
    port     = 5432

[client]
    ipcPath  = "ws://127.0.0.1:8546"
    ipfsPath = "/Users/user/.ipfs"

With an additional field, client.ipcPath, that is either the ws url or the ipc path that Geth has exposed (the url and path output when the geth sync was started), and client.ipfsPath which is the path the ipfs datastore directory.

syncPublishScreenAndServe

syncPublishScreenAndServe does everything that syncAndPublish does, plus it opens up an RPC server which exposes an endpoint to allow transformers to subscribe to subsets of the sync-and-published data that are relevant to their transformations

Usage:

./vulcanizedb syncPublishScreenAndServe --config=<config_file.toml>

The config file for the syncPublishScreenAndServe command has two additional fields and looks like:

[database]
    name     = "vulcanize_demo"
    hostname = "localhost"
    port     = 5432

[client]
    ipcPath  = "ws://127.0.0.1:8546"
    ipfsPath = "/Users/user/.ipfs"

[server]
    ipcPath = "/Users/user/.vulcanize/vulcanize.ipc"
    wsEndpoint = "127.0.0.1:80"
    
[superNodeBackFill]
    on = false
    ipcPath = ""
    frequency = 5

The additional server.ipcPath and server.wsEndpoint fields are used to set what ipc endpoint and ws url the syncPublishScreenAndServe rpc server will expose itself to subscribing transformers over, respectively. Any valid and available path and endpoint is acceptable, but keep in mind that this path and endpoint need to be known by transformers for them to subscribe to the super node.

Because the super node syncs data from a geth full node as it progresses through its block synchronization, there is potential for the super node to miss data both at the beginning of the sync due to lag between initialization of the two processes and anywhere throughout the sync if the processes are interrupted. The superNodeBackFill config mapping is used to optionally configure the super node with an archival geth client that exposes a statediff.StateDiffAt rpc endpoint, to enable it to fill in these data gaps. superNodeBackFill.on turns the backfill process on, the superNodeBackFill.ipcPath is the rpc path for the archival geth node, and superNodeBackFill.frequency sets at what frequency (in minutes) the backfill process checks for and fills in gaps.

Dockerfile Setup

The below provides step-by-step directions for how to setup the super node using the provided Dockerfile on an AWS Linux AMI instance. Note that the instance will need sufficient memory and storage for this to work.

  1. Install basic dependencies
sudo yum update
sudo yum install -y curl gpg gcc gcc-c++ make git
  1. Install Go 1.12
wget https://dl.google.com/go/go1.12.6.linux-amd64.tar.gz
tar -xzf go1.12.6.linux-amd64.tar.gz
sudo mv go /usr/local
  1. Edit .bash_profile to export GOPATH
export GOROOT=/usr/local/go
export GOPATH=$HOME/go
export PATH=$GOPATH/bin:$GOROOT/bin:$PATH
  1. Install and setup Postgres
sudo yum install postgresql postgresql96-server
sudo service postgresql96 initdb
sudo service postgresql96 start
sudo -u postgres createuser -s ec2-user
sudo -u postgres createdb ec2-user
sudo su postgres
psql
ALTER USER "ec2-user" WITH SUPERUSER;
/q
exit

4b. Edit hba_file to trust connections

psql
SHOW hba_file;
/q
sudo vim {PATH_TO_FILE}

4c. Stop and restart Postgres server to affect changes

sudo service postgresql96 stop
sudo service postgresql96 start
  1. Install and start Docker (exit and re-enter ec2 instance afterwards to affect changes)
sudo yum install -y docker
sudo service  docker start
sudo usermod -aG docker ec2-user
  1. Fetch the repository and switch to this working branch
go get github.com/vulcanize/vulcanizedb
cd $GOPATH/src/github.com/vulcanize/vulcanizedb
git checkout ipfs_concurrency
  1. Create the db
createdb vulcanize_public
  1. Build and run the Docker image
cd $GOPATH/src/github.com/vulcanize/vulcanizedb/dockerfiles/super_node
docker build .
docker run --network host -e VDB_PG_CONNECT=postgres://localhost:5432/vulcanize_public?sslmode=disable {IMAGE_ID}

Subscribing

A transformer can subscribe to the syncPublishScreenAndServe service over its ipc or ws endpoints, when subscribing the transformer specifies which subsets of the synced data it is interested in and the server will forward only these data.

The streamSubscribe command serves as a simple demonstration/example of subscribing to the super-node feed, it subscribes with a set of parameters defined in the loaded config file, and prints the streamed data to stdout. To build transformers that subscribe to and use super-node data, the shared/libraries/streamer can be used.

Usage:

./vulcanizedb streamSubscribe --config=<config_file.toml>

The config for streamSubscribe has the subscribe set of parameters, for example:

[subscription]
    path = "ws://127.0.0.1:8080"
    backfill = true
    backfillOnly = false
    startingBlock = 0
    endingBlock = 0
    [subscription.headerFilter]
        off = false
        uncles = false
    [subscription.trxFilter]
        off = false
        src = [
            "0xde0B295669a9FD93d5F28D9Ec85E40f4cb697BAe",
        ]
        dst = [
            "0xde0B295669a9FD93d5F28D9Ec85E40f4cb697BAe",
        ]
    [subscription.receiptFilter]
        off = false
        topic0s = [
            "0xddf252ad1be2c89b69c2b068fc378daa952ba7f163c4a11628f55a4df523b3ef",
            "0x930a61a57a70a73c2a503615b87e2e54fe5b9cdeacda518270b852296ab1a377"
        ]
    [subscription.stateFilter]
        off = false
        addresses = [
           "0xde0B295669a9FD93d5F28D9Ec85E40f4cb697BAe"
       ]
       intermediateNodes = false
    [subscription.storageFilter]
        off = true
        addresses = [
            "",
            ""
        ]
        storageKeys = [
            "",
            ""
        ]
        intermediateNodes = false

subscription.path is used to define the ws url OR ipc endpoint we will subscribe to the super-node over (the server.ipcPath or server.wsEndpoint that the super-node has defined in their config file).

subscription.backfill specifies whether or not the super-node should look up historical data in its cache and send that to the subscriber, if this is set to false then the super-node only forwards newly synced/incoming data.

subscription.backfillOnly will tell the super-node to only send historical data and not stream incoming data going forward.

subscription.startingBlock is the starting block number for the range we want to receive data in.

subscription.endingBlock is the ending block number for the range we want to receive data in; setting to 0 means there is no end/we will continue indefinitely.

subscription.headerFilter has two sub-options: off and uncles. Setting off to true tells the super-node to not send any headers to the subscriber; setting uncles to true tells the super-node to send uncles in addition to normal headers.

subscription.trxFilter has three sub-options: off, src, and dst. Setting off to true tells the super-node to not send any transactions to the subscriber; src and dst are string arrays which can be filled with ETH addresses we want to filter transactions for, if they have any addresses then the super-node will only send transactions that were sent or received by the addresses contained in src and dst, respectively.

subscription.receiptFilter has four sub-options: off, topics, contracts and matchTxs. Setting off to true tells the super-node to not send any receipts to the subscriber; topic0s is a string array which can be filled with event topics we want to filter for, if it has any topics then the super-node will only send receipts that contain logs which have that topic0. Similarly, contracts is a string array which can be filled with contract addresses we want to filter for, if it contains any contract addresses the super-node will only send receipts that correspond to one of those contracts. matchTrxs is a bool which when set to true any receipts that correspond to filtered for transactions will be sent by the super-node, regardless of whether or not the receipt satisfies the topics or contracts filters.

subscription.stateFilter has three sub-options: off, addresses, and intermediateNodes. Setting off to true tells the super-node to not send any state data to the subscriber; addresses is a string array which can be filled with ETH addresses we want to filter state for, if it has any addresses then the super-node will only send state leafs (accounts) corresponding to those account addresses. By default the super-node only sends along state leafs, if we want to receive branch and extension nodes as well intermediateNodes can be set to true.

subscription.storageFilter has four sub-options: off, addresses, storageKeys, and intermediateNodes. Setting off to true tells the super-node to not send any storage data to the subscriber; addresses is a string array which can be filled with ETH addresses we want to filter storage for, if it has any addresses then the super-node will only send storage nodes from the storage tries at those state addresses. storageKeys is another string array that can be filled with storage keys we want to filter storage data for. It is important to note that the storageKeys are the actual keccak256 hashes, whereas the addresses in the addresses fields are the ETH addresses and not their keccak256 hashes that serve as the actual state keys. By default the super-node only sends along storage leafs, if we want to receive branch and extension nodes as well intermediateNodes can be set to true.