Docker Compose for ZKsync Era RPC "external"/zklink/treasure node.
This repo supports mainnet only. It would need to be adjusted for testnet.
cp default.env .env, then nano .env and adjust values, including NODE_DOCKER_TAG, SQL_NODE and PG_SNAPSHOT. PG_SNAPSHOT
should be a URL and will be downloaded.
If PG_SNAPSHOT is omitted, the node will sync from a network snapshot. This is fast and takes very little storage,
but does not have any historical data.
If PRUNING is set to true, the node will keep data only for PRUNING_RETENTION seconds. This means it won't
have historical data past that point.
This repo supports splitting PostgreSQL and era node to two different machines, see SQL_NODE.
Note its important to use a unique value for SQL_NODE even if running on single machine to allow running multiple versions of the repo on same machine.
Meant to be used with central-proxy-docker for traefik
and Prometheus remote write; use :ext-network.yml in COMPOSE_FILE inside .env in that case.
Note there is a memory leak in the Era node when Prometheus metrics are configured but not scraped. Unset
METRICS_PORT if you are not going to scrape metrics.
If you want the RPC ports exposed locally, use rpc-shared.yml in COMPOSE_FILE inside .env.
The ./erad script can be used as a quick-start:
./erad install brings in docker-ce, if you don't have Docker installed already.
cp default.env .env
nano .env and adjust variables as needed
./erad up
Do not restart the machine or the stack until the initial load has been completed, which takes ~ 2 days.
To update the software, run ./erad update and then ./erad up
Seriously consider splitting the era node (RocksDB) and the PostgreSQL server, if you have a limit to your storage.
Storage use climbs by around 350 GiB/month.
March 2024: RocksDB ~2.4 TiB, PostgreSQL without call_traces ~4 TiB.
June 20th 2024: RocksDB 3.2 TB (~3 TiB), PostgreSQL without call_traces 5.8 TB, ~5.4 TiB.
Nov 14th 2024: RocksDB 3.5TB (~3.3 TiB), PostgreSQL without call_traces 6.3 TB, ~5.9 TiB
The storage requirements depend on whether the node is configured to prune, and are roughly:
- 40GB + ~5GB/day of retained data of disk space for RocksDB / the node
- 300GB + ~15GB/day of retained data of disk space for PostgreSQL
Keep in mind the initial PostgreSQL snapshot load will require roughly twice the space of the DB.
Provision 6 or 8 cores and 64 GiB RAM. Initial snapshot load takes up to 60 GiB RAM with ZKsync Era node and PostgreSQL, and up to 4 cores. Steady state takes ~ 37 GiB RAM and ~ 1/5th of a core when 64 GiB are available.
When using a separate machine for PostgreSQL, 32 GiB RAM, 4 or 6 cores and 14 TiB+ NVMe work well.
The era node and PostgreSQL can be run on separate machines. Ideally, they can communicate over a private network.
On the one running the era node with era.yml, adjust the SQL_NODE variable to the IP of the PostgreSQL server.
On the one running PostgreSQL with psql.yml:psql-shared.yml, adjust PG_PARAMS for better performance and set
SHARE_IP to the IP of the interface on the private network.
If you don't have a private network, place ufw in front of Docker and allow tcp 5432 from the trusted IP of the era node, disallow all other access to tcp 5432.
Both of this are very similar to era external node but different enough to require their own yaml files. Use zklink.yml and treasure.yml for zklink and treasure respectively in place of era.yml.
Era Docker uses semver versioning. First digit breaking changes, second digit non-breaking changes and additions, third digit bugfixes.
This is Era Docker v4.1.0