Disk Usage Optimization

Customize the configuration settings to lower the disk requirements for your validator node

Blockchain database tends to grow over time, depending e.g. on block speed and transaction amount. For HAQQ, we are talking about close to 100GB of disk usage in first two weeks.

There are few configurations that can be done to reduce the required disk usage quite significantly. Some of these changes take full effect only when you do the configuration and start syncing from start with them in use.

Storage Configuration Options

Set to true to discard ABCI responses from the state store, which can save a considerable amount of disk space. On config.toml setß

[storage]
discard_abci_responses = true

Indexing

If you do not need to query transactions from the specific node, you can disable indexing. On config.toml set

indexer = "null"

If you do this on already synced node, the collected index is not purged automatically, you need to delete it manually. The index is located under the database directory with name data/tx_index.db/.

Consensus Configuration Options

[consensus]
skip_timeout_commit = true

State-sync snapshots

I believe this was disabled by default on HAQQ, but listing it in any case here. On app.toml set

snapshot-interval = 0

Note that if state-sync was enabled on the network and working properly, it would allow one to sync a new node in few minutes. But this node would not have the history.

Configure pruning

By default every 500th state, and the last zero states are kept. This consumes a lot of disk space on long run, and can be optimized with following custom configuration in app.toml:

pruning = "everything"

pruning-keep-recent = "0"
pruning-interval = "0"
min-retain-blocks = 400000

API

To reduce the load, we recommend disabling all APIs in in app.toml:

[api]
enable = false

[rosetta]
enable = false

[grpc]
enable = false

[grpc-web]
enable = false

Logging

By default the logging level is set to info, and this produces a lot of logs. This log level might be good when starting up to see that the node starts syncing properly. However, after you see the syncing is going smoothly, you can lower the log level to warn (or error). On config.toml set the following

log_level = "warn"

Also ensure your log rotation is configured properly.

Results

Below is the disk usage after two weeks of HAQQ testnet. The default configuration results in disk usage of 90GB.

5.3G    ./state.db
70G     ./application.db
20K     ./snapshots/metadata.db
24K     ./snapshots
9.0G    ./blockstore.db
20K     ./evidence.db
1018M   ./cs.wal
4.7G    ./tx_index.db
90G     .

This optimized configuration has reduced the disk usage to 17 GB.

17G     .
1.1G    ./cs.wal
946M    ./application.db
20K     ./evidence.db
9.1G    ./blockstore.db
24K     ./snapshots
20K     ./snapshots/metadata.db
5.3G    ./state.db