upgrade.md

October 15, 2023 · View on GitHub

This document describes the steps to upgrade a node running go-opera v1.1.1 to the go-opera v1.1.3-rc.5

Update go-opera version

Stop the node

  • Make sure to stop the go-opera node first and that the node is not restarted before processing the next steps
pkill opera

Update and build go-opera

cd go-opera/
git checkout release/1.1.3-rc.5
make
  • Confirm your go-opera version
./build/opera version
Go-Opera
Version: 1.1.3-rc.5

Migration flags in 1.1.3-rc.5

There's two migration options for upgrading to 1.1.3-rc.5:

  • Preserving legacy DBs layout: --db.migration.mode reformat --db.preset legacy-ldb. This option is instant but involves no performance improvement for blocks/events processing.
  • Rebuilding DBs according to a new layout: --db.migration.mode rebuild --db.preset X, where X is a selected DBs layout. Migration will take a lot of time - for an NVMe SSD drive, the number of hours would be roughly datadir size in GB divided by 100 GB/hour. If the remaining disk space is insufficient to store a copy of largest DB in datadir/chaindata, then the migration will take up to 3 times longer, depending on available disk space.

Flags --db.migration.mode and --db.preset are mandatory only for first launch.

There are 2 options for the new --db.preset flag:

  • ldb-1 - fastest LevelDB layout (default option).
  • pbl-1 - fastest Pebble layout.

Pebble offers slightly better performance and smaller IO utilization on average, especially with extended cache. We don't recommend using pbl-1 or legacy-pbl layouts for validators or any other mission-critical nodes.

Recommendations

Migrate an existing node

To migrate from an existing opera version (1.1.1-rc.2), please use:

  • --db.migration.mode reformat --db.preset legacy-ldb

This applies to all types of nodes (including API, rpc and validator nodes). Reformatting will finish almost instantly.

Sync a new node from scratch

Validator nodes and critical nodes

You can use: --db.preset ldb-1
You don't need --db.migration.mode flag, if you don’t migrate an existing DB.

Non-critical nodes (including rpc, api and tracing)

You can use either:

  • --db.preset ldb-1 (This flag will sync slightly faster, but may respond slower on multiple parallel API queries)
  • --db.preset pbl-1 (This flag will sync longer, but have more stable API response time).

Start the read-only node

Follow Launching go-opera readonly node and substitute migration flags above.

Start as a validator

Follow Launching go-opera validator node and substitute migration flags above.

Cache size

Regarding cache size, you may check your node available memory size and consider changing the default Opera cache size to a bigger value based on that. For example: --cache 16000 for a node with more than 16GB of RAM. There is an imposed limit of 60% of the available RAM of your machine, if cache size is specified larger than that value.