Migrating to a UTxO-HD node as a user

This is a checklist of modifications needed to be performed by node users when migrating to UTxO-HD. It aims to ease the transition for node 10.4.

Install the LMDB library

As the Consensus layer now allows for the use of the LMDB backend, the development library needs to be present on the system. If using Nix, this library will be available in the Nix shell. Otherwise, it will have to be manually installed:

Debian/Ubuntu

sudo apt-get update -y
sudo apt-get install liblmdb-dev -y

Fedora, RedHat or CentOS

sudo yum update -y
sudo yum install lmdb -y

macOS

brew install lmdb

Windows MSYS2

pacman -S mingw-w64-<env>-x86_64-lmdb

Which for the CLANG64 environment will be:

pacman -S mingw-w64-clang-x86_64-lmdb

Populate the new configuration options

The shape of the configuration file was slightly altered to put together all the options related to the LedgerDB into a JSON object "LedgerDB". The shape of this object is as follows:

{
  ..., // other top level configuration options
  "LedgerDB": {
    "SnapshotInterval":   INT, //in seconds
    "NumOfDiskSnapshots": INT,
    "QueryBatchSize":     INT,
    "Backend":            <backend>
  }
}

Notice that "SnapshotInterval" and "NumOfDiskSnapshots" have moved from the top-level to the "LedgerDB" object. The option "DoDiskSnapshotChecksum" has been deleted as snapshots now include the metadata unconditionally.

By default, "SnapshotInterval" is set to $2 k = 4320$ seconds, and "NumOfDiskSnapshots" is set to 2.

"QueryBatchSize" sets the size of the batches in which the store will be read when querying the store for a big range of UTxOs (such as with QueryUTxOByAddress). By default we read the store in batches of 100,000 values.

The value of <backend> depends on your choice of backend for UTxO-HD. It can be one of these:

In memory

{
  ...,
  "LedgerDB": {
    ...,
    "Backend": "V2InMemory"
  }
}

LMDB

{
  ...,
  "LedgerDB": {
    ...,
    "Backend":        "V1LMDB",
    "FlushFrequency": INT,
    "MapSize":        INT,  // number of GBs
    "LiveTablesPath": PATH
  }
}

The explanation for the new options applicable only to this backend follows:

• "FlushFrequency": The number of immutable blocks that need to exists before we flush the sequence of differences to the disk. Smaller implies more frequent disk writing, bigger implies more retained memory and slower consulting of values. By default it is set to 100 blocks.
• "MapSize": LMDB needs to be given the desired mapsize with which the database is open. See the LMDB docs. By default we set this value to 16, which is fine on Linux as the file progressively grows. On Windows this will write a 16GB file directly on the disk. This is something that LMDB does, out of our control.
• "LiveTablesPath": LMDB will use a live database on the file system to store the live UTxO set. Taking snapshots will "copy" this database to the snapshot path. Therefore this "live" database is not needed for starting a node (as the one from the snapshot will be used instead), but it will be used by the running node process. It is recommended that this path is on SSD storage. The default is mainnet/db/lmdb.

Convert the existing Ledger snapshots with snapshot-converter

warning

Backup your ledger state snapshots before converting them to be safe in the unlikely case that something fails during conversion.

The format of the Ledger snapshots in the ChainDB has changed with UTxO-HD. The simplest way to use the new format is deleting your snapshots and replaying the chain from Genesis, however this can take some hours.

We provide the snapshot-converter tool which can load a snapshot in the Legacy format and write it either in the in-memory or on-disk formats for UTxO-HD. Supposing you have copied a legacy snapshot to <tmpdir>/snapshots/<slotno> you can run the following command:

In memory

$ cd <tmpdir>
$ snapshot-converter Legacy ./snapshots/<slotno> Mem ./snapshots/<slotno>_mem cardano --config <path-to-config.json>
$ rm -rf <dbDir>/ledger/*
$ cp -r ./snapshots/<slotno>_mem <dbDir>/ledger/<slotno>

LMDB

$ cd <tmpdir>
$ snapshot-converter Legacy ./snapshots/<slotno> LMDB ./snapshots/<slotno>_lmdb cardano --config <path-to-config.json>
$ rm -rf <dbDir>/ledger/*
$ cp -r ./snapshots/<slotno>_lmdb <dbDir>/ledger/<slotno>

info

You can do this process directly in the snapshots directory in the database path, but you have to make sure you delete the old format snapshots before running the node as otherwise you would have two snapshots with the same slot number and the node would be free to pick the one with the old format again. Doing the conversion in a separate directory and then copying only the new format snapshot prevents this situation.

This will create a snapshot with the proper format for UTxO-HD. See the UTxO-HD in depth document for more information on the contents of the different files in the snapshot.

If the snapshot does not have the expected structure or the backend for which it was created is different from the backend used by the node, the node will issue a warning mentioning the snapshot-converter and will ignore such snapshot.