Custom Transformers

When the capabilities of the generic contractWatcher are not sufficient, custom transformers tailored to a specific purpose can be leveraged.

Individual custom transformers can be composed together from any number of external repositories and executed as a single process using the compose and execute commands or the composeAndExecute command. This is accomplished by generating a Go plugin which allows the vulcanizedb binary to link to the external transformers, so long as they abide by one of the standard interfaces.

Writing custom transformers

For help with writing different types of custom transformers please see below:

Storage Transformers: transform data derived from contract storage tries

Event Transformers: transform data derived from Ethereum log events

Contract Transformers: transform data derived from Ethereum log events and use it to poll public contract methods

Preparing custom transformers to work as part of a plugin

To plug in an external transformer we need to:

Create a package that exports a variable TransformerInitializer, StorageTransformerInitializer, or ContractTransformerInitializer that are of type TransformerInitializer or StorageTransformerInitializer, or ContractTransformerInitializer, respectively
Design the transformers to work in the context of their event, storage, or contract watcher execution modes
Create db migrations to run against vulcanizeDB so that we can store the transformer output
- Do not goose fix the transformer migrations, this is to ensure they are always ran after the core vulcanizedb migrations which are kept in their fixed form
- Specify migration locations for each transformer in the config with the exporter.transformer.migrations fields
- If the base vDB migrations occupy this path as well, they need to be in their goose fixed form as they are here

To update a plugin repository with changes to the core vulcanizedb repository, use your dependency manager to install the desired version of vDB.

Building and Running Custom Transformers

Commands

The compose, execute, composeAndExecute commands require Go 1.11+ and use [Go plugins](https://golang .org/pkg/plugin/) which only work on Unix-based systems.
There is an ongoing conflict between Go plugins and the use of vendored dependencies which imposes certain limitations on how the plugins are built.
Separate compose and execute commands allow pre-building and linking to the pre-built .so file. So, if these are run independently, instead of using composeAndExecute, a couple of things need to be considered:
- It is necessary that the .so file was built with the same exact dependencies that are present in the execution environment, i.e. we need to compose and execute the plugin .so file with the same exact version of vulcanizeDB.
- The plugin migrations are run during the plugin's composition. As such, if execute is used to run a prebuilt .so in a different environment than the one it was composed in, then the database structure will need to be loaded into the environment's Postgres database. This can either be done by manually loading the plugin's schema into Postgres, or by manually running the plugin's migrations.
The compose and composeAndExecute commands assume you are in the vulcanizdb directory located at your system's $GOPATH, and that the plugin dependencies are present at their $GOPATH directories.
The execute command does not require the plugin transformer dependencies be located in their $GOPATH directories, instead it expects a .so file (of the name specified in the config file) to be in $GOPATH/src/github.com/vulcanize/vulcanizedb/plugins/ and, as noted above, also expects the plugin db migrations to have already been ran against the database.
Usage:
- compose: ./vulcanizedb compose --config=environments/config_name.toml
- execute: ./vulcanizedb execute --config=environments/config_name.toml
- composeAndExecute: ./vulcanizedb composeAndExecute --config=environments/config_name.toml

Flags

The execute and composeAndExecute commands can be passed optional flags to specify the operation of the watchers:

--recheck-headers/-r - specifies whether to re-check headers for events after the header has already been queried for watched logs. Can be useful for redundancy if you suspect that your node is not always returning all desired logs on every query. Argument is expected to be a boolean: e.g. -r=true. Defaults to false.
query-recheck-interval/-q - specifies interval for re-checking storage diffs that haven been queued for later processing (by default, the storage watched queues storage diffs if transformer execution fails, on the assumption that subsequent data derived from the event transformers may enable us to decode storage keys that we don't recognize right now). Argument is expected to be a duration (integer measured in nanoseconds): e.g. -q=10m30s (for 10 minute, 30 second intervals). Defaults to 5m (5 minutes).

Configuration

A .toml config file is specified when executing the commands. The config provides information for composing a set of transformers from external repositories:

[database]
    name     = "vulcanize_public"
    hostname = "localhost"
    user     = "vulcanize"
    password = "vulcanize"
    port     = 5432

[client]
    ipcPath  = "/Users/user/Library/Ethereum/geth.ipc"
    wsPath   = "ws://127.0.0.1:8546"

[exporter]
    home     = "github.com/vulcanize/vulcanizedb"
    name     = "exampleTransformerExporter"
    save     = false
    transformerNames = [
        "transformer1",
        "transformer2",
        "transformer3",
        "transformer4",
    ]
    [exporter.transformer1]
        path = "path/to/transformer1"
        type = "eth_event"
        repository = "github.com/account/repo"
        migrations = "db/migrations"
        rank = "0"
    [exporter.transformer2]
        path = "path/to/transformer2"
        type = "eth_contract"
        repository = "github.com/account/repo"
        migrations = "db/migrations"
        rank = "0"
    [exporter.transformer3]
        path = "path/to/transformer3"
        type = "eth_event"
        repository = "github.com/account/repo"
        migrations = "db/migrations"
        rank = "0"
    [exporter.transformer4]
        path = "path/to/transformer4"
        type = "eth_storage"
        repository = "github.com/account2/repo2"
        migrations = "to/db/migrations"
        rank = "1"

home is the name of the package you are building the plugin for, in most cases this is github.com/vulcanize/vulcanizedb
name is the name used for the plugin files (.so and .go)
save indicates whether or not the user wants to save the .go file instead of removing it after .so compilation. Sometimes useful for debugging/trouble-shooting purposes.
transformerNames is the list of the names of the transformers we are composing together, so we know how to access their submaps in the exporter map
exporter.<transformerName>s are the sub-mappings containing config info for the transformers
- repository is the path for the repository which contains the transformer and its TransformerInitializer
- path is the relative path from repository to the transformer's TransformerInitializer directory (initializer package).
  - Transformer repositories need to be cloned into the user's $GOPATH (go get)
- type is the type of the transformer; indicating which type of watcher it works with (for now, there are only two options: eth_event and eth_storage)
  - eth_storage indicates the transformer works with the storage watcher that fetches state and storage diffs from an ETH node (instead of, for example, from IPFS)
  - eth_event indicates the transformer works with the event watcher that fetches event logs from an ETH node
  - eth_contract indicates the transformer works with the contract watcher that is made to work with contract_watcher pkg based transformers which work with either a header or full sync vDB to watch events and poll public methods (example1, example2)
- migrations is the relative path from repository to the db migrations directory for the transformer
- rank determines the order that migrations are ran, with lower ranked migrations running first
  - this is to help isolate any potential conflicts between transformer migrations
  - start at "0"
  - use strings
  - don't leave gaps
  - transformers with identical migrations/migration paths should share the same rank
Note: If any of the imported transformers need additional config variables those need to be included as well
Note: If the storage transformers are processing storage diffs from geth, we need to configure the websocket endpoint client.wsPath for them

This information is used to write and build a Go plugin which exports the configured transformers. These transformers are loaded onto their specified watchers and executed.

Transformers of different types can be run together in the same command using a single config file or in separate instances using different config files

The general structure of a plugin .go file, and what we would see built with the above config is shown below

package main

import (
	interface1 "github.com/vulcanize/vulcanizedb/libraries/shared/transformer"
	transformer1 "github.com/account/repo/path/to/transformer1"
	transformer2 "github.com/account/repo/path/to/transformer2"
	transformer3 "github.com/account/repo/path/to/transformer3"
	transformer4 "github.com/account2/repo2/path/to/transformer4"
)

type exporter string

var Exporter exporter

func (e exporter) Export() []interface1.EventTransformerInitializer, []interface1.StorageTransformerInitializer, []interface1.ContractTransformerInitializer {
	return []interface1.TransformerInitializer{
            transformer1.TransformerInitializer,
            transformer3.TransformerInitializer,
        },     []interface1.StorageTransformerInitializer{
            transformer4.StorageTransformerInitializer,
        },     []interface1.ContractTransformerInitializer{
            transformer2.TransformerInitializer,
        }
}

Storage backfilling

Storage transformers stream data from a geth subscription or parity csv file where the storage diffs are produced and emitted as the full sync progresses. If the transformers have missed consuming a range of diffs due to lag in the startup of the processes or due to misalignment of the sync, we can configure our storage transformers to backfill missing diffs from a modified archival geth client.

To do so, add the following field to the config file.

[storageBackFill]
    on = false

on is set to true to turn the backfill process on

This process uses the regular client.ipcPath rpc path, it assumes that it is either an http or ipc path that supports the StateDiffAt endpoint.

12 KiB Raw Blame History