Skip to content

Latest commit

 

History

History
67 lines (45 loc) · 4.41 KB

README.md

File metadata and controls

67 lines (45 loc) · 4.41 KB
Raw

subgraph-cache

Purpose

This deploys a Google Cloud Function that periodically fetches records from subgraphs and stores them in Google Cloud Storage

Potential Uses

  • Expose records in BigQuery through the external tables functionality
  • Post-processing of records (e.g. to generate daily balances)

Architecture

This project has a few components:

  • Google Cloud Storage (GCS) bucket to store records (in JSONL files)
    • Files are stored in the following location: <bucket>/<object>/dt=<YYYY-MM-DD>/records.jsonl
    • Files are stored in the JSONL (newline-delimited) format in order to make it easy to ingest into BigQuery
    • The parent directory of the records.jsonl file contains dt= followed by the date, so that Hive partitioning is supported without further transformation.
  • Google Cloud Function (GCF) to fetch records from the subgraph
    • It will fetch records from the latest date present in GCS, or the finish date specified in a PubSub message (which is useful when there's a timeout).
    • If the function detects that the timeout is approaching, it will exit fetching the records.
    • Before exit, the function sends a message on the specified PubSub Topic. This includes both the start date and the finish date for the fetching.
  • Google Cloud Scheduler Job to trigger the function in GCF.
  • BigQuery table exposing the data stored in the GCS bucket.
  • Alert policies to inform about failures and extraneous executions in Discord/email.
  • Monitoring dashboard for key metrics: function executions and status, GCS network activity

The Pulumi tool is used to manage the orchestration of resources in Google Cloud Platform, and has dev and prod stacks (environments).

This project is designed to cache any Graph Protocol subgraph that is specified, effectively creating a BigQuery pipeline. See the Subgraph Configuration section for more information.

Discord Integration

GCP monitoring does not have a direct integration with Discord, and it seemed like overkill to write a GCP -> Discord webhook integration.

Instead, a scenario is defined in Make (formerly Integromat) that does the following:

  • Watches an email address supplied by the custom mailhook
  • Send a HTTP post request to the Discord webhook

This results in a small message being sent into the alerts channel.

Caveats

  • The token-holder-balances project has details of the resources in this project hard-coded into its configuration. Those values will need to be manually updated if the resource ids change (which is rare).
  • If a new subgraph version is deployed that changes historical data (such as a new token being indexed, or a different calcultion being used), this function will not (yet) detect those changes. To force re-fetching of the transactions, delete the token-holders-transactions directory in the GCS bucket. Upon the next schedule, records will be fetched automatically.
  • If specifying a BigQuery type override for a date in ISO string format (e.g. 2022-10-11T10:05.001Z), the type to specify is TIMESTAMP (not DATE).
  • After changing any of the contents of the subgraph/*.json files, run yarn codegen.
  • BigQuery does not support replacing the schema of an existing BigQuery table. If the type overrides are changed, you will need to delete the BigQuery table manually, run pulumi refresh, and then run pulumi up.

Subgraph Configuration

For each subgraph, there is a corresponding JSON file in the subgraphs/ directory. Note: the name of the file is irrelevant. This is used to:

  • generate a query to fetch all fields in the subgraph
  • generate a schema that is used in conjunction with Pulumi to create a BigQuery table

See the TokenHolderTransaction file as an example.

To cache a new subgraph, perform the following:

  • Add a new definition file in subgraphs/, adhering to the SubgraphConfig type
    • If a number is large or has a large number of decimal places, it should have a type override to be BIGNUMERIC
    • See the data types
  • Run yarn codegen to generate the GraphQL & BigQuery schemas, and the Typescript typings
  • Run pulumi up on the appropriate stack to update the resources