Investigate how to use AutoAgora

[Web3Slimchance]

https://github.com/semiotic-ai/autoagora

An Agora cost model automation tool for The Graph indexers:

• Automates the creation of relative prices for commonly seen query skeletons. (Off by default).
• Continuously tunes a per-subgraph absolute price multiplier, using continuous online reinforcement learning.

AutoAgora will continously:

• Watch for the indexer's current allocations by querying the indexer-agent's management GraphQL endpoint.
• Analyze the query logs stored in a PostgreSQL database -- logs that were previously processed by the AutoAgora indexer-service wrapper and the AutoAgora Processor.
• Gather query metrics from the indexer-service's prometheus metrics endpoint.
• Update the allocated subgraph's cost models by sending mutations to the indexer-agent's management GraphQL endpoint.

Dependencies

For AutoAgora to function correctly, you will also need to set up:

• AutoAgora indexer-service
• AutoAgora Processor

Build

docker build -t autoagora .

https://github.com/semiotic-ai/autoagora-indexer-service

A wrapper around the indexer-service that captures and filters its logs.

The point of this is to capture "query timing" logs from the indexer-service and send them to the AutoAgora Processor through RabbitMQ. All the other logs are sent back to stdout.

Works only with indexer-service v0.20.1+, which generates the logs that are needed for AutoAgora (see https://github.com/graphprotocol/indexer/pull/428).

Build

docker build --build-arg "INDEXER_SERVICE_TAG=v0.20.3" -t autoagora-indexer-service:v0.1.2-0.20.3 .

https://github.com/semiotic-ai/autoagora-processor

Processes raw indexer-service logs from a RabbitMQ queue, saves the results to a PostgreSQL database.

The processing:

• Extracts details about query execution form the log lines: The subgraph deployment, timestamp, query, variables, execution time and fees paid.
• Normalizes the queries, separates the "query skeleton" from all values and variables.
• Saves the query skeleton (if never seen before), as well as a logs all the information about the query execution (including the variable values used).

Build

docker build -t autoagora-processor .

[alex-pakalniskis]

Notes from [MIPs Workshop] The Graph's Cost Model & AutoAgora

Zoom link: https://us06web.zoom.us/j/86758885752

Agora & AutoAgora

GraphQL query cost modeling and automation

Agora: A primer

• GraphQL query complexity
  • Execution time spans 3 orders of magnitude (Uniswap example)
• Queries are very repetitive
  • most queries an indexer receives are similar

Indexer components and the gateway (only showing flow of queries)

• End user talks to the gateway (pre-pays for queries with GRT)
• The Gateway selects an indexer for each query
  • selection is a function of economic security and other quality/pricing metrics
• Query + payment data from gateway goes through indexer service. Proxies query to query node to actually process the query.
• Graph-node gets its data from a postgres database

Indexer service serves a cost model to the gateway

• gateway polls indexer about every thirty seconds

Agora domain specific language

• used to write the cost models
• similar to graphql queries with addition of costs added (see presentation slides)

query {
    swaps(first: $first) {
        id
        transaction
    }
}  => 0.042 + ..... (more cost info goes here)

Agora cost modeling

1. Run query-node with GRAPH_LOG_QUERY_TIMING=qgl + Capture logs
2. Run the logs through github.com/graphprotocol/qlog for analysis + generate .jsonl log files
3. Write Agora models for each subgraph
4. Try the Agora models using the .jsonl files + gihub.com/graphprotocol/agora

Writing Agora models is hard and repetitive :(

default => 0.00001;

AutoAgora

• Automating the building of relative cost models for frequent queries
• How prices are adjusted in real-time to current market conditions