Content

This PR includes a rework of the retrieval of the network configuration parameters in the mithril-aggregator in order to prepare their decentralization.

Details

Mithril-aggregator

Changes on both leader and follower:

• Rework EpochService to use the MithrilNetworkConfigurationProvider system from internal/mithril-protocol-config
  • inform_epoch now fetch its current/next/registration AggregatorEpochSettings from the configuration provider
  • rework how future epoch settings are inserted in db : instead of doing it at an offset of +2 in a update_epoch_settings that is run by the state machine after running inform_new_epoch, it's now done within inform_epoch right after retrieving the parameters from the configuration provider and at an offset of +1. This means that the parameters are now stored at the last moment right before their usage instead of ahead of time.
• Epoch setting store: use an insert or ignore instead of an insert or replace when storing the epoch settings. Now stored epoch settings are considered final, that's why we moved the time of their insertion to right before their usage.
• Rework handle_discrepancies_at_startup with the aim of making it ready for decentralization:
  • Source its data from the MithrilNetworkConfigurationProvider instead of the aggregator configuration
  • Last change + store change implies that data are now registered for the work epoch window (-1, 0, +1) instead of from -1 to +2
  • Run it later in the dependency injection: after building the ServeCommandDependenciesContainer instead of after building the epoch setting store. This changes limits its call to the serve command, previously other commands could call it even if they did not need it at all.
• Configuration:
  • protocol_parameters and cardano_transactions_signing_config are now options, but still mandatory for a leader aggregator (the missing configuration error is now handled manually instead of automatically by the config crate)
  • Add optional preload_security_parameter with a default value of 2160. Used to configure the transactions preloader instead of fetching the security parameter in the cardano_transactions_signing_config configuration.
• Single signature authenticator: log the inner error when the authentication fails, before no context were available in we could not know why an authentication failed
• Test:
  • strengthen create_certificate_follower integration test by making it check that the follower aggregator works without configured protocol_parameters
  • rework and simplify test tooling attach to the ServeCommandDependenciesContainer:
    • now longer stores data in the epoch_settings table, but only in the signer and signer_registration table. This means that the epoch_settings must have been filled beforehand (either by running handle discrepancies or manually).
    • update usages, notably in the certifier service tests
    • remove now unused methods

Leader aggregator specific:

• Add LocalMithrilNetworkConfigurationProvider: a MithrilNetworkConfigurationProvider that fetch its data first from the epoch_settings table in the sqlite database, and if an entry is missing for an epoch, it fallback to the usual configuration parameters (protocol_parameters and cardano_transactions_signing_config)

Follower aggregator specific:

• Use mithril-protocol-config::http::HttpMithrilNetworkConfigurationProvider as its network configuration provider, fetching data from its configured leader aggregator

Mithril-end-to-end

• Make update_protocol_parameters step leader only. Now it doesn't matter for the follower since it retrieve its configuration from the leader and no longer read its configuration, and keeping it introduced a flakiness since sometimes the follower aggregator restarted before the leader could restart its http server, making the handle discrepancies of the follower fails (since it do a call to the configuration provider).
• Wrap tailed logs and extracted errors in a ::group:: when running in github action. Disabled for now as there's an remaining issue to tackle first: when the e2e is retry the logs in those groups for the previous iteration are missing in the action output.

Pre-submit checklist

• Branch
  • Tests are provided (if possible)
  • Crates versions are updated (if relevant)
  • CHANGELOG file is updated (if relevant)
  • Commit sequence broadly makes sense
  • Key commits have useful messages
• PR
  • All check jobs of the CI have succeeded
  • Self-reviewed the diff
  • Useful pull request description
  • Reviewer requested
• Documentation
  • Update README file (if relevant)
  • Update documentation website (if relevant)
  • No new TODOs introduced

Comments

There's a know issue for the updated handle_discrepancies_at_startup: it run twice.
This is because the ServeCommandDependenciesContainer is also built twice, once for its purpose, a second time for the http server.
This is harmless as this will record twice the same epoch settings and the second recording will be ignored, but we should probably construct a HttpServerDependenciesContainer instead of reusing the one of the serve command.

Issue(s)

Relates to #2692