indexer-agent: Improve batch action preparation

[fordN]

Background

Multiple actions are executed in a single transaction using the multicall paradigm, so the call data for each action is prepared and then combined into a single transaction. The call data for each action in the batch are prepared concurrently for efficiency; however, there was some innefficiencies in the concurrent preparation of call data due to redundancy in data used between the functions and a naive approach being used to validate that a deployment is healthy.

Improvements

Share data

Instead of refetching the same data during the preparation step for each and every action in the batch let's just fetch the data once and share it across all of them. To do this I've created a new context interface:

export interface TransactionPreparationContext {
  activeAllocations: Allocation[]
  currentEpoch: BigNumber
  indexingStatuses: IndexingStatus[]
}

I've updated the prepareTransactions() function to create an instance of TransactionPreparationContext and pass that into each call of prepareTransaction() (for each action).

More robust and efficient validation of healthy deployment syncing

Another efficiency and robustness improvement made in this PR is specific to preparing an allocate action for execution. In the legacy code GraphNode.ensure() was called to deploy the subgraph to the graph-node. The idempotent nature of ensure() was leaned on heavily here as it was used in this case to validate that the graph-node is already syncing that deployment. Instead of using ensure() we now access the indexingStatuses data to check if the deployment is already syncing and is healthy. This change should also improve robustness when allocating because separating the deployment process from the allocation process should allow more time between deploy and allocate for syncing errors to be caught.

[hopeyen]

lgtm!