Include loaded accounts data size in TransactionResults

[tao-stones]

Problem

Total loaded-accounts-data-size for successfully loaded transaction is accumulated during load_transaction_accounts(). It is useful information, could be included in TransactionResults that is returned from load_execute_and_commit_transactions(), so call sites can access stats of loaded accounts of sanitized transactions.

For now, such stats includes the actual loaded-accounts-data-size, count of loaded accounts.

The use of this information is planned for:

1. cost_tracker uses it to adjust reserved block space for transaction, in the same fashion as it adjust for estimated/actual execution cost. (note: loaded accounts data size contributes to transaction cost in term of CU, charging CU for loaded accounts data size #1356 )
2. replay_stage uses it to report histogram of loaded accounts data size per transaction.

Summary of Changes

• add pub struct TransactionLoadedAccountsStats
• include it into TransactionResults

Fixes #

[jstarry]

Main question is why not put this info in TransactionExecutionDetails instead of TransactionResults? For the most part, any transaction which is loaded will also be executed. It's not a big deal either way but putting this info in execution details could be easier to work with depending on what you're doing.

[tao-stones]

Main question is why not put this info in TransactionExecutionDetails instead of TransactionResults? For the most part, any transaction which is loaded will also be executed. It's not a big deal either way but putting this info in execution details could be easier to work with depending on what you're doing.

Good question. It'd be easier to ride on TransactionExecutionDetails, however there are few errors can happen after accounts loaded but NotExecuted, for example:

• When program cache full -> NotExecuted(TransactionError::ProgramCacheHitMaxLimit)
• Any error from ComputeBudget::try_from_instructions() -> NotExecuted(...)

In these cases, we would lost information about loaded accounts, that'd skew histogram data report to metrics, and make banking_stage adjusting block reserve by actual loaded-accounts-data-size-cost inaccurate (potentially?). Thought it's better to separate loaded and executed anyway.

[jstarry]

Ok sounds good, makes sense.