Skip to main content

Runtime

The main code can be found in src/runtime

note: Last written for 31b6111c

Current Status Overview

Currently, the system_program has been implemented along with basic test cases. For now, it is probably best to comence the implementation of other native programs rather than focusing on rigorous testing of the system_program. A Github issue has been opened to address the need for additional system program unit testing.

The current plan moving forward is to reveiw and merge system program and related context PR and begin the implementation of both the vote (@dadepo) and bpf loader (@yewman) programs. Once the vote and bpf loader programs are implemented, we will consider re-prioritising the implementation of the transaction processing pipeline over more program implementations in order to facilitate @dadepo's work on consensus which may require producing a set of 'state' changes for a given sequence of vote transactions.

Current Scope

Currently, this document is scoped to the logic contained within the execute_loaded_transaction method in Agave. Among other things, this includes:

Data Structures and Modules

Account Shared Data​

  • AccountSharedData holds account information with a shared reference to the account data field
  • AccountSharedData's are loaded from accounts_db during the transaction loading phase
  • It should be moved to accounts_db at in the future

Borrowed Account​

  • BorrowedAccount represents an account which has been 'borrowed' from the TransactionContext
  • It contains the accounts Pubkey, a mutable reference to an AccountSharedData with an associated single threaded write guard, and a borrow_context which represents the context under which account was borrowed
  • The borrow_context is an unamed struct which contains:
    • program_id: Pubkey: the program which borrowed the account
    • is_writable: bool: whether the account is writable within the program instruction which borrowed the account
  • BorrowedAccount provides methods for accessing and modifying account state with necessary checks

InstructionContext​

  • InstructionContext handles all state required for executing a single program instruction
  • Functionality is limited to only support the execution of SystemProgramInstruction's and will evolve as more programs are implemented
  • It defines the program_id of currently executing instruction, an array of InstructionAccountInfo's which contain account meta data, and the instruction which is the serialized program instruction
  • It provides methods for borrowing accounts from the TransactionContext, loading sysvars from the SysvarCache, and performing checks during program execution

TransactionContext​

  • TransactionContext handles all state required for executing a transaction
  • Functionality is limited to only providing access to data required during the execution of a single instruction. In time, functionality will be extended to executing multiple instructions
  • It has an array of TransactionAccount's which contain the account Pubkey, and AccountSharedData, as well as constructs for basic single threaded read/write locking of the AccountSharedData
  • These accounts are borrowed by programs during execution in order to perform state changes
  • For convenience, it contains dependencies that are required for transaction execution but should ultimately be located in a broader context. For example:
    • sysvar_cache
    • lamports_per_signature
    • prev_blockhash
    • feature_set

Feature Set​

  • FeatureSet is used to perform inference on currently active features during program execution
  • It should exist above the TransactionContext, however, it is defined here for convenience at present
  • Its implementation is trivial and does not include any feature definitions yet

Ids​

  • ids module defines system id's for programs, sysvars and other reserved accounts
  • The id defenitions are re-exported from relevant data structures where necessary
  • Perhaps they do not all need to be defined in one location, however, it is convenient for reference

Log Collector​

  • LogCollector is used to collect logs at the transaction level
  • Each TransactionContext has its own log collector which may be used to collect and emit logs as part of the transaction processing result

Nonce​

  • nonce implements types for nonce accounts
  • It probably belongs somewhere other than src/runtime/nonce.zig, perhaps in an SDK of some sort when we get to it

Pubkey Utils​

  • pubkey_utils defines the createWithSeed method which creates a Pubkey from a given base, seed, and owner
  • It returns a PubkeyError on failure which is set as a custom error in the TransactionContext when failure occures during program execution
  • We may consider moving this logic to src/core/pubkey.zig

SysvarCache​

  • SysvarCache provides the runtime with access to sysvars during program execution
  • Currently its implementation is trivial, and only serves to facilitate an equivalent implementation of Agave's get_sysvar_with_account_check module

Tmp Utils​

  • tmp_utils currently only redefines the hashv method which is currently defined in src/ledger/shred.zig
  • Rather than import from this location, the current approach was chosen to emphasize the need to extract this method to an appropriate module
  • The extraction is not performed as part of the current PR to simplify review, keeping refactor and feature implementations separate