Skip to content

Latest commit

 

History

History
150 lines (110 loc) · 6.23 KB

README.md

File metadata and controls

150 lines (110 loc) · 6.23 KB

Ledger

The ledger represents the state of an Algorand node. The core state of the ledger is a sequence of blocks (the blockchain). The ledger also maintains several trackers: state machines that consume the blockchain as input. One example of a tracker is the account state tracker, which keeps track of the state of individual accounts (balance, status, etc).

The external API for the ledger (for use by callers outside of this package) is defined in ledger.go.

Blocks

The ledger exposes the following functions for managing the blocks:

  • AddBlock(block, cert) adds a new block to the ledger, along with a certificate for that block. The ledger does not check the certificate's validity. If the block is not from the correct round (i.e., latest known plus one), an error is returned. The block is added to an in-memory queue of pending blocks, and is flushed to disk in the background for performance.

  • WaitForCommit(round) waits for the block for round to be written to persistent storage. After WaitForCommit(round) returns, the ledger guarantees that the block will be present even after a crash.

  • Latest() returns the last block added to the ledger.

  • LatestCommitted() returns the last block written to durable storage as well as the round of the latest block added to the ledger.

  • Block(round) returns the block for round, or ErrNoEntry if no such block has been added. Similarly, BlockCert(round) will return the block and the associated certificate.

  • Wait(round) allows the caller to wait for a block to be added to the ledger for round, by returning a channel that will be closed when a block for round is added.

Tracker API

The ledger comes with a set of trackers. Each tracker maintains a state machine based on the blockchain contents. Trackers are logically stateless: that is, they can reconstruct their state by consuming all blocks from the beginning of the blockchain. As an optimization, the ledger allows trackers to store persistent state, so that they can reconstruct their state quickly, without considering every block.

The interface between ledger and trackers is defined in tracker.go.

Trackers have access to the ledger through a restricted API defined by ledgerForTracker. This allows trackers to access the ledger's SQLite database, to query for blocks, etc.

Conversely, the ledger accesses trackers through the ledgerTracker interface:

  • loadFromDisk(ledgerForTracker) initializes the state of the tracker. The tracker can use the ledgerForTracker argument to load persistent state (e.g., for the accounts database). The tracker can also query for recent blocks, if the tracker's state depends only on recent blocks (e.g., for the tracker that keeps track of the recently committed transactions).

  • newBlock(rnd, delta) tells the tracker about a new block added to the ledger. delta describes the changes made by this block; this will be described in more details under block evaluation later.

  • committedUpTo(rnd) tells the tracker that all blocks up to and including rnd are written to persistent storage. This call is important for trackers that store persistent state themselves, since the tracker must be able to restore its state correctly after a crash, and may need to answer queries about older blocks after a crash if some recent non-committed blocks are lost.

  • close() frees up any resources held by this tracker.

The ledger serializes all updates to the trackers with a reader-writer lock.

Trackers

An individual tracker exposes tracker-specific APIs for accessing the state maintained by that tracker. These are currently passed through the Ledger object, so that the ledger can provide appropriate reader-writer locking.

Account tracker

  • Lookup(round, address) uses the account tracker to look up the state of an account as of round.

  • AllBalances(round) uses the account tracker to return the set of all account states as of round. This is likely to be large, so it's intended for debug purposes only.

  • Totals(round) returns the totals of accounts, using the account tracker.

Recent transactions tracker

  • Committed(txnid) returns whether txid has been recently committed, using the transaction tail tracker.

Participation tracker

  • ParticipationThresholds() returns the participation thresholds, from the participation tracker.

Delegator tracker

  • DumpDelegatorsTree() returns the tree of delegators for offline rewards, from the delegator tracker, for debug purposes.

Notification tracker

  • RegisterBlockListeners(listeners) registers listeners for new blocks, based on the pools.BlockListener API.

Block evaluation

Finally, the ledger implements logic to evaluate blocks. It supports three functions:

  • Construct a new block, based on a pool of potential transactions and rewards, that will be valid. This is done by using the Ledger.StartEvaluator(hdr, txcache) method. This returns a BlockEvaluator, which can then accept tentative transactions and rewards (using BlockEvaluator.Transaction() and BlockEvaluator.Reward()). The caller can finalize the block by calling BlockEvaluator.GenerateBlock(). txcache represents a cache of previously verified transactions, to avoid repeated checking of transaction signatures.

  • Validate a block. This is done by calling Ledger.Validate(block, txcache). Under the covers, it executes the same logic using a BlockEvaluator.

  • Evaluate a block to produce a delta describing the changes that this block implies for all of the trackers. This is the delta passed to the newBlock() method of trackers.

Block evaluation also produces auxiliary state, evalAux, which describes the state needed to evaluate a block. Currently, this consists of the set of offline rewards. Computing evalAux may require access to past blocks or the old state of various state trackers (in particular, for the offline rewards, it requires access to the past state of the delegator tracker). However, given an evalAux, it is possible to evaluate the block. The BlockEvaluator computes the evalAux when first evaluating a block, and the ledger saves the evalAux state to re-evaluate the block after a crash as needed.