Skip to content

Latest commit

 

History

History

ethmonitor

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

README.md

ethmonitor

Follows an EVM chain block by block and publishes each new block, and optionally its logs, to subscribers. Reorgs come through as explicit events.

ethmonitor keeps a short canonical chain in memory. When the node's view of the chain changes, it emits Removed events for the orphaned blocks and Added events for their replacements, in order, so a subscriber can keep its own state consistent with the chain.

Usage

provider, err := ethrpc.NewProvider("https://your-node-url")
if err != nil {
	return err
}

opts := ethmonitor.DefaultOptions
opts.WithLogs = true

monitor, err := ethmonitor.NewMonitor(provider, opts)
if err != nil {
	return err
}

go func() {
	if err := monitor.Run(ctx); err != nil {
		// Run returns when ctx is cancelled, Stop is called, or on a fatal error
	}
}()
defer monitor.Stop()

sub := monitor.Subscribe("my-service")
defer sub.Unsubscribe()

for {
	select {
	case blocks := <-sub.Blocks():
		for _, block := range blocks {
			switch block.Event {
			case ethmonitor.Added:
				// a new canonical block; block.Logs is set when WithLogs is on
			case ethmonitor.Removed:
				// a reorged block: undo whatever was done for its Added event
			}
		}
	case <-sub.Done():
		return sub.Err()
	}
}

To get new heads over a websocket instead of polling, pass ethrpc.WithStreaming(wsURL) to ethrpc.NewProvider. The monitor falls back to polling if streaming keeps failing (see Options).

How it works

 new head (poll or stream)
        │
        ▼
 fetch next block ──► canonical chain ──► fetch logs ──► publish queue ──► subscribers
                      (last N blocks,     (WithLogs)     (in order,
                       reorg detection)                   ready blocks only)
  1. Fetch. The monitor fetches the next block by number.
  2. Chain. If the block's parent is the current head, it's added to the in-memory chain. If not, the monitor walks back to the common ancestor, emitting Removed for each orphaned block, then adds the new branch. The chain keeps the last BlockRetentionLimit + TrailNumBlocksBehindHead blocks, and never fewer than 10.
  3. Logs. With WithLogs, the block's logs are fetched with eth_getLogs by block hash and validated before the block counts as ready (Block.OK). See Logs.
  4. Publish. Events are queued and published in order, and publishing stops at the first block that isn't ready yet. With TrailNumBlocksBehindHead, events are held until they are that many blocks behind the head, which cuts down on Added/Removed churn on chains that reorg often.

Events

Each Blocks batch a subscription receives holds one or more *Block events, oldest first:

  • Added: the block joined the canonical chain.
  • Removed: a block that was published as Added is no longer canonical. Match it by block hash.

Blocks.Reorg() reports whether a batch contains a removal. Consumers that store anything per block must handle Removed.

Reading chain state

The monitor also answers questions about the chain it holds in memory:

  • Latest blocks:
    • LatestBlock() returns the head of the in-memory chain.
    • LatestReadyBlock() returns the newest block whose data (including logs) is complete.
    • LatestFinalBlock(n) returns the block n blocks behind the head.
  • Lookups: GetBlock(hash) and GetTransaction(hash) search the retained blocks.
  • Chain statistics: GetAverageBlockTime(), and FeeHistory(...), an approximation of eth_feeHistory computed from the retained blocks without fetching receipts. It uses gas limits and priority fees in place of gas used and effective priority fees.
  • The chain itself: Chain() gives access to it directly.

Logs

With WithLogs, every published Added block carries its logs in Block.Logs. LogTopics limits them to logs whose first topic is in the list.

How logs are validated

A node can answer eth_getLogs with [] before it has indexed a block, so an empty result isn't always trusted:

  • Most chains: if eth_getLogs returns [] but the header's logsBloom is non-zero, the result is rejected and retried, since the bloom says the block has logs.

  • Chains with settled headers: on Avalanche C-Chain after ACP-194 (Helicon), the logsBloom describes the settled blocks rather than the block itself, so it can't be used. The monitor detects these chains from the headers' settledHeight field, with no configuration needed. It confirms an empty result against the block's receipts, which a node only serves once the block has executed. If the receipts have logs, those are used.

    Note: this check only runs when LogTopics is empty. With a topic filter set, an empty result on these chains is accepted without confirmation, so a premature one can drop a block's logs. On settled-header chains, leave LogTopics empty and filter the logs yourself.

  • Receipts methods: receipts are fetched with eth_getBlockReceipts. If the node says the method doesn't exist, the monitor switches to eth_getTransactionReceipt per transaction, in batches of 20, plus eth_getBlockTransactionCountByHash. It uses that path for the rest of its lifetime.

When logs keep failing

A block whose logs can't be fetched blocks everything after it, because publishing is in order. To keep the monitor moving:

  1. Receipts fallback. After MaxLogsAttempts failed attempts (default 10), the block's logs are taken from its receipts. On chains without settled headers, empty receipts for a block with a non-zero bloom are still rejected.
  2. Leaving the chain cache. If the block is still unresolved when it leaves the chain cache (BlockRetentionLimit + TrailNumBlocksBehindHead blocks later), one more receipts attempt is made. If that fails too:
    • Default: the block is published with empty Logs and Block.LogsUnconfirmed = true. An Error is logged and the Alerter is called.
    • StrictLogs: true: the block is never published without confirmed logs. The monitor keeps retrying, and when the publish queue (2 × (BlockRetentionLimit + TrailNumBlocksBehindHead) events) overflows, Run returns ErrFatal wrapping ErrQueueFull.

Which mode to use: keep the default when a short gap is better than downtime, for example live notifications. Use StrictLogs when a missing log is worse than stopping. For example, an indexer that persists what it receives would rather restart from its last checkpoint than record a block as having no events.

Options

Start from ethmonitor.DefaultOptions and override what you need.

Option Default What it does
PollingInterval 1.5s How often to poll for new blocks.
StreamingDisabled false Poll even if the provider has a websocket URL.
StreamingErrNumToSwitchToPolling 3 Streaming errors before switching to polling.
StreamingErrorResetInterval 2m How often the streaming error count resets.
StreamingRetryAfter 5m How long to poll before retrying streaming.
StreamHealthCheckInterval 10s How often to verify the last received head is still fresh while streaming. 0 disables the check.
StreamHealthCheckStaleness 60s Max age of the streamed head before the node's head is fetched over http to decide if the subscription is stalled.
Timeout 20s Timeout for node requests.
StartBlockNumber nil (latest) Block to start from. A negative value starts that many blocks behind the latest.
Bootstrap false Start from a chain restored with Chain().BootstrapFromBlocks / BootstrapFromBlocksJSON. Takes precedence over StartBlockNumber.
TrailNumBlocksBehindHead 0 Publish only once a block is this many blocks behind the head.
BlockRetentionLimit 200 Blocks kept in the in-memory chain, on top of TrailNumBlocksBehindHead. This is also the deepest reorg the monitor can follow. Keep the sum above 5: smaller values can overflow the publish queue.
WithLogs false Fetch and publish each block's logs.
LogTopics all Only include logs whose first topic is in this list.
MaxLogsAttempts 10 Failed eth_getLogs attempts before falling back to receipts.
StrictLogs false Never publish a block without confirmed logs (see Logs).
PrefetchConcurrency 0 (off) Workers fetching blocks and logs ahead of the monitor while it trails the head. Useful on fast chains.
PrefetchWindow 4 × concurrency How far ahead prefetching may go.
CacheBackend nil Shared cache for block and log payloads. In most cases, leave it nil.
CacheExpiry 10m Lifetime of cache entries.
UnsubscribeOnStop false Close all subscriptions when Run returns.
ChainID from node Set it to skip asking the node.
Alerter no-op Called on conditions that need attention, such as a block published without confirmed logs.
Logger discard *slog.Logger for warnings and debug output.
DebugLogging false Verbose logging.

Restarts and snapshots

Chain().Snapshot() serializes the retained chain to JSON. A new monitor created with Bootstrap: true can then be restored with Chain().BootstrapFromBlocksJSON(data) (or BootstrapFromBlocks) before Run. It continues from the snapshot's head and can still follow reorgs into the restored blocks.

Errors

Run returns when its context is cancelled, when Stop is called, or on an error the monitor can't recover from. Fatal errors wrap ErrFatal. ErrQueueFull is the main one: the publish queue overflowed. In default mode that only happens for reorgs deeper than the chain cache, or when BlockRetentionLimit + TrailNumBlocksBehindHead is 5 or less.

A subscription's Done() channel closes when it ends, and Err() returns the reason.