Expand description
Node logging (blueprint §9.1’s log_{error..trace} / log_with_fields).
A node’s logs go two places at once, and both matter:
- Over the wire, as
kind = Logframes carrying anastrs_wire::LogRecord, soastrs logsand the TUI see them interleaved with every other node’s by HLC order (§13). - Into
tracing, so a node that already instruments itself withtracingkeeps its local subscriber, its spans, and itsRUST_LOG.
Sending only over the wire would make a node undebuggable without a daemon; emitting only locally would make a dataflow undebuggable without ssh. Doing both costs one extra macro call per record.
§Stamping
Every record carries this node’s HLC reading, its node id and its dataflow
id before it leaves. That is what lets the merger in astrs-log order
records from machines whose wall clocks disagree.
§A dropped log is not an error
Logging must never take a node down, and a node that cannot log is usually
a node whose daemon has already gone. The log_* methods therefore return
(); Node::try_log is there for the rare caller that wants to know.