You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
feat(graph): add per-node timing and datastore-read attribution to check debug trace
Adds start_time and datastore_queries to the internal dispatch CheckDebugTrace
so a single Check / CheckBulk debug trace can answer two questions that
duration alone cannot: (1) whether sub-problems were traversed concurrently,
and (2) where time was spent blocked on datastore reads.
start_time is the wall-clock anchor for each node (the monotonic duration stays
authoritative; end = start_time + duration). A context-scoped, concurrency-safe
collector is installed per dispatch node; the observable datastore proxy records
each relationship query (shape, start, duration, row count) into it, so reads
attribute to the node that issued them. Both only activate when debug tracing is
enabled, leaving the non-debug hot path untouched.
This is the internal-proto-only slice; the public authzed/api CheckDebugTrace
fields and the internal->public conversion follow once that field lands.
Refs #3198
0 commit comments