What held each operation up in the most recent epoch, and where the observed client's resources go. This page is deliberately one point in time: the series page carries the history.
Bottleneck
A time budget per operation, from four independent sources: the operation's own duration, the daemon's cgroup CPU delta across exactly that operation, goroutine blocking on real dependencies, and summed postgres statement time. Whichever dominates names the class. Rows are clickable for the contributing functions and queries. The operator's function names are not published, because the lumos repository is private; its budget and verdict are.
| operation | node | verdict | wall | cpu | block | mutex | db |
|---|
Balance and vtxos are the workload's ground truth; the sqlite
footprint includes the WAL, and unbackfilled is how many WAL
frames a checkpoint has not folded into the main file yet: a figure that
only rises means checkpointing has stalled and the WAL will grow without
bound.
| client | balance | live vtxos | sqlite | unbackfilled | goroutines | go heap |
|---|
| name | flat | share | cumulative |
|---|