Traces

The list

A raw span list is unreadable on a phone, so the trace list uses OpenObserve’s per-trace_id roll-up endpoint, which reports the things that make a trace list scannable:

• total duration;
• span count and error count;
• the services involved, with how many spans each contributed;
• the root operation.

It polls while live updates are on, so a trace that happens while you are looking shows up.

The waterfall

Tapping a trace reads its spans back with an ordinary search on trace_id and lays them out as a waterfall: bars on one shared time axis, offset and width taken from each span’s own start and end, so you can see where the time actually went rather than reading a list of durations. Any span opens its full record, attributes included.

Re-treeing spans, safely

The spans come back as a flat list, so the parent/child tree has to be rebuilt in the app. That code is pure and unit-tested, and it is written to survive real-world damage:

• a span whose parent was never ingested becomes a root, rather than vanishing;
• a parent cycle cannot loop the layout;
• every span appears exactly once.

A trace you look at during an incident is exactly the trace most likely to be incomplete — losing spans silently would be the worst possible failure mode.