[Milestone] Quality of Samples & Connection to Trace

[matejminar]

Problem

The current trace correlation experience brings up samples that are somewhat "random" or at least not the most useful. We have an opportunity to improve on:

• the quality/relevance of the samples we show
• the way they are visualised in the chart
• the way they are displayed in the table
• the connection of samples to different time-series (multiple queries or group by)

Potential solutions

Chart:

• Have a good distribution of samples around the line, to have good coverage
• Display less samples (maximum 10-15, TBD), but ensure they are useful
  • Always cover the spike/dip case
  • if there’s an arrow, in span card show +100% or x times bigger than average (to show how much of outlier it is)
• Have samples also for count and sum (just an even spread of samples), to ensure consistency
• Connect sample on chart with table on hover
• [strechy] Lo-fi preview trace

Samples table:

• Give users the ability to filter samples that are shown to them, keeping the same date range and bounds. Also give the possibility to choose the ordering in a user friendly way (e.g., slowest, fastest)
• Revise table columns or make them configurable
• Make hover cards the same on chart and samples table
• When you do max and min, sort also the table by default
• Samples roll-up to the trace or root-span
• Improve the empty state of the samples table - CTA to enable tracing or troubleshoot
• Highlight specific samples (for example when coming from trace view)

Trace view:

• Have icon to span with metrics when you’re coming from metrics page

[ale-cota]

Jesse's analysis: https://www.notion.so/sentry/UX-Correlating-Samples-to-Metrics-a5e9b1ccba5f4c3983f19fde9cd5b19a