Performance Monitoring

CEDAR records both the server calculation and the browser-visible wait for its main reports. Open Data & Usage → Cache → Report Timing Estimates and click Refresh Timing Stats to compare report types and fresh/cache paths.

The median describes a typical run. The 90th percentile (P90) describes the slow tail and is usually the better prioritization measure for a shared Shiny process. The table separates:

Measure	Interpretation
Total	Time from the user’s request until the output is delivered and painted.
Compute	Time inside the report’s server timer.
Wait/delivery	Event-loop waiting and delivery time before the browser receives the completion message. This can rise when other sessions occupy the Shiny worker.
Browser settle	Output handling and painting after the completion message arrives.
CPU	Process CPU consumed during the server timer.
Worker growth	Change in the R worker’s resident memory across the operation.
Result size	Size of the retained R result object returned by an instrumented report.
Payload	Approximate JSON value size delivered to report outputs.
Connected sessions	Browsers connected to that R worker at the observation boundary.

Use the measures together. High compute and CPU points toward the R analysis. High total with modest compute, especially when connected sessions are high, points toward queueing or delivery contention. High browser settle and payload points toward output serialization or rendering. Large result objects and worker growth identify candidates for narrower payloads, precomputation, or cache changes.

The loading overlay learns from the most recent 100 observations for each report and cache path. Once three browser observations exist, it shows the recent median-to-P90 end-to-end range. Until then it falls back to server calculation history and finally to the configured static default. This avoids presenting a fast calculation or cache lookup as if it were the user’s full wait.

Log files

report_timing.csv contains calculation, CPU, worker memory, result size, cache, session, success, and report-parameter fields. client_render_timing.csv contains total, compute, wait/delivery, browser settle, payload, viewport, cache, and session fields. New observations share an operation_id so the two files can be joined for run-level analysis.

CEDAR migrates older schemas at startup and keeps the original file beside the new one with a .legacy suffix. Appends use a small cross-process lock so multiple Shiny workers do not migrate or write the same file simultaneously.

Worker growth is a boundary measurement for the whole R process, not a record of peak allocations. A negative value can mean that garbage collection freed memory during the report. Result size measures retained output and may differ from transient working memory. Connected sessions provide load context but do not mean every connected browser was actively running a report.

After a release, exercise at least one cold and warm path for each important report, then review P90 total, P90 compute, wait/delivery, result size, and worker memory. Compare one-session and many-session observations before deciding whether the next improvement belongs in analysis code, caching, payload shaping, or deployment capacity.