Contents

visualize$execution$performance

visualize execution performance as ggplot figures

Renders the execution dashboard's performance charts as live ggplot2 objects from one execution's <exec_id>.jsonl event stream. These are the same themed figures the dashboard embeds, returned to R so they can be restyled, composed, and saved with ggplot2::ggsave().

Usage

visualize.execution.performance(
  execution = NULL,
  metric = NULL,
  by = "time",
  flip = FALSE,
  color = "resource"
)

Arguments

Argument	Description
`execution`	A run result object, a character scalar, or NULL (default). Which execution to plot: NULL takes the most recent stream in the workspace execution directory; a run result object (e.g. the PGSLibrary returned by generate$models()) is resolved through its provenance; a character scalar is an execution id (e.g. "20260627-204722-0001"), a path to a .jsonl stream, or a directory holding one or more streams (the most recent is used). Aborts when no stream resolves, and when the resolved stream holds no completed task.
`metric`	One of "occupancy" (core usage against the allocated pool), "concurrency" (running-task parallelism), "packing" (task bin-packing over the pool), "memfootprint", "throughput", "queue", "completions", "walltime", "peakmem", "corecost", or NULL (default). NULL returns every chart that has data, and makes by, flip and color inert.
`by`	One of "time" (default), over elapsed run time, or "rule", aggregated per rule. Aborts when the metric has no view for it.
`flip`	Logical scalar, default FALSE. TRUE flips the axes to horizontal rules; read only when by = "rule".
`color`	One of "resource" (default), the resource computed, or "rule", the pipeline rule. Tints the task rectangles; read only when metric = "packing".

Value

With metric = NULL, a named list of ggplot objects keyed by the dashboard's export id ("walltime-rule", "concurrency-time", ...) in .pg.perf.builders() order; a chart with no data is omitted, so the list may be empty. With metric set, one ggplot, or an abort when that chart has no data. Each element composes with + like any other ggplot; the list itself does not.

Details

The reducer reads whatever the stream holds at call time, so a still-running execution yields a partial picture rather than an error. Not every metric/by pair exists -- occupancy, packing, throughput and queue depth are run-global and have only a "time" view; asking for another names the views that do exist.

Examples

models <- generate$models(sources = sources, algorithms = algorithms)
plots <- visualize$execution$performance(models)
names(plots)
visualize$execution$performance(models, metric = "walltime", by = "rule")