Skip to contents

Plot the full history of every logged run, including reruns

Source: R/ga_autometric_runs.R
ga_autometric_history.Rd

lifecycle-experimental

Draw a Gantt-style timeline of every process run recorded in an autometric log, faceted by pipeline session. Each (pid, phase) execution becomes a horizontal segment from its start to its end time, colored by mean CPU usage.

A log usually accumulates several pipeline executions that can be days or weeks apart, so plotting them on one absolute time axis would squeeze every run into an invisible sliver. Runs are therefore grouped into sessions (consecutive runs separated by a gap larger than session_gap) and drawn in their own facet with a session-relative time axis. A target rebuilt in several sessions appears in several facets, which exposes the trial-and-error and re-run cost that the "last run only" views (ga_targets_meta_plot(), ga_autometric_lastrun()) hide by design. This full-history view is the closest to greenAlgoR's original motivation of accounting for the whole development cost of a pipeline, not just its final build.

Usage

ga_autometric_history(
  log,
  units_time = "hours",
  units_memory = "gigabytes",
  session_gap = 1,
  min_duration = 0,
  ...
)

Arguments

log

Either a path to an autometric log file (read with autometric::log_read()) or a data frame already returned by autometric::log_read() (with at least the columns pid, phase, time, cpu).

units_time, units_memory

Passed to autometric::log_read() when log is a file path (defaults "hours" and "gigabytes").

session_gap

(numeric, default 1) Gap, expressed in units_time, above which two consecutive runs are considered to belong to different pipeline sessions (facets). Increase it to merge runs into fewer sessions, decrease it to split more finely. Set to Inf to force a single session.

min_duration

(numeric, default 0) Drop runs shorter than this duration (in units_time). Useful to hide near-instant phases and keep the figure legible.

...

Additional arguments forwarded to autometric::log_read().

Value

A ggplot object: one facet per session, each with one segment per logged run.

See also

ga_autometric_lastrun(), ga_targets_meta_plot(), ga_autometric_plot()

Author

Adrien Taudière

Examples

if (requireNamespace("autometric", quietly = TRUE)) {
  log_df <- data.frame(
    version = "0.1.2",
    phase = c(rep("raw", 4), rep("scaled", 4), rep("raw", 4)),
    pid = c(rep(1L, 4), rep(2L, 4), rep(3L, 4)),
    name = "local",
    status = 0L,
    # Two sessions: the first at t ~ 0 h, a "raw" rerun ~ 5 h later.
    time = c(0, 0.1, 0.2, 0.3, 0.3, 0.35, 0.4, 0.45, 5.0, 5.1, 5.2, 5.3),
    core = runif(12, 0, 40),
    cpu = runif(12, 0, 90),
    resident = runif(12, 70, 90),
    virtual = runif(12, 900, 1000)
  )
  ga_autometric_history(log_df)
}