Summarise a garry.task_log CSV.
garry_task_report.RdSummarises the task log CSV the distributed scheduler writes when
the garry.task_log option is set (see garry_options(); schema
time,event,key,pool,slot,mb,store_mb,ready): per-stage task counts
and run/queue-wait quantiles, maximum concurrency, the drain vs
host-tail split, and the peak measured (per-daemon anonymous RSS) vs
modelled (in-flight + resident) memory, with the fleet peak split by
daemon role. Logs that carry them also report the calling process's
peak anonymous RSS, the binding memory cgroup's peak usage against
its limit, peak /dev/shm use, and the tightest compute admission
budget. It answers "where did the time and memory go" for a
distributed run.
Value
Invisibly, a list: events (event counts), tasks (one row
per launch/done pair: key, pool, slot, mb, store_mb, wait_s,
queue_s, run_s, harvest_s), stages (per-stage counts, p50/p95 run
and wait seconds, p50 queue and harvest seconds), pools (per-pool
task counts, summed and p50 run seconds over every run, and the last
run's daemon count and utilisation, its busy seconds over daemons x
the run's span), max_concurrency,
drain_s, host_tail_s,
peak_model_mb, peak_rss_mb, peak_rss_by_role (named MB),
peak_host_mb, peak_cgroup_mb, cgroup_limit_mb, peak_shm_mb
and min_comp_budget_mb (NA when the log has no such rows).
Printed as a cli summary.