Skip to content

ci.balance

Predict each check platform's wall at one to several shards; propose the counts.

Reads the metrics series the gate job writes; CI is its only writer, a local run reads. Per platform, a runner and a Python, the newest ten rows whose check jobs ran the full gate and succeeded give the medians of four measures: F, the setup and teardown the forge timed around the call; O, the summed wall of the gate steps that are not test steps, which the first shard alone runs; S, the summed test time over every package; and p, the parallelism the test step reached, S over its wall. The shards of a platform fold into one reading. The first shard's slice is lighter by the tests O holds back, so the wall at n shards is predicted as F + max(O, (S / p + O) / n) and the runner seconds as about n F + S / p + O; the table prints both for every count up to the cap, [ci] shards-max or --max under it.

The proposal: the slowest platform, the largest wall at one shard, takes the largest count whose every added shard saved at least F; every other platform takes the smallest count whose wall is within 15% of the slowest platform's, or the cap, named. The pull request counts follow the same rule over [ci] pull-request-runners when that list is shorter than [ci] runners. The proposal prints as the TOML [ci] shards and [ci] pull-request-shards would hold, with the predicted wall of the slowest leg beside its wall at one shard. A platform with fewer than three such rows keeps its count and is named. A store that cannot be read prints why and proposes nothing.

Without a flag the verb changes nothing. --write rewrites the two keys in the root workshop.toml when a count differs from the contract's and the predicted wall at the proposed counts differs from the wall at the contract's counts by more than 10%, so a count does not move back and forth; every comment and every other key stay, a count of one is left out of a table, a table that would be empty is removed, and [ci] shards-max is read and never written. It prints what each key becomes, or why nothing was written: no platform with enough rows, no count changed, or a wall that moved less. The workflows follow at the next fm sync, which the last line says. --submit writes the same change on the day's branch, chore/balance-<date>, renders the workflows, commits it as chore(ci): the check legs run as <counts> and submits it unarmed, for a person to merge; it runs from origin's default branch alone and says so otherwise, and a day whose branch origin already holds at these counts submits nothing and says so. The nightly runs it once a day. A count can also be set by hand.

fm ci.balance [--max=INT] [--write] [--submit]
Parameter Type Default Description
--max=INT int 0 the largest shard count a platform may take; 0 reads [ci] shards-max, 4 when the contract is silent
--write flag false rewrite [ci] shards and [ci] pull-request-shards when due
--submit flag false write, commit on the day's branch and open its pull request, unarmed (implies --write)

Example

$ fm ci.balance --write