Struct QueuedDemand 

pub struct QueuedDemand { /* private fields */ }

Queued jobs, per repository, each carrying the runs-on it requires.

This is a count of jobs, not of runs, and it is not filtered by any policy’s routing labels — but it carries everything needed to filter it. Both facts are the module documentation’s subject and are repeated on the type because this is what a caller holds.

The unfiltered totals (QueuedDemand::total, QueuedDemand::for_repository) are the raw queue depth, which is what g2 renders for an operator: “what is waiting in this repository”, independent of which host could serve it. The number e1 clamps is not either of those. It comes from tallying QueuedDemand::jobs_for against one policy’s routing labels, and the difference between the two is exactly the jobs this host cannot serve.

A count can be short in two different ways, and both have to say so

A repository can fail to answer at all (QueuedDemand::unavailable), and a repository can answer with a number that is only a floor (QueuedDemand::truncated) — more runs were active than MAX_QUEUED_RUNS_PER_REPOSITORY_PER_POLL and MAX_IN_PROGRESS_RUNS_PER_REPOSITORY_PER_POLL allow resolving, or one run’s job listing walked to MAX_JOB_PAGES_PER_RUN. QueuedDemand::is_complete is false for either.

The shape mirrors crate::rest::ActivityCount deliberately, down to the method names, because g2 renders the two side by side and a caller that has learned one should not have to learn the other. They stay separate types for the reason c3 keeps the busy-runner count and the in-progress count apart: a type that can hold either number is a type that will eventually add them.

Implementations

impl QueuedDemand

pub fn new(per_repository: BTreeMap<OwnerRepo, Vec<RunsOn>>) -> Self

pub fn of(repository: OwnerRepo, jobs: impl IntoIterator<Item = RunsOn>) -> Self

One repository’s queued jobs, for the repository-target case and for tests.

pub fn with_truncated(self, repository: OwnerRepo) -> Self

Record that repository’s count is a floor rather than a total.

pub fn with_unavailable( self, repository: OwnerRepo, reason: impl Into<String>, ) -> Self

Record that repository could not be read at all, and why.

Deliberately not the same as a count of zero: nothing was learned about it, and a zero would render a possibly-busy repository as idle.

pub fn total(&self) -> u32

Queued jobs across every repository that answered, unfiltered.

The raw queue depth, not the demand any one policy should serve. See the type documentation for the distinction, which is the whole reason QueuedDemand::jobs exists beside this.

Saturating rather than wrapping: a total wider than a u32 is not a number to wrap around zero, and u32::MAX runners is refused by the capacity ceilings long before it means anything.

pub fn per_repository(&self) -> &BTreeMap<OwnerRepo, Vec<RunsOn>>

pub fn for_repository(&self, repository: &OwnerRepo) -> Option<u32>

One repository’s unfiltered count, or None when it did not answer.

pub fn jobs_for(&self, repository: &OwnerRepo) -> &[RunsOn]

One repository’s queued jobs, as the runs-on each requires.

This is the input b1’s predicate was written for and never had. An empty slice for a repository that answered means it really is idle; a repository that did not answer is in QueuedDemand::unavailable instead, and the two must not be conflated.

pub fn jobs(&self) -> impl Iterator<Item = &RunsOn>

Every queued job across every repository that answered.

What an organization target tallies, for the reason a repository target tallies QueuedDemand::jobs_for: one policy watching an organization serves any repository in it, so its demand is the whole scope’s.

pub fn unavailable(&self) -> &[UnavailableRepository]

pub fn truncated(&self) -> &BTreeSet<OwnerRepo>

pub fn is_truncated(&self, repository: &OwnerRepo) -> bool

pub fn is_complete(&self) -> bool

Whether every repository in scope answered with an exact count.

false means the total is a floor. Scaling up from a floor is safe; concluding “idle” from one is not, which is the mistake this exists to make expressible.

Trait Implementations

impl Clone for QueuedDemand

fn clone(&self) -> QueuedDemand

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for QueuedDemand

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for QueuedDemand

fn default() -> QueuedDemand

Returns the “default value” for a type.

impl Eq for QueuedDemand

impl PartialEq for QueuedDemand

fn eq(&self, other: &QueuedDemand) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for QueuedDemand