Struct Disk 

pub struct Disk {
    pub min_free_bytes: u64,
    pub auto_fold: bool,
    pub fold_grace_secs: u64,
    pub cache_limit_bytes: u64,
}

Disk hygiene: how hard magi is allowed to press on the machine’s free space.

The numbers below come from one incident, not from theory: a machine with 951.8 GB free ran a few competitions and plans and best read 6.7 GB free. Three multi-gigabyte classes of junk accumulated side by side - per-run worktrees that end as Merged/Ready/Failed, a shared build cache whose each verify round and each implementation wave recompiles the derived section of the project into, and the outputs of runs that were removed but whose folders nobody deleted.

Fields

min_free_bytes: u64

Free space, in bytes, below which no new run may start: the daemon and the magi run gate answer with “the disk is full” instead of letting the graph fill it the rest of the way. 0 turns the gate off.

Default 8 GiB. The incident ran down to 6.7 GB free of 951.8 GB total before anybody noticed; 8 GiB is enough headroom for the compile a fresh competition triggers and small enough that a 1 TB disk with 100 GB free is nowhere near the threshold.

auto_fold: bool

Fold finished runs without being asked. Merged, Ready and Failed runs older than fold_grace_secs have their worktrees removed. 0 turns the janitor off.

Default true.

fold_grace_secs: u64

How old a finished run must be before the janitor folds it, seconds.

Default 6 hours. A run that Ready at 8am is the operator’s answer; a run that Ready a week ago is worktrees holding a compile each. Six hours is long enough that nobody loses an answer in the gap between reading the report and starting from it, and short enough that a backlog cannot pile up across two nights.

cache_limit_bytes: u64

Ceiling for the shared build cache (CARGO_TARGET_DIR in the rendered verify commands), in bytes. When the janitor runs and the cache is over it, files are dropped oldest-first until it is not. 0 turns pruning off - the cache then only ever grows, which is the operator’s call.

Default 10 GiB. This is what the incident measured: 30.61 GB sat in the shared cache on top of ~16 GB in the primary target directory and 6.7- 11.15 GB in each of four per-worktree targets. 10 GiB holds a healthy stack of prebuilt dependencies (cargo’s per-file fingerprinting means pruning only costs the rebuild of the dropped files, not of the world) without letting one addled cache swallow the machine.

Trait Implementations

impl Clone for Disk

fn clone(&self) -> Disk

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Disk

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

Formats the value using the given formatter.

impl Default for Disk

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for Disk

where Disk: Default,

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for Disk

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

Equality operator ==.

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

Inequality operator !=.

impl Serialize for Disk

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for Disk