Struct MaintenanceReport 

pub struct MaintenanceReport {
    pub time_spent: Duration,
    pub completed_merges: Vec<PathBuf>,
    pub rows_merged: u64,
    pub bytes_written: u64,
    pub pending_compaction: bool,
    pub dropped_whole: Vec<PathBuf>,
    pub direct_stream_partitions: u64,
    pub pre_pass_time: Duration,
    pub pre_pass_short_circuited: bool,
}

Maintenance report from a maintenance_step() call (M5.2, Issue #384)

Fields

time_spent: Duration

Time spent in this maintenance step

completed_merges: Vec<PathBuf>

Completed merge output files (if any merge completed)

rows_merged: u64

Number of rows merged in this step

bytes_written: u64

Number of bytes written in this step

pending_compaction: bool

Whether there is pending compaction work

dropped_whole: Vec<PathBuf>

SSTables DROPPED WHOLE by the fully-expired fast path in the merge that completed this step (issue #1388), distinct from the merged inputs: each was proven fully expired by authoritative Statistics.db metadata and overlap-safe, so it was excluded from the K-way merger (never read/decoded) and its components were reclaimed after the merged output published. Empty when nothing was dropped. Paths are input Data.db paths.

direct_stream_partitions: u64

Number of partitions written via the issue #2299 direct-stream path (rows fed straight into the writer session with no whole-partition buffer). Taken only when authoritative Statistics.db proves no input carries a deletion. Exposed so tests can positively assert the memory-bounding path actually executed (byte-identity alone holds on both the direct and buffered paths, so it cannot distinguish them) — roborev job 1723. 0 when every partition used the buffered path.

pre_pass_time: Duration

Wall-clock time spent in the dropped-column survivor pre-pass during THIS step (issue #1667, budget honesty). Non-zero only on the step that STARTS a merge for a table with dropped columns — the pre-pass (compute_surviving_dropped_columns) is a FULL, one-shot merge scan over every input that runs before any partition is emitted and is NOT bounded by budget (making it incremental is deferred to Q5). It is nonetheless counted inside time_spent (the budget Instant starts before start_merge), and when the pre-pass alone exhausts the budget the partition loop is short-circuited for this step so the pre-pass cannot silently precede an unbudgeted partition loop. Duration::ZERO on every step that does not run the pre-pass (no dropped columns, or a step that resumes an already-started merge). Always <= time_spent.

pre_pass_short_circuited: bool

true when this step OPENED a fresh merge whose dropped-column survivor pre-pass (pre_pass_time) ALONE met or exceeded the budget, so the partition loop was SKIPPED for this step and the merge left pending (issue #1667). This makes the budget-honesty guarantee observable: an over-budget pre-pass can never silently precede a partition pass in the same step. false on every step that ran (or resumed) the partition loop, on a table with no dropped columns (no pre-pass runs, so #1668’s mid-partition pause still governs a tiny-budget first step), and on a freshly-started merge whose pre-pass fit inside the budget.

Trait Implementations

impl Clone for MaintenanceReport

fn clone(&self) -> MaintenanceReport

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MaintenanceReport

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

Formats the value using the given formatter.