Struct RunSummary 

pub struct RunSummary {

    pub run_id: String,
    pub export_name: String,
    pub status: String,
    pub total_rows: i64,
    pub files_produced: usize,
    pub bytes_written: u64,
    pub files_committed: usize,
    pub duration_ms: i64,
    pub peak_rss_mb: i64,
    pub retries: u32,
    pub validated: Option<bool>,
    pub schema_changed: Option<bool>,
    pub quality_passed: Option<bool>,
    pub error_message: Option<String>,
    pub tuning_profile: String,
    pub batch_size: usize,
    pub batch_size_memory_mb: Option<usize>,
    pub format: String,
    pub mode: String,
    pub compression: String,
    pub pg_temp_bytes_delta: Option<i64>,
    pub skip_reason: Option<String>,
    pub source_count: Option<i64>,
    pub reconciled: Option<bool>,
    pub manifest_parts: Vec<ManifestPart>,
    pub schema_fingerprint: Option<String>,
    pub manifest_verification: Option<ManifestVerification>,
    pub apply_context: Option<ApplyContext>,
    pub journal: RunJournal,
}

Accumulates operational data during a pipeline run for summary and metrics.

The embedded journal is the structured event log for this run. Use summary.journal.record(event) at any call site that already holds &mut RunSummary.

Fields

run_id: String

export_name: String

status: String

total_rows: i64

files_produced: usize

bytes_written: u64

files_committed: usize

Incremented after each successful dest.write(). Non-zero means a previous attempt already committed data — retrying from the same cursor would duplicate rows.

duration_ms: i64

peak_rss_mb: i64

retries: u32

validated: Option<bool>

schema_changed: Option<bool>

quality_passed: Option<bool>

error_message: Option<String>

tuning_profile: String

profile from YAML, or balanced (default) if omitted.

batch_size: usize

Configured batch_size from YAML/profile (FETCH cap before batch_size_memory_mb override).

batch_size_memory_mb: Option<usize>

When set, actual FETCH size is derived from schema (see logs).

format: String

mode: String

compression: String

pg_temp_bytes_delta: Option<i64>

Postgres pg_stat_database.temp_bytes delta around the run. None for non-Postgres sources or when the snapshot probe failed (no admin perms not required — the view is readable by any role). When set and large, indicates cursor / sort spill to pgsql_tmp/ — the safe action is to shrink tuning.batch_size or set tuning.batch_size_memory_mb below PG’s work_mem.

skip_reason: Option<String>

Human-readable parenthetical attached to status: skipped so the operator knows why there was nothing to export this run (e.g. "no new rows since cursor 'updated_at'"). Always None when status != "skipped". Surfaced in the console summary card as status: skipped (<reason>).

source_count: Option<i64>

Source COUNT(*) result for reconciliation (None = not requested or not applicable).

reconciled: Option<bool>

Whether reconciliation passed (Some(true) = match, Some(false) = mismatch, None = skipped).

manifest_parts: Vec<ManifestPart>

Committed parts accumulated during the run, in commit order. Populated by pipeline::manifest_writer::record_committed_part at each dest.write site (ADR-0012 M1 — Parts Before Manifest). Drained at finalize into a RunManifest by [crate::pipeline::manifest_writer::write_manifest].

schema_fingerprint: Option<String>

xxh3 fingerprint of the dest-facing column schema for this run, in the canonical xxh3:<16-hex> form produced by crate::state::schema_fingerprint.

Recorded by [crate::pipeline::manifest_writer::record_run_schema_fingerprint] the first time the sink has resolved a schema (i.e. on the first batch of any chunk). Idempotent within a run — the schema is identical across chunks, so later writes are no-ops.

finalize_manifest reads this directly so the manifest’s schema_fingerprint no longer depends on the per-export schema row happening to land in state before the manifest write. The state lookup remains a fallback for resume scenarios where the summary was reconstructed without ever seeing a live schema.

manifest_verification: Option<ManifestVerification>

Result of the manifest-aware --validate pass (ADR-0012 M5/M6, ADR-0013). Populated by pipeline::job::finalize_validate_manifest after finalize_manifest succeeds; None when the run targeted a streaming destination, when --validate was not requested, or when the run failed before any manifest could be written.

apply_context: Option<ApplyContext>

Apply-time context (plan_id, –force usage, bypassed checks). None when the run came from rivet run rather than rivet apply. See [ApplyContext] and finding F5 of the 0.7.5 audit.

journal: RunJournal

Structured event log for this run. Answers the four DoD observability questions.

Implementations

impl RunSummary

pub fn check_post_run_invariants(&self) -> Result<(), String>

Sanity-check the post-run summary ↔ manifest_parts coherence. Used as a debug_assert!-style runtime gate from finalize_manifest so any future runner that bumps bytes_written / files_committed / files_produced without going through pipeline::commit::record_part is caught the moment it finishes a real export. Compiled out in release builds via the cfg!(debug_assertions) guard at the call site.

Resume-safe inequalities only: on resume, manifest_parts carries prior runs’ parts via chunked::resume_m8 while bytes_written / files_committed reflect only the current invocation — so strict equality is wrong across resume boundaries. Strict equality on the non-resume path is pinned by pipeline::commit::tests.

Returns Ok(()) when the summary satisfies the invariants, else an Err(String) naming which one was violated and by how much.

Trait Implementations

impl Clone for RunSummary

fn clone(&self) -> RunSummary

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for RunSummary

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

Formats the value using the given formatter.