Struct CheckpointManifest 

pub struct CheckpointManifest {

    pub version: u32,
    pub checkpoint_id: u64,
    pub epoch: u64,
    pub timestamp_ms: u64,
    pub source_offsets: HashMap<String, ConnectorCheckpoint>,
    pub sink_epochs: HashMap<String, u64>,
    pub sink_commit_statuses: HashMap<String, SinkCommitStatus>,
    pub table_offsets: HashMap<String, ConnectorCheckpoint>,
    pub operator_states: HashMap<String, OperatorCheckpoint>,
    pub table_store_checkpoint_path: Option<String>,
    pub wal_position: u64,
    pub per_core_wal_positions: Vec<u64>,
    pub watermark: Option<i64>,
    pub source_watermarks: HashMap<String, i64>,
    pub source_names: Vec<String>,
    pub sink_names: Vec<String>,
    pub pipeline_hash: Option<u64>,
    pub inflight_data: HashMap<String, Vec<InFlightRecord>>,
    pub size_bytes: u64,
    pub is_incremental: bool,
    pub parent_id: Option<u64>,
    pub state_checksum: Option<String>,
}

A point-in-time snapshot of all pipeline state.

This is the single source of truth for checkpoint persistence, replacing the three previously disconnected checkpoint systems:

• PipelineCheckpoint (source offsets + sink epochs)
• DagCheckpointSnapshot (operator state — in-memory only)
• CheckpointMetadata (WAL position + watermark)

Fields

version: u32

Manifest format version (for future evolution).

checkpoint_id: u64

Unique, monotonically increasing checkpoint ID.

epoch: u64

Epoch number for exactly-once coordination.

timestamp_ms: u64

Timestamp when checkpoint was created (millis since Unix epoch).

source_offsets: HashMap<String, ConnectorCheckpoint>

Per-source connector offsets (key: source name).

sink_epochs: HashMap<String, u64>

Per-sink last committed epoch (key: sink name).

sink_commit_statuses: HashMap<String, SinkCommitStatus>

Per-sink commit status (key: sink name).

Populated during the commit phase (Step 6) and saved to the manifest afterward. Recovery uses this to decide which sinks need rollback (those with SinkCommitStatus::Pending or SinkCommitStatus::Failed).

table_offsets: HashMap<String, ConnectorCheckpoint>

Per-table source offsets for reference tables (key: table name).

operator_states: HashMap<String, OperatorCheckpoint>

Per-operator checkpoint data (key: operator/node name).

Small state is inlined as base64. Large state is stored in a separate state.bin file and this map holds only a reference marker.

table_store_checkpoint_path: Option<String>

Path to the table store checkpoint, if any.

wal_position: u64

WAL position for single-writer mode.

per_core_wal_positions: Vec<u64>

Per-core WAL positions at the time of operator snapshot.

During recovery, WAL replay must start after these positions to avoid replaying entries already reflected in the operator state. Index i corresponds to the WAL segment for core i.

watermark: Option<i64>

Global watermark at checkpoint time.

source_watermarks: HashMap<String, i64>

Per-source watermarks (key: source name).

source_names: Vec<String>

Sorted names of all registered sources at checkpoint time.

Used during recovery to detect topology changes (added/removed sources) and warn the operator.

sink_names: Vec<String>

Sorted names of all registered sinks at checkpoint time.

pipeline_hash: Option<u64>

Hash of the pipeline configuration at checkpoint time.

Computed from SQL queries, source/sink configuration, and connector options. Recovery logs a warning when this changes, indicating operator state may be incompatible with the new configuration.

inflight_data: HashMap<String, Vec<InFlightRecord>>

In-flight channel data captured during unaligned checkpoints.

Key: operator name. Value: list of in-flight records from input channels. Empty for aligned checkpoints. During recovery, these events are replayed before resuming normal processing.

size_bytes: u64

Total size of all checkpoint data in bytes (manifest + state.bin).

is_incremental: bool

Whether this is an incremental checkpoint (only deltas since parent).

parent_id: Option<u64>

Parent checkpoint ID for incremental checkpoints.

state_checksum: Option<String>

SHA-256 hex digest of the sidecar state.bin file (if any).

Written during checkpoint commit so that recovery can verify the sidecar hasn’t been corrupted or truncated on disk/S3.

Implementations

impl CheckpointManifest

pub fn validate(&self) -> Vec<ManifestValidationError>

Validates manifest consistency before recovery.

Returns a list of issues found. An empty list means the manifest is valid. Callers should treat non-empty results as warnings (recovery may still proceed) or errors depending on severity.

pub fn new(checkpoint_id: u64, epoch: u64) -> Self

Creates a new manifest with the given ID and epoch.

Trait Implementations

impl Clone for CheckpointManifest

fn clone(&self) -> CheckpointManifest

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CheckpointManifest

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for CheckpointManifest

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for CheckpointManifest

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

Tests for self and other values to be equal, and is used by ==.

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

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for CheckpointManifest

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 CheckpointManifest