Struct CheckpointConfig 

pub struct CheckpointConfig {
    pub interval: Option<Duration>,
    pub max_retained: usize,
    pub alignment_timeout: Duration,
    pub pre_commit_timeout: Duration,
    pub persist_timeout: Duration,
    pub commit_timeout: Duration,
    pub incremental: bool,
    pub state_inline_threshold: usize,
    pub max_checkpoint_bytes: Option<usize>,
    pub max_pending_changelog_entries: usize,
    pub adaptive: Option<AdaptiveCheckpointConfig>,
    pub unaligned: Option<UnalignedCheckpointConfig>,
}

Unified checkpoint configuration.

Fields

interval: Option<Duration>

Interval between checkpoints. None = manual only.

max_retained: usize

Maximum number of retained checkpoints.

alignment_timeout: Duration

Maximum time to wait for barrier alignment at fan-in nodes.

pre_commit_timeout: Duration

Maximum time to wait for all sinks to pre-commit.

A stuck sink will not block checkpointing indefinitely. Defaults to 30 seconds.

persist_timeout: Duration

Maximum time to wait for manifest persist (filesystem I/O).

A hung or degraded filesystem will not stall the runtime indefinitely. Defaults to 120 seconds.

commit_timeout: Duration

Maximum time to wait for all sinks to commit (phase 2).

Defaults to 60 seconds.

incremental: bool

Whether to use incremental checkpoints.

state_inline_threshold: usize

Maximum operator state size (bytes) to inline in the JSON manifest.

States larger than this threshold are written to a state.bin sidecar file and referenced by offset/length in the manifest. This avoids inflating the manifest with base64-encoded blobs (~33% overhead).

Default: 1 MB (1_048_576). Set to usize::MAX to inline everything.

max_checkpoint_bytes: Option<usize>

Maximum total checkpoint size in bytes (manifest + sidecar).

If the packed operator state exceeds this limit, the checkpoint is rejected with [LDB-6014] and not persisted. This prevents unbounded state from creating multi-GB sidecar files.

None means no limit (default). A warning is logged at 80% of the cap.

max_pending_changelog_entries: usize

Maximum pending changelog entries per drainer before forced clear.

On checkpoint failure, clear_pending() is normally skipped. If failures repeat, pending entries grow unboundedly → OOM. This cap is a safety valve: if any drainer exceeds it after a failure, all drainers are cleared and an [LDB-6010] warning is logged.

Default: 10,000,000 entries.

adaptive: Option<AdaptiveCheckpointConfig>

Adaptive checkpoint interval configuration.

When Some, the coordinator dynamically adjusts the checkpoint interval based on observed checkpoint durations using an exponential moving average. When None (default), the static interval is used unchanged.

unaligned: Option<UnalignedCheckpointConfig>

Unaligned checkpoint configuration.

When Some, the coordinator can fall back to unaligned checkpoints when barrier alignment takes too long under backpressure. When None (default), only aligned checkpoints are used.

Trait Implementations

impl Clone for CheckpointConfig

fn clone(&self) -> CheckpointConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CheckpointConfig

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

Formats the value using the given formatter.

impl Default for CheckpointConfig

fn default() -> Self

Returns the “default value” for a type.