pub struct FlushConfig {
pub max_bytes: usize,
pub max_batches: usize,
pub linger: Duration,
pub max_inflight_flushes: usize,
pub max_buffered_bytes: usize,
pub budget: BudgetConfig,
}Expand description
Flush-trigger policy for the engine’s group-commit buffer.
Fields§
§max_bytes: usizeHard ceiling on one sealed object’s payload size (memory / object-store
physics). Not the primary packing control: under normal load packing
is ingest_rate × linger. Keep this high enough that linger binds
before size (default 1 GiB). Lower only for tight RAM or backend caps.
max_batches: usizeFlush once this many batches are buffered (secondary ceiling).
linger: DurationMaximum time a batch may wait before a deadline flush (hard ceiling).
This is the latency↔throughput control surface: longer wait ⇒ more bytes
per seal ⇒ fewer durable ops/s. The budget controller may use a shorter
effective linger when media is idle (TD-004). ZERO = seal as soon as
any data is buffered (no co-buffer wait). Default 50ms.
max_inflight_flushes: usizeMax sealed objects PUT concurrently. Default 1 (single-flight bulk path). Raise for parallel S3 PUTs. Also bounds ready objects per atomic commit when the sequencer opts in to multi-object commits.
max_buffered_bytes: usizeMax bytes in the mutable queue plus in-flight seals. Producers block when exceeded. Default 2 GiB.
budget: BudgetConfigDurable-ops budget controller (default on for Fjord).