Module barrier

Modules§

BackfillNode
BackfillOrderState: Actor done -> update fragment_id state Operator done -> update downstream operator dependency Operator’s dependencies done -> queue operator for backfill
BackfillProgress 🔒
BarrierScheduler: The sender side of the barrier scheduling queue. Can be cloned and held by other managers to schedule and run barriers.
BarrierWorkerRuntimeInfoSnapshot 🔒
CreateStreamingJobCommandInfo
DatabaseRuntimeInfoSnapshot 🔒
FragmentBackfillProgress 🔒
GlobalBarrierManager
IcebergPkIndexCompactionOverwrite
ReplaceStreamJobPlan: Replacing an old job with a new one. All actors in the job will be rebuilt.
Reschedule: Reschedule describes per-fragment changes in a resolved reschedule plan, used for actor scaling or migration.
RescheduleContext: Preloaded context for rescheduling, built outside the barrier worker.
ReschedulePlan
Scheduled 🔒: Scheduled command with its notifier.
SinceEpochInfo
SnapshotBackfillInfo
TracedEpoch: A wrapper of Epoch with tracing span, used for issuing epoch-based tracing from the barrier manager on the meta service. This structure is free to clone, which’ll extend the lifetime of the underlying span.
UpdateDatabaseBarrierRequest 🔒

BarrierKind
BarrierManagerRequest 🔒
BarrierManagerStatus 🔒: Status of barrier manager.
Command: Command is the input of crate::barrier::worker::GlobalBarrierWorker. For different commands, it will build different barriers to send via corresponding *_to_mutation helpers, and may do different stuffs after the barrier is collected.
CreateStreamingJobType
IndependentStreamingJobType
RecoveryReason 🔒: The reason why the cluster is recovering.
RecoveryStatus 🔒: Recovery state exposed by the barrier manager, without completion notifiers.
ResumeBackfillTarget

snapshot_backfill_max_pending_barrier_num: The maximum number of upstream barriers a snapshot backfill partial graph may queue before the upstream is back-pressured.