perf: defer per-super-step read_channels deep clone in non-streaming mode - #31

Merged

Onelevenvy merged 15 commits into

Onelevenvy:mainfrom

MuFengMuXue:feat/lazy-read-channels

Aug 11, 2026

Commits on Aug 11, 2026

perf: defer per-super-step read_channels deep clone in non-streaming mode
MuFengMuXue
committed
079d53e
perf: wire updated_channels candidate-set through prepare_next_tasks
MuFengMuXue
committed
485d891
chore: ignore AGENTS.md
MuFengMuXue
committed
255d4f8
perf: eliminate per-superstep O(#channels) scans in apply_writes
MuFengMuXue
committed
e6490dd
perf: cache pregel_nodes/trigger_to_nodes on CompiledStateGraph
MuFengMuXue
committed
42bb230
perf: run tasks concurrently via JoinSet in PregelRunner
MuFengMuXue
committed
cbb0b6c
perf: eliminate per-step checkpoint serde round-trip and sync/async bridge
MuFengMuXue
committed
5b9843e
perf: emit delta-only channel_values from build_checkpoint
MuFengMuXue
committed
26c7909
checkpoint: add BlobCache, a version-keyed cache of deserialized blobs
MuFengMuXue
committed
41d6d29
checkpoint-sqlite: cache deserialized blobs on the read path
MuFengMuXue
committed
ba629c4
bench: assert static-context integrity and isolate BlobCache read wins
MuFengMuXue
committed
76bfd5e
style: run rustfmt on bench_pregel
MuFengMuXue
committed
2fbea25
perf: cache a process-global Tokio runtime for sync invoke() entry points
MuFengMuXue
committed
3b3cf79
perf: drop per-task config clone in PregelRunner by mutating task.config in place
MuFengMuXue
committed
16fb2a1
perf: drain task write buffers in apply_writes instead of cloning
MuFengMuXue
committed
f1f3d64