perf(memtrack): reduce serialization bottleneck#436

Open

not-matthias wants to merge 13 commits into

mainfrom

cod-3071-fix-serialization-bottleneck

Commits on Jul 7, 2026

perf(memtrack): avoid buffered serde event serialization
not-matthias
committed
perf(memtrack): return encoded frames from writer
not-matthias
committed
perf(memtrack): encode event batches in parallel
not-matthias
committed
perf(memtrack): buffer serializer input to fix write regression
not-matthias
committed
perf(memtrack): batch ring-buffer drain and drop forwarder threads
not-matthias
committed
refactor(memtrack): extract encode pipeline into a benchable library fn
not-matthias
committed
refactor(memtrack): drop unused Compression::None raw encode path
not-matthias
committed
perf(memtrack): default encode to fast zstd level -5
not-matthias
committed
perf(memtrack): reuse zstd compressor + scratch buffer per worker
not-matthias
committed
perf(memtrack): use all CPUs for parallel encode
not-matthias
committed
perf(memtrack): stage ring-buffer events in a thread-local, drop the per-event lock
not-matthias
committed
perf(memtrack): drain ring-buffer stragglers adaptively at shutdown
not-matthias
committed
perf(memtrack): recycle frame buffers between writer and workers
not-matthias
committed