Latest Results
perf(vortex-file): bypass async I/O pipeline for in-memory buffer reads
Add BufferSegmentSource that resolves segment requests synchronously by
slicing a ByteBuffer directly, and use it in open_buffer() instead of
routing through FileSegmentSource.
Previously, open_buffer() delegated to open_read() which created a
FileSegmentSource backed by the full async I/O pipeline: mpsc channels,
IoRequestStream coalescing, buffer_unordered concurrency limiter, and
oneshot callbacks. For an in-memory or mmap'd ByteBuffer, the actual
data access is a zero-copy slice (~0µs), but each segment request paid
the async pipeline overhead.
With BufferSegmentSource, each request() returns future::ready() with
the sliced buffer. No channels, no coalescing, no callbacks.
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com> Active Branches
#6804-11%
#68100%
#68130%
© 2026 CodSpeed Technology