Closed cthulhu-rider closed 7 months ago
Not clear why we need it
tried to explain in the commit. This is 100%-needed functionality for any storage IMO - just get what u store as []byte
increased test coverage. Fixed old bug on the way
We need to have some broader context for this. It looks like a beginning of a custom memory allocator which just shouldn't be done. We can mmap things, we can stream things, we can pool some chunks (but it doesn't work well for objects of different sizes unfortunately). If nothing else works in-place allocation is an option too (which we had for years anyway, just in a bit different form), especially given that it will be more effective already (one flat buffer, no decoding).
ok lets use make only for now
dropped options
we need object streaming RPC for a correct optimization
two different topics. Here we cover redundant decoding, while ur talking about smaller bufferization (which is also good to be done, but unrelated to the current PR/issue)
two different topics
Not different to me, the main idea is the same.
ur talking about smaller bufferization
Im talking about less memory consumption and/or allocations number. RPC that streams an object with partial reading may improve important things by orders of magnitude, while reading the whole object into memory and transferring it as a single message should only make it a few times better.
Codecov Report
Attention: Patch coverage is
79.05759%
with40 lines
in your changes are missing coverage. Please review.Additional details and impacted files
```diff @@ Coverage Diff @@ ## master #2753 +/- ## ========================================== + Coverage 21.89% 22.03% +0.14% ========================================== Files 787 787 Lines 46643 46763 +120 ========================================== + Hits 10211 10304 +93 - Misses 35555 35576 +21 - Partials 877 883 +6 ```:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.