After calling Recover(), an infinite loop occurs.

[KazuhiroNomura]

Context: I am testing Tsavorite 2.0-beta (from NuGet) . I'm using Blittable structs for Key and Value.
Issue: I am encountering an infinite loop during Recover() at ClearBitsOnPage (where AllocatedSize becomes 0). I suspect an inconsistency between SpanByteAllocator and the checkpoint metadata (info.dat.0).
Questions:
Is there an official sample code for 2.0-beta that demonstrates the correct sequence of TakeFullCheckpoint and Recover for Blittable types?
Specifically, how should IStoreFunctions be implemented to ensure LogScan correctly identifies record boundaries during recovery in v2?
Will the full source code for the 2.0-beta NuGet package be merged into the main Tsavorite/Garnet repository soon?
namespace Tsavorite.core{ internal sealed unsafe class ObjectAllocatorImpl<TStoreFunctions> : AllocatorBase<TStoreFunctions, ObjectAllocator<TStoreFunctions>> private void AsyncReadPageWithObjectsCallback<TContext>(uint errorCode, uint numBytes, object context){ while (recordAddress < endAddress){ recordAddress += logRecord.AllocatedSize;//←AllocatedSize==0
namespace Tsavorite.core{ public partial class TsavoriteKV<TStoreFunctions, TAllocator> : TsavoriteBase private unsafe void ClearBitsOnPage(long page, RecoveryOptions options) { while (recordOffset < untilLogicalAddressInPage){ long recordSize = logRecord.AllocatedSize;//←AllocatedSize==0

[TedHartMS]

Thanks for the report! I've made fix for this since the initial beta release.

[TedHartMS]

This fix is in yesterday's beta.2 drop

[KazuhiroNomura]

Subject: Infinite loop still persists in 2.0.0-beta.2

Thank you for the update. I have tested this with the Microsoft.Garnet 2.0.0-beta.2 NuGet package, but unfortunately, the infinite loop during Recover() still occurs.

As shown in the attached debugger screenshot:

In ClearBitsOnPage, logRecord.AllocatedSize returns 0.

Consequently, recordOffset never increments, causing the while (recordOffset < endOffset) loop to run indefinitely.

Environment Details:

Package: Microsoft.Garnet 2.0.0-beta.2

Key/Value types: Blittable structs

Observation: The Debug.Assert(recordSize > 0) at line 1254 seems to be skipped or not triggered in the release/beta build, and the logic falls into an infinite loop.

Is it possible that the fix didn't cover the specific case for SpanByteAllocator with Blittable structs, or is there a specific IStoreFunctions configuration required to avoid AllocatedSize being zero during recovery?

I would appreciate any further insights or a potential fix for this.