Skip to content

fix: prevent auto-embedding model cache leaks and key collisions #205

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
glookka merged 1 commit into from
Sep 6, 2026
Merged

fix: prevent auto-embedding model cache leaks and key collisions #205

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
90 changes: 69 additions & 21 deletions knn/embeddings.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@

#include <unordered_map>
#include <algorithm>
#include <mutex>

#ifdef WIN32
#define WIN32_LEAN_AND_MEAN
Expand Down Expand Up @@ -72,6 +73,8 @@ bool LoadFunc ( T & pFunc, void * pHandle, const char * szFunc, const std::strin

using GetLibFuncs_fn = const EmbedLib * (*)();

using ModelRef_t = std::shared_ptr<void>;

class LoadedLib_c
{
public:
Expand All @@ -80,14 +83,17 @@ class LoadedLib_c

bool Initialize ( std::string & sError );

void AddModel ( const std::string & sKey, TextModelWrapper pModel ) { m_hModels.insert ( { sKey, pModel } ); }
TextModelWrapper GetModel ( const std::string & sKey ) const;
ModelRef_t AcquireModel ( const std::string & sKey );
ModelRef_t AddModel ( const std::string & sKey, TextModelWrapper pModel );
const EmbedLib * GetLibFuncs() const { return m_pLibFuncs; }

private:
void FreeModel ( TextModelWrapper pModel );

std::string m_sLibPath;
void * m_pHandle = nullptr;
std::unordered_map<std::string, TextModelWrapper> m_hModels;
std::mutex m_tModelsMutex;
std::unordered_map<std::string, std::weak_ptr<void>> m_hModels;
const EmbedLib * m_pLibFuncs = nullptr;
};

Expand Down Expand Up @@ -121,21 +127,52 @@ LoadedLib_c::~LoadedLib_c()
if ( !m_pHandle )
return;

if ( m_pLibFuncs )
for ( auto i : m_hModels )
dlclose(m_pHandle);
}


ModelRef_t LoadedLib_c::AcquireModel ( const std::string & sKey )
{
std::lock_guard<std::mutex> tLock(m_tModelsMutex);
const auto & tFound = m_hModels.find(sKey);
if ( tFound==m_hModels.end() )
return {};

auto pModel = tFound->second.lock();
if ( !pModel )
m_hModels.erase(tFound);

return pModel;
}


ModelRef_t LoadedLib_c::AddModel ( const std::string & sKey, TextModelWrapper pModel )
{
ModelRef_t pNewModel ( pModel, [this] ( void * pValue ) { FreeModel(pValue); } );

ModelRef_t pExistingModel;
{
std::lock_guard<std::mutex> tLock(m_tModelsMutex);
auto & pCachedModel = m_hModels[sKey];
pExistingModel = pCachedModel.lock();
if ( !pExistingModel )
{
TextModelResult tResult = { i.second };
m_pLibFuncs->free_model_result(tResult);
pCachedModel = pNewModel;
return pNewModel;
}
}

dlclose(m_pHandle);
return pExistingModel;
}


TextModelWrapper LoadedLib_c::GetModel ( const std::string & sKey ) const
void LoadedLib_c::FreeModel ( TextModelWrapper pModel )
{
const auto & tFound = m_hModels.find(sKey);
return tFound==m_hModels.end() ? nullptr : tFound->second;
if ( !pModel || !m_pLibFuncs )
return;

TextModelResult tResult = { pModel };
m_pLibFuncs->free_model_result(tResult);
}

///////////////////////////////////////////////////////////////////////////////
Expand Down Expand Up @@ -175,16 +212,24 @@ bool EmbeddingsLib_c::Load ( std::string & sError )

///////////////////////////////////////////////////////////////////////////////

static void AddKeyPart ( std::string & sKey, const std::string & sValue )
{
sKey += std::to_string ( sValue.length() );
sKey += ':';
sKey += sValue;
}


std::string ToKey ( const ModelSettings_t & tSettings )
{
std::string sKey;
sKey += tSettings.m_sModelName;
sKey += tSettings.m_sCachePath;
sKey += tSettings.m_sAPIKey;
sKey += tSettings.m_sAPIUrl;
sKey += std::to_string ( tSettings.m_iAPITimeout );
sKey += std::to_string ( tSettings.m_bUseGPU ? 1 : 0 );
sKey += std::to_string ( tSettings.m_iMaxInputTokens );
AddKeyPart ( sKey, tSettings.m_sModelName );
AddKeyPart ( sKey, tSettings.m_sCachePath );
AddKeyPart ( sKey, tSettings.m_sAPIKey );
AddKeyPart ( sKey, tSettings.m_sAPIUrl );
AddKeyPart ( sKey, std::to_string ( tSettings.m_iAPITimeout ) );
AddKeyPart ( sKey, std::to_string ( tSettings.m_bUseGPU ? 1 : 0 ) );
AddKeyPart ( sKey, std::to_string ( tSettings.m_iMaxInputTokens ) );
return sKey;
}

Expand All @@ -202,6 +247,7 @@ class TextToEmbeddings_c : public TextToEmbeddings_i
private:
ModelSettings_t m_tSettings;
std::shared_ptr<LoadedLib_c> m_pLib;
ModelRef_t m_pModelRef;
TextModelWrapper m_pModel = nullptr;
};

Expand All @@ -211,7 +257,9 @@ bool TextToEmbeddings_c::Initialize ( std::shared_ptr<LoadedLib_c> pLib, std::st
assert ( !m_pModel && pLib );

m_pLib = pLib;
m_pModel = m_pLib->GetModel ( ToKey(m_tSettings) );
const std::string sKey = ToKey(m_tSettings);
m_pModelRef = m_pLib->AcquireModel(sKey);
m_pModel = m_pModelRef.get();
if ( m_pModel )
return true;

Expand Down Expand Up @@ -250,8 +298,8 @@ bool TextToEmbeddings_c::Initialize ( std::shared_ptr<LoadedLib_c> pLib, std::st
}
}

m_pModel = tResult.m_pModel;
m_pLib->AddModel ( ToKey(m_tSettings), m_pModel );
m_pModelRef = m_pLib->AddModel ( sKey, tResult.m_pModel );
m_pModel = m_pModelRef.get();
return true;
}

Expand Down