Merge pull request #24367 from vespa-engine/toregge/share-code-for-loading-and-saving-tensor-attributev8.65.41

Share code for loading and saving tensor attribute between

22 files changed, 170 insertions, 267 deletions

diff --git a/searchlib/src/tests/tensor/dense_tensor_store/dense_tensor_store_test.cpp b/searchlib/src/tests/tensor/dense_tensor_store/dense_tensor_store_test.cpp
index 149662cd266..26ef57aab65 100644
--- a/searchlib/src/tests/tensor/dense_tensor_store/dense_tensor_store_test.cpp
+++ b/searchlib/src/tests/tensor/dense_tensor_store/dense_tensor_store_test.cpp
@@ -32,15 +32,15 @@ struct Fixture
     {}
     void assertSetAndGetTensor(const TensorSpec &tensorSpec) {
         Value::UP expTensor = makeTensor(tensorSpec);
-        EntryRef ref = store.setTensor(*expTensor);
-        Value::UP actTensor = store.getTensor(ref);
+        EntryRef ref = store.store_tensor(*expTensor);
+        Value::UP actTensor = store.get_tensor(ref);
         EXPECT_EQUAL(*expTensor, *actTensor);
         assertTensorView(ref, *expTensor);
     }
     void assertEmptyTensor(const TensorSpec &tensorSpec) {
         Value::UP expTensor = makeTensor(tensorSpec);
         EntryRef ref;
-        Value::UP actTensor = store.getTensor(ref);
+        Value::UP actTensor = store.get_tensor(ref);
         EXPECT_TRUE(actTensor.get() == nullptr);
         assertTensorView(ref, *expTensor);
     }
diff --git a/searchlib/src/vespa/searchlib/tensor/CMakeLists.txt b/searchlib/src/vespa/searchlib/tensor/CMakeLists.txt
index 75f453ddcbc..0d87881711c 100644
--- a/searchlib/src/vespa/searchlib/tensor/CMakeLists.txt
+++ b/searchlib/src/vespa/searchlib/tensor/CMakeLists.txt
@@ -9,7 +9,6 @@ vespa_add_library(searchlib_tensor OBJECT
     dense_tensor_attribute_saver.cpp
     dense_tensor_store.cpp
     direct_tensor_attribute.cpp
-    direct_tensor_saver.cpp
     direct_tensor_store.cpp
     distance_calculator.cpp
     distance_function_factory.cpp
@@ -29,13 +28,13 @@ vespa_add_library(searchlib_tensor OBJECT
     nearest_neighbor_index_saver.cpp
     serialized_fast_value_attribute.cpp
     small_subspaces_buffer_type.cpp
-    streamed_value_saver.cpp
     tensor_attribute.cpp
     tensor_buffer_operations.cpp
     tensor_buffer_store.cpp
     tensor_buffer_type_mapper.cpp
     tensor_deserialize.cpp
     tensor_store.cpp
+    tensor_store_saver.cpp
     reusable_set_visited_tracker.cpp
     DEPENDS
 )
diff --git a/searchlib/src/vespa/searchlib/tensor/dense_tensor_attribute.cpp b/searchlib/src/vespa/searchlib/tensor/dense_tensor_attribute.cpp
index d4353532309..636c949be08 100644
--- a/searchlib/src/vespa/searchlib/tensor/dense_tensor_attribute.cpp
+++ b/searchlib/src/vespa/searchlib/tensor/dense_tensor_attribute.cpp
@@ -112,7 +112,7 @@ void
 DenseTensorAttribute::internal_set_tensor(DocId docid, const vespalib::eval::Value& tensor)
 {
     consider_remove_from_index(docid);
-    EntryRef ref = _denseTensorStore.setTensor(tensor);
+    EntryRef ref = _denseTensorStore.store_tensor(tensor);
     setTensorRef(docid, ref);
 }
 
@@ -229,10 +229,7 @@ DenseTensorAttribute::getTensor(DocId docId) const
     if (docId < getCommittedDocIdLimit()) {
         ref = acquire_entry_ref(docId);
     }
-    if (!ref.valid()) {
-        return {};
-    }
-    return _denseTensorStore.getTensor(ref);
+    return _denseTensorStore.get_tensor(ref);
 }
 
 vespalib::eval::TypedCells
diff --git a/searchlib/src/vespa/searchlib/tensor/dense_tensor_store.cpp b/searchlib/src/vespa/searchlib/tensor/dense_tensor_store.cpp
index 1bc84a7216d..ba7e8526146 100644
--- a/searchlib/src/vespa/searchlib/tensor/dense_tensor_store.cpp
+++ b/searchlib/src/vespa/searchlib/tensor/dense_tensor_store.cpp
@@ -11,6 +11,7 @@
 using vespalib::datastore::CompactionContext;
 using vespalib::datastore::CompactionSpec;
 using vespalib::datastore::CompactionStrategy;
+using vespalib::datastore::EntryRef;
 using vespalib::datastore::Handle;
 using vespalib::datastore::ICompactionContext;
 using vespalib::eval::CellType;
@@ -147,19 +148,8 @@ DenseTensorStore::start_compact(const CompactionStrategy& compaction_strategy)
     return std::make_unique<CompactionContext>(*this, std::move(compacting_buffers));
 }
 
-std::unique_ptr<Value>
-DenseTensorStore::getTensor(EntryRef ref) const
-{
-    if (!ref.valid()) {
-        return {};
-    }
-    vespalib::eval::TypedCells cells_ref(getRawBuffer(ref), _type.cell_type(), getNumCells());
-    return std::make_unique<vespalib::eval::DenseValueView>(_type, cells_ref);
-}
-
-template <class TensorType>
-TensorStore::EntryRef
-DenseTensorStore::setDenseTensor(const TensorType &tensor)
+EntryRef
+DenseTensorStore::store_tensor(const Value& tensor)
 {
     assert(tensor.type() == _type);
     auto cells = tensor.cells();
@@ -170,10 +160,29 @@ DenseTensorStore::setDenseTensor(const TensorType &tensor)
     return raw.ref;
 }
 
-TensorStore::EntryRef
-DenseTensorStore::setTensor(const vespalib::eval::Value &tensor)
+EntryRef
+DenseTensorStore::store_encoded_tensor(vespalib::nbostream& encoded)
+{
+    (void) encoded;
+    abort();
+}
+
+std::unique_ptr<Value>
+DenseTensorStore::get_tensor(EntryRef ref) const
+{
+    if (!ref.valid()) {
+        return {};
+    }
+    vespalib::eval::TypedCells cells_ref(getRawBuffer(ref), _type.cell_type(), getNumCells());
+    return std::make_unique<vespalib::eval::DenseValueView>(_type, cells_ref);
+}
+
+bool
+DenseTensorStore::encode_stored_tensor(EntryRef ref, vespalib::nbostream& target) const
 {
-    return setDenseTensor(tensor);
+    (void) ref;
+    (void) target;
+    abort();
 }
 
 }
diff --git a/searchlib/src/vespa/searchlib/tensor/dense_tensor_store.h b/searchlib/src/vespa/searchlib/tensor/dense_tensor_store.h
index bd83772ee55..1b25bdad464 100644
--- a/searchlib/src/vespa/searchlib/tensor/dense_tensor_store.h
+++ b/searchlib/src/vespa/searchlib/tensor/dense_tensor_store.h
@@ -51,10 +51,6 @@ private:
     BufferType _bufferType;
     ValueType _type; // type of dense tensor
     std::vector<char> _emptySpace;
-
-    template <class TensorType>
-    TensorStore::EntryRef
-    setDenseTensor(const TensorType &tensor);
 public:
     DenseTensorStore(const ValueType &type, std::shared_ptr<vespalib::alloc::MemoryAllocator> allocator);
     ~DenseTensorStore() override;
@@ -70,12 +66,15 @@ public:
     EntryRef move(EntryRef ref) override;
     vespalib::MemoryUsage update_stat(const vespalib::datastore::CompactionStrategy& compaction_strategy) override;
     std::unique_ptr<vespalib::datastore::ICompactionContext> start_compact(const vespalib::datastore::CompactionStrategy& compaction_strategy) override;
-    std::unique_ptr<vespalib::eval::Value> getTensor(EntryRef ref) const;
+    EntryRef store_tensor(const vespalib::eval::Value &tensor) override;
+    EntryRef store_encoded_tensor(vespalib::nbostream &encoded) override;
+    std::unique_ptr<vespalib::eval::Value> get_tensor(EntryRef ref) const override;
+    bool encode_stored_tensor(EntryRef ref, vespalib::nbostream &target) const override;
+
     vespalib::eval::TypedCells get_typed_cells(EntryRef ref) const {
         return vespalib::eval::TypedCells(ref.valid() ? getRawBuffer(ref) : &_emptySpace[0],
                                           _type.cell_type(), getNumCells());
     }
-    EntryRef setTensor(const vespalib::eval::Value &tensor);
     // The following method is meant to be used only for unit tests.
     uint32_t getArraySize() const { return _bufferType.getArraySize(); }
 };
diff --git a/searchlib/src/vespa/searchlib/tensor/direct_tensor_attribute.cpp b/searchlib/src/vespa/searchlib/tensor/direct_tensor_attribute.cpp
index 7730f340e01..d9fe025b4e5 100644
--- a/searchlib/src/vespa/searchlib/tensor/direct_tensor_attribute.cpp
+++ b/searchlib/src/vespa/searchlib/tensor/direct_tensor_attribute.cpp
@@ -1,16 +1,8 @@
 // Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
 #include "direct_tensor_attribute.h"
-#include "direct_tensor_saver.h"
-
 #include <vespa/eval/eval/fast_value.h>
 #include <vespa/eval/eval/value.h>
-#include <vespa/searchlib/attribute/readerbase.h>
-#include <vespa/searchlib/util/fileutil.h>
-#include <vespa/vespalib/util/array.h>
-
-#include "blob_sequence_reader.h"
-#include "tensor_deserialize.h"
 
 using vespalib::eval::FastValueBuilderFactory;
 
@@ -27,39 +19,6 @@ DirectTensorAttribute::~DirectTensorAttribute()
     _tensorStore.clearHoldLists();
 }
 
-bool
-DirectTensorAttribute::onLoad(vespalib::Executor *)
-{
-    BlobSequenceReader tensorReader(*this);
-    if (!tensorReader.hasData()) {
-        return false;
-    }
-    setCreateSerialNum(tensorReader.getCreateSerialNum());
-    assert(tensorReader.getVersion() == getVersion());
-    uint32_t numDocs = tensorReader.getDocIdLimit();
-    _refVector.reset();
-    _refVector.unsafe_reserve(numDocs);
-    vespalib::Array<char> buffer(1024);
-    for (uint32_t lid = 0; lid < numDocs; ++lid) {
-        uint32_t tensorSize = tensorReader.getNextSize();
-        if (tensorSize != 0) {
-            if (tensorSize > buffer.size()) {
-                buffer.resize(tensorSize + 1024);
-            }
-            tensorReader.readBlob(&buffer[0], tensorSize);
-            auto tensor = deserialize_tensor(&buffer[0], tensorSize);
-            EntryRef ref = _direct_store.store_tensor(std::move(tensor));
-            _refVector.push_back(AtomicEntryRef(ref));
-        } else {
-            EntryRef invalid;
-            _refVector.push_back(AtomicEntryRef(invalid));
-        }
-    }
-    setNumDocs(numDocs);
-    setCommittedDocIdLimit(numDocs);
-    return true;
-}
-
 void
 DirectTensorAttribute::set_tensor(DocId lid, std::unique_ptr<vespalib::eval::Value> tensor)
 {
@@ -129,15 +88,4 @@ DirectTensorAttribute::get_tensor_ref(DocId docId) const
     return *ptr;
 }
 
-std::unique_ptr<AttributeSaver>
-DirectTensorAttribute::onInitSave(vespalib::stringref fileName)
-{
-    vespalib::GenerationHandler::Guard guard(getGenerationHandler().takeGuard());
-    return std::make_unique<DirectTensorAttributeSaver>
-        (std::move(guard),
-         this->createAttributeHeader(fileName),
-         getRefCopy(),
-         _direct_store);
-}
-
 } // namespace
diff --git a/searchlib/src/vespa/searchlib/tensor/direct_tensor_attribute.h b/searchlib/src/vespa/searchlib/tensor/direct_tensor_attribute.h
index 2dfb5c1efcd..6466c6f7537 100644
--- a/searchlib/src/vespa/searchlib/tensor/direct_tensor_attribute.h
+++ b/searchlib/src/vespa/searchlib/tensor/direct_tensor_attribute.h
@@ -21,8 +21,6 @@ public:
                        const document::TensorUpdate &update,
                        bool create_empty_if_non_existing) override;
     std::unique_ptr<vespalib::eval::Value> getTensor(DocId docId) const override;
-    bool onLoad(vespalib::Executor *executor) override;
-    std::unique_ptr<AttributeSaver> onInitSave(vespalib::stringref fileName) override;
     void set_tensor(DocId docId, std::unique_ptr<vespalib::eval::Value> tensor);
     const vespalib::eval::Value &get_tensor_ref(DocId docId) const override;
     bool supports_get_tensor_ref() const override { return true; }
diff --git a/searchlib/src/vespa/searchlib/tensor/direct_tensor_saver.cpp b/searchlib/src/vespa/searchlib/tensor/direct_tensor_saver.cpp
deleted file mode 100644
index 0de4491cfcc..00000000000
--- a/searchlib/src/vespa/searchlib/tensor/direct_tensor_saver.cpp
+++ /dev/null
@@ -1,54 +0,0 @@
-// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
-
-#include "direct_tensor_saver.h"
-#include "direct_tensor_store.h"
-
-#include <vespa/eval/eval/value_codec.h>
-#include <vespa/searchlib/attribute/iattributesavetarget.h>
-#include <vespa/searchlib/util/bufferwriter.h>
-#include <vespa/vespalib/objects/nbostream.h>
-
-using vespalib::GenerationHandler;
-
-namespace search::tensor {
-
-DirectTensorAttributeSaver::
-DirectTensorAttributeSaver(GenerationHandler::Guard &&guard,
-                            const attribute::AttributeHeader &header,
-                            RefCopyVector &&refs,
-                            const DirectTensorStore &tensorStore)
-    : AttributeSaver(std::move(guard), header),
-      _refs(std::move(refs)),
-      _tensorStore(tensorStore)
-{
-}
-
-
-DirectTensorAttributeSaver::~DirectTensorAttributeSaver()
-{
-}
-
-bool
-DirectTensorAttributeSaver::onSave(IAttributeSaveTarget &saveTarget)
-{
-    auto datWriter = saveTarget.datWriter().allocBufferWriter();
-    const uint32_t docIdLimit(_refs.size());
-    vespalib::nbostream stream;
-    for (uint32_t lid = 0; lid < docIdLimit; ++lid) {
-        const vespalib::eval::Value *tensor = _tensorStore.get_tensor_ptr(_refs[lid]);
-        if (tensor) {
-            stream.clear();
-            encode_value(*tensor, stream);
-            uint32_t sz = stream.size();
-            datWriter->write(&sz, sizeof(sz));
-            datWriter->write(stream.peek(), stream.size());
-        } else {
-            uint32_t sz = 0;
-            datWriter->write(&sz, sizeof(sz));
-        }
-    }
-    datWriter->flush();
-    return true;
-}
-
-}  // namespace search::tensor
diff --git a/searchlib/src/vespa/searchlib/tensor/direct_tensor_saver.h b/searchlib/src/vespa/searchlib/tensor/direct_tensor_saver.h
deleted file mode 100644
index 132e1570f0f..00000000000
--- a/searchlib/src/vespa/searchlib/tensor/direct_tensor_saver.h
+++ /dev/null
@@ -1,35 +0,0 @@
-// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
-
-#pragma once
-
-#include <vespa/searchlib/attribute/attributesaver.h>
-#include "tensor_attribute.h"
-
-namespace search::tensor {
-
-class DirectTensorStore;
-
-/*
- * Class for saving a tensor attribute.
- */
-class DirectTensorAttributeSaver : public AttributeSaver
-{
-public:
-    using RefCopyVector = TensorAttribute::RefCopyVector;
-private:
-    using GenerationHandler = vespalib::GenerationHandler;
-
-    RefCopyVector _refs;
-    const DirectTensorStore &_tensorStore;
-
-    bool onSave(IAttributeSaveTarget &saveTarget) override;
-public:
-    DirectTensorAttributeSaver(GenerationHandler::Guard &&guard,
-                               const attribute::AttributeHeader &header,
-                               RefCopyVector &&refs,
-                               const DirectTensorStore &tensorStore);
-
-    virtual ~DirectTensorAttributeSaver();
-};
-
-} // namespace search::tensor
diff --git a/searchlib/src/vespa/searchlib/tensor/direct_tensor_store.cpp b/searchlib/src/vespa/searchlib/tensor/direct_tensor_store.cpp
index fba1d494690..1184cca37e7 100644
--- a/searchlib/src/vespa/searchlib/tensor/direct_tensor_store.cpp
+++ b/searchlib/src/vespa/searchlib/tensor/direct_tensor_store.cpp
@@ -1,7 +1,10 @@
 // Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
 #include "direct_tensor_store.h"
+#include "tensor_deserialize.h"
+#include <vespa/eval/eval/fast_value.h>
 #include <vespa/eval/eval/value.h>
+#include <vespa/eval/eval/value_codec.h>
 #include <vespa/vespalib/datastore/compacting_buffers.h>
 #include <vespa/vespalib/datastore/compaction_context.h>
 #include <vespa/vespalib/datastore/compaction_strategy.h>
@@ -14,6 +17,8 @@ using vespalib::datastore::CompactionSpec;
 using vespalib::datastore::CompactionStrategy;
 using vespalib::datastore::EntryRef;
 using vespalib::datastore::ICompactionContext;
+using vespalib::eval::FastValueBuilderFactory;
+using vespalib::eval::Value;
 
 namespace search::tensor {
 
@@ -54,13 +59,6 @@ DirectTensorStore::DirectTensorStore()
 
 DirectTensorStore::~DirectTensorStore() = default;
 
-EntryRef
-DirectTensorStore::store_tensor(std::unique_ptr<vespalib::eval::Value> tensor)
-{
-    assert(tensor);
-    return add_entry(TensorSP(std::move(tensor)));
-}
-
 void
 DirectTensorStore::holdTensor(EntryRef ref)
 {
@@ -100,4 +98,42 @@ DirectTensorStore::start_compact(const CompactionStrategy& compaction_strategy)
     return std::make_unique<CompactionContext>(*this, std::move(compacting_buffers));
 }
 
+EntryRef
+DirectTensorStore::store_tensor(std::unique_ptr<Value> tensor)
+{
+    assert(tensor);
+    return add_entry(std::move(tensor));
+}
+
+EntryRef
+DirectTensorStore::store_tensor(const Value& tensor)
+{
+    return add_entry(FastValueBuilderFactory::get().copy(tensor));
+}
+
+EntryRef
+DirectTensorStore::store_encoded_tensor(vespalib::nbostream& encoded)
+{
+    return add_entry(deserialize_tensor(encoded));
+}
+
+std::unique_ptr<Value>
+DirectTensorStore::get_tensor(EntryRef ref) const
+{
+    if (!ref.valid()) {
+        return {};
+    }
+    return FastValueBuilderFactory::get().copy(*_tensor_store.getEntry(ref));
+}
+
+bool
+DirectTensorStore::encode_stored_tensor(EntryRef ref, vespalib::nbostream& target) const
+{
+    if (!ref.valid()) {
+        return false;
+    }
+    vespalib::eval::encode_value(*_tensor_store.getEntry(ref), target);
+    return true;
+}
+
 }
diff --git a/searchlib/src/vespa/searchlib/tensor/direct_tensor_store.h b/searchlib/src/vespa/searchlib/tensor/direct_tensor_store.h
index 658d1ab0549..ff9540a27b3 100644
--- a/searchlib/src/vespa/searchlib/tensor/direct_tensor_store.h
+++ b/searchlib/src/vespa/searchlib/tensor/direct_tensor_store.h
@@ -52,6 +52,10 @@ public:
     EntryRef move(EntryRef ref) override;
     vespalib::MemoryUsage update_stat(const vespalib::datastore::CompactionStrategy& compaction_strategy) override;
     std::unique_ptr<vespalib::datastore::ICompactionContext> start_compact(const vespalib::datastore::CompactionStrategy& compaction_strategy) override;
+    EntryRef store_tensor(const vespalib::eval::Value& tensor) override;
+    EntryRef store_encoded_tensor(vespalib::nbostream& encoded) override;
+    std::unique_ptr<vespalib::eval::Value> get_tensor(EntryRef ref) const override;
+    bool encode_stored_tensor(EntryRef ref, vespalib::nbostream& target) const override;
 };
 
 }
diff --git a/searchlib/src/vespa/searchlib/tensor/serialized_fast_value_attribute.cpp b/searchlib/src/vespa/searchlib/tensor/serialized_fast_value_attribute.cpp
index 94bf3f1a37b..a24059b3f7c 100644
--- a/searchlib/src/vespa/searchlib/tensor/serialized_fast_value_attribute.cpp
+++ b/searchlib/src/vespa/searchlib/tensor/serialized_fast_value_attribute.cpp
@@ -1,17 +1,13 @@
 // Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
 #include "serialized_fast_value_attribute.h"
-#include "streamed_value_saver.h"
 #include <vespa/eval/eval/value.h>
-#include <vespa/fastlib/io/bufferedfile.h>
 #include <vespa/searchcommon/attribute/config.h>
 
 #include <vespa/log/log.h>
 
 LOG_SETUP(".searchlib.tensor.serialized_fast_value_attribute");
 
-#include "blob_sequence_reader.h"
-
 using namespace vespalib;
 using namespace vespalib::eval;
 
@@ -50,50 +46,4 @@ SerializedFastValueAttribute::getTensor(DocId docId) const
     return _tensorBufferStore.get_tensor(ref);
 }
 
-bool
-SerializedFastValueAttribute::onLoad(vespalib::Executor *)
-{
-    BlobSequenceReader tensorReader(*this);
-    if (!tensorReader.hasData()) {
-        return false;
-    }
-    setCreateSerialNum(tensorReader.getCreateSerialNum());
-    assert(tensorReader.getVersion() == getVersion());
-    uint32_t numDocs(tensorReader.getDocIdLimit());
-    _refVector.reset();
-    _refVector.unsafe_reserve(numDocs);
-    vespalib::Array<char> buffer(1024);
-    for (uint32_t lid = 0; lid < numDocs; ++lid) {
-        uint32_t tensorSize = tensorReader.getNextSize();
-        if (tensorSize != 0) {
-            if (tensorSize > buffer.size()) {
-                buffer.resize(tensorSize + 1024);
-            }
-            tensorReader.readBlob(&buffer[0], tensorSize);
-            vespalib::nbostream source(&buffer[0], tensorSize);
-            EntryRef ref = _tensorBufferStore.store_encoded_tensor(source);
-            _refVector.push_back(AtomicEntryRef(ref));
-        } else {
-            EntryRef invalid;
-            _refVector.push_back(AtomicEntryRef(invalid));
-        }
-    }
-    setNumDocs(numDocs);
-    setCommittedDocIdLimit(numDocs);
-    return true;
-}
-
-
-std::unique_ptr<AttributeSaver>
-SerializedFastValueAttribute::onInitSave(vespalib::stringref fileName)
-{
-    vespalib::GenerationHandler::Guard guard(getGenerationHandler().
-                                             takeGuard());
-    return std::make_unique<StreamedValueSaver>
-        (std::move(guard),
-         this->createAttributeHeader(fileName),
-         getRefCopy(),
-         _tensorBufferStore);
-}
-
 }
diff --git a/searchlib/src/vespa/searchlib/tensor/serialized_fast_value_attribute.h b/searchlib/src/vespa/searchlib/tensor/serialized_fast_value_attribute.h
index 5dd49a2bbc4..2124ddeb70a 100644
--- a/searchlib/src/vespa/searchlib/tensor/serialized_fast_value_attribute.h
+++ b/searchlib/src/vespa/searchlib/tensor/serialized_fast_value_attribute.h
@@ -24,8 +24,6 @@ public:
     ~SerializedFastValueAttribute() override;
     void setTensor(DocId docId, const vespalib::eval::Value &tensor) override;
     std::unique_ptr<vespalib::eval::Value> getTensor(DocId docId) const override;
-    bool onLoad(vespalib::Executor *executor) override;
-    std::unique_ptr<AttributeSaver> onInitSave(vespalib::stringref fileName) override;
 };
 
 }
diff --git a/searchlib/src/vespa/searchlib/tensor/tensor_attribute.cpp b/searchlib/src/vespa/searchlib/tensor/tensor_attribute.cpp
index cdaea07176a..99a30b59bd1 100644
--- a/searchlib/src/vespa/searchlib/tensor/tensor_attribute.cpp
+++ b/searchlib/src/vespa/searchlib/tensor/tensor_attribute.cpp
@@ -1,6 +1,8 @@
 // Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
 #include "tensor_attribute.h"
+#include "blob_sequence_reader.h"
+#include "tensor_store_saver.h"
 #include <vespa/document/base/exceptions.h>
 #include <vespa/document/datatype/tensor_data_type.h>
 #include <vespa/searchlib/attribute/address_space_components.h>
@@ -277,6 +279,51 @@ TensorAttribute::getRefCopy() const
     return result;
 }
 
+bool
+TensorAttribute::onLoad(vespalib::Executor*)
+{
+    BlobSequenceReader tensorReader(*this);
+    if (!tensorReader.hasData()) {
+        return false;
+    }
+    setCreateSerialNum(tensorReader.getCreateSerialNum());
+    assert(tensorReader.getVersion() == getVersion());
+    uint32_t numDocs = tensorReader.getDocIdLimit();
+    _refVector.reset();
+    _refVector.unsafe_reserve(numDocs);
+    vespalib::Array<char> buffer(1024);
+    for (uint32_t lid = 0; lid < numDocs; ++lid) {
+        uint32_t tensorSize = tensorReader.getNextSize();
+        if (tensorSize != 0) {
+            if (tensorSize > buffer.size()) {
+                buffer.resize(tensorSize + 1024);
+            }
+            tensorReader.readBlob(&buffer[0], tensorSize);
+            vespalib::nbostream source(&buffer[0], tensorSize);
+            EntryRef ref = _tensorStore.store_encoded_tensor(source);
+            _refVector.push_back(AtomicEntryRef(ref));
+        } else {
+            EntryRef invalid;
+            _refVector.push_back(AtomicEntryRef(invalid));
+        }
+    }
+    setNumDocs(numDocs);
+    setCommittedDocIdLimit(numDocs);
+    return true;
+}
+
+std::unique_ptr<AttributeSaver>
+TensorAttribute::onInitSave(vespalib::stringref fileName)
+{
+    vespalib::GenerationHandler::Guard guard(getGenerationHandler().
+                                             takeGuard());
+    return std::make_unique<TensorStoreSaver>
+        (std::move(guard),
+         this->createAttributeHeader(fileName),
+         getRefCopy(),
+         _tensorStore);
+}
+
 void
 TensorAttribute::update_tensor(DocId docId,
                                const document::TensorUpdate &update,
diff --git a/searchlib/src/vespa/searchlib/tensor/tensor_attribute.h b/searchlib/src/vespa/searchlib/tensor/tensor_attribute.h
index 411efcd8fea..7cfbb68eac7 100644
--- a/searchlib/src/vespa/searchlib/tensor/tensor_attribute.h
+++ b/searchlib/src/vespa/searchlib/tensor/tensor_attribute.h
@@ -36,6 +36,8 @@ protected:
     void populate_state(vespalib::slime::Cursor& object) const;
     void populate_address_space_usage(AddressSpaceUsage& usage) const override;
     EntryRef acquire_entry_ref(DocId doc_id) const noexcept { return _refVector.acquire_elem_ref(doc_id).load_acquire(); }
+    bool onLoad(vespalib::Executor *executor) override;
+    std::unique_ptr<AttributeSaver> onInitSave(vespalib::stringref fileName) override;
 
 public:
     using RefCopyVector = vespalib::Array<EntryRef>;
diff --git a/searchlib/src/vespa/searchlib/tensor/tensor_buffer_store.cpp b/searchlib/src/vespa/searchlib/tensor/tensor_buffer_store.cpp
index eff6ac9f374..800311adfd6 100644
--- a/searchlib/src/vespa/searchlib/tensor/tensor_buffer_store.cpp
+++ b/searchlib/src/vespa/searchlib/tensor/tensor_buffer_store.cpp
@@ -1,6 +1,7 @@
 // Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
 #include "tensor_buffer_store.h"
+#include <vespa/document/util/serializableexceptions.h>
 #include <vespa/eval/eval/value_codec.h>
 #include <vespa/eval/streamed/streamed_value_builder_factory.h>
 #include <vespa/vespalib/datastore/array_store.hpp>
@@ -10,6 +11,7 @@
 #include <vespa/vespalib/datastore/datastore.hpp>
 #include <vespa/vespalib/util/size_literals.h>
 
+using document::DeserializeException;
 using vespalib::alloc::MemoryAllocator;
 using vespalib::datastore::CompactionContext;
 using vespalib::datastore::CompactionStrategy;
@@ -90,6 +92,9 @@ TensorBufferStore::store_encoded_tensor(vespalib::nbostream &encoded)
 {
     const auto &factory = StreamedValueBuilderFactory::get();
     auto val = vespalib::eval::decode_value(encoded, factory);
+    if (!encoded.empty()) {
+        throw DeserializeException("Leftover bytes deserializing tensor attribute value.", VESPA_STRLOC);
+    }
     return store_tensor(*val);
 }
 
diff --git a/searchlib/src/vespa/searchlib/tensor/tensor_buffer_store.h b/searchlib/src/vespa/searchlib/tensor/tensor_buffer_store.h
index 585bbd7a0c3..6611660b410 100644
--- a/searchlib/src/vespa/searchlib/tensor/tensor_buffer_store.h
+++ b/searchlib/src/vespa/searchlib/tensor/tensor_buffer_store.h
@@ -30,10 +30,10 @@ public:
     EntryRef move(EntryRef ref) override;
     vespalib::MemoryUsage update_stat(const vespalib::datastore::CompactionStrategy& compaction_strategy) override;
     std::unique_ptr<vespalib::datastore::ICompactionContext> start_compact(const vespalib::datastore::CompactionStrategy& compaction_strategy) override;
-    EntryRef store_tensor(const vespalib::eval::Value &tensor);
-    EntryRef store_encoded_tensor(vespalib::nbostream &encoded);
-    std::unique_ptr<vespalib::eval::Value> get_tensor(EntryRef ref) const;
-    bool encode_stored_tensor(EntryRef ref, vespalib::nbostream &target) const;
+    EntryRef store_tensor(const vespalib::eval::Value& tensor) override;
+    EntryRef store_encoded_tensor(vespalib::nbostream& encoded) override;
+    std::unique_ptr<vespalib::eval::Value> get_tensor(EntryRef ref) const override;
+    bool encode_stored_tensor(EntryRef ref, vespalib::nbostream& target) const override;
 };
 
 }
diff --git a/searchlib/src/vespa/searchlib/tensor/tensor_deserialize.cpp b/searchlib/src/vespa/searchlib/tensor/tensor_deserialize.cpp
index a8399d9ddeb..791662caed7 100644
--- a/searchlib/src/vespa/searchlib/tensor/tensor_deserialize.cpp
+++ b/searchlib/src/vespa/searchlib/tensor/tensor_deserialize.cpp
@@ -5,6 +5,7 @@
 #include <vespa/eval/eval/fast_value.h>
 #include <vespa/eval/eval/value_codec.h>
 #include <vespa/eval/eval/value.h>
+#include <vespa/vespalib/objects/nbostream.h>
 
 using document::DeserializeException;
 using vespalib::eval::FastValueBuilderFactory;
@@ -25,10 +26,4 @@ std::unique_ptr<Value> deserialize_tensor(vespalib::nbostream &buffer)
     }
 }
 
-std::unique_ptr<Value> deserialize_tensor(const void *data, size_t size)
-{
-    vespalib::nbostream wrapStream(data, size);
-    return deserialize_tensor(wrapStream);
-}
-
 } // namespace
diff --git a/searchlib/src/vespa/searchlib/tensor/tensor_deserialize.h b/searchlib/src/vespa/searchlib/tensor/tensor_deserialize.h
index 7d1ede29167..18b9731b30b 100644
--- a/searchlib/src/vespa/searchlib/tensor/tensor_deserialize.h
+++ b/searchlib/src/vespa/searchlib/tensor/tensor_deserialize.h
@@ -1,12 +1,13 @@
 // Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
-#include <vespa/eval/eval/value.h>
-#include <vespa/vespalib/objects/nbostream.h>
+#pragma once
 
-namespace search::tensor {
+#include <memory>
 
-extern std::unique_ptr<vespalib::eval::Value>
-deserialize_tensor(const void *data, size_t size);
+namespace vespalib { class nbostream; }
+namespace vespalib::eval { struct Value; }
+
+namespace search::tensor {
 
 extern std::unique_ptr<vespalib::eval::Value>
 deserialize_tensor(vespalib::nbostream &stream);
diff --git a/searchlib/src/vespa/searchlib/tensor/tensor_store.h b/searchlib/src/vespa/searchlib/tensor/tensor_store.h
index 90bc82c4fde..e2426d2e899 100644
--- a/searchlib/src/vespa/searchlib/tensor/tensor_store.h
+++ b/searchlib/src/vespa/searchlib/tensor/tensor_store.h
@@ -8,6 +8,7 @@
 #include <vespa/vespalib/datastore/i_compactable.h>
 #include <vespa/vespalib/util/generationhandler.h>
 
+namespace vespalib { class nbostream; }
 namespace vespalib::datastore { struct ICompactionContext; }
 namespace vespalib::eval { struct Value; }
 
@@ -41,6 +42,11 @@ public:
 
     virtual std::unique_ptr<vespalib::datastore::ICompactionContext> start_compact(const vespalib::datastore::CompactionStrategy& compaction_strategy) = 0;
 
+    virtual EntryRef store_tensor(const vespalib::eval::Value& tensor) = 0;
+    virtual EntryRef store_encoded_tensor(vespalib::nbostream& encoded) = 0;
+    virtual std::unique_ptr<vespalib::eval::Value> get_tensor(EntryRef ref) const = 0;
+    virtual bool encode_stored_tensor(EntryRef ref, vespalib::nbostream& target) const = 0;
+
     // Inherit doc from DataStoreBase
     void trimHoldLists(generation_t usedGen) {
         _store.trimHoldLists(usedGen);
diff --git a/searchlib/src/vespa/searchlib/tensor/streamed_value_saver.cpp b/searchlib/src/vespa/searchlib/tensor/tensor_store_saver.cpp
index 4c188bb3370..0963e79b0dd 100644
--- a/searchlib/src/vespa/searchlib/tensor/streamed_value_saver.cpp
+++ b/searchlib/src/vespa/searchlib/tensor/tensor_store_saver.cpp
@@ -1,7 +1,7 @@
 // Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
-#include "streamed_value_saver.h"
-#include "tensor_buffer_store.h"
+#include "tensor_store_saver.h"
+#include "tensor_store.h"
 
 #include <vespa/searchlib/attribute/iattributesavetarget.h>
 #include <vespa/searchlib/util/bufferwriter.h>
@@ -11,21 +11,21 @@ using vespalib::GenerationHandler;
 
 namespace search::tensor {
 
-StreamedValueSaver::
-StreamedValueSaver(GenerationHandler::Guard &&guard,
-                   const attribute::AttributeHeader &header,
-                   RefCopyVector &&refs,
-                   const TensorBufferStore &tensorStore)
+TensorStoreSaver::
+TensorStoreSaver(GenerationHandler::Guard &&guard,
+                 const attribute::AttributeHeader &header,
+                 RefCopyVector &&refs,
+                 const TensorStore &tensorStore)
   : AttributeSaver(std::move(guard), header),
     _refs(std::move(refs)),
     _tensorStore(tensorStore)
 {
 }
 
-StreamedValueSaver::~StreamedValueSaver() = default;
+TensorStoreSaver::~TensorStoreSaver() = default;
 
 bool
-StreamedValueSaver::onSave(IAttributeSaveTarget &saveTarget)
+TensorStoreSaver::onSave(IAttributeSaveTarget &saveTarget)
 {
     auto datWriter = saveTarget.datWriter().allocBufferWriter();
     const uint32_t docIdLimit(_refs.size());
diff --git a/searchlib/src/vespa/searchlib/tensor/streamed_value_saver.h b/searchlib/src/vespa/searchlib/tensor/tensor_store_saver.h
index 0ce864769f7..a4bf6e07519 100644
--- a/searchlib/src/vespa/searchlib/tensor/streamed_value_saver.h
+++ b/searchlib/src/vespa/searchlib/tensor/tensor_store_saver.h
@@ -7,12 +7,10 @@
 
 namespace search::tensor {
 
-class TensorBufferStore;
-
 /*
  * Class for saving a tensor attribute.
  */
-class StreamedValueSaver : public AttributeSaver
+class TensorStoreSaver : public AttributeSaver
 {
 public:
     using RefCopyVector = TensorAttribute::RefCopyVector;
@@ -20,16 +18,16 @@ private:
     using GenerationHandler = vespalib::GenerationHandler;
 
     RefCopyVector _refs;
-    const TensorBufferStore &_tensorStore;
+    const TensorStore& _tensorStore;
 
     bool onSave(IAttributeSaveTarget &saveTarget) override;
 public:
-    StreamedValueSaver(GenerationHandler::Guard &&guard,
-                       const attribute::AttributeHeader &header,
-                       RefCopyVector &&refs,
-                       const TensorBufferStore &tensorStore);
+    TensorStoreSaver(GenerationHandler::Guard &&guard,
+                     const attribute::AttributeHeader &header,
+                     RefCopyVector &&refs,
+                     const TensorStore &tensorStore);
 
-    virtual ~StreamedValueSaver();
+    virtual ~TensorStoreSaver();
 };
 
 } // namespace search::tensor