Skip to content

[None][fix] Fix KV cache manager test warnings #9103

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
Tabrizian merged 1 commit into from
Nov 13, 2025
+14 −9
Merged

[None][fix] Fix KV cache manager test warnings #9103

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
23 changes: 14 additions & 9 deletions cpp/tests/unit_tests/batch_manager/kvCacheManagerTest.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,10 +34,12 @@
#include <chrono>
#include <cmath>
#include <cstddef>
#include <fcntl.h>
#include <filesystem>
#include <memory>
#include <set>
#include <thread>
#include <unistd.h>
#include <variant>

using namespace tensorrt_llm::batch_manager;
Expand Down Expand Up @@ -212,7 +214,10 @@ void writePatternToOffloadedBlocksGDS(
{
buffer[i] = i & mask;
}
::write(fd, buffer.data(), poolBlockSize * sizeof(T));
auto const bytesToWrite = static_cast<size_t>(poolBlockSize) * sizeof(T);
auto const written = ::write(fd, buffer.data(), bytesToWrite);
EXPECT_EQ(written, static_cast<ssize_t>(bytesToWrite))
<< "Failed to write pattern to offloaded block file " << filename;
::close(fd);
}
}
Expand Down Expand Up @@ -3575,7 +3580,7 @@ TEST_F(KVCacheManagerTest, KVCacheManagerMaxAttentionWindowWithReuseTest)
auto numAllocatedPrimaryBlocks = blockManager.getNumAllocatedBlocks() - blocksInSecondaryPool;
EXPECT_THAT(seq0.getCacheBlockIds(onlyWindowSize).at(beamIdx), ::testing::ElementsAreArray({0, 1, 2, 3, 4}));

EXPECT_NO_THROW(kvCacheManager.removeSequence(requestId, llmRequest));
EXPECT_NO_THROW(static_cast<void>(kvCacheManager.removeSequence(requestId, llmRequest)));
numAllocatedPrimaryBlocks = blockManager.getNumAllocatedBlocks() - blocksInSecondaryPool;
EXPECT_EQ(numAllocatedPrimaryBlocks, 0);
// store blocks 0, 1, 2, 3, 4 for reuse ([1000,1001,1002,1003], [1004,1005,1006,1007], [1008,1009,1010,1011],
Expand All @@ -3601,7 +3606,7 @@ TEST_F(KVCacheManagerTest, KVCacheManagerMaxAttentionWindowWithReuseTest)
kvCacheManager.addToken(requestId);
numTokens = llmRequest->getNumTokens(beamIdx);
EXPECT_THAT(seq1.getCacheBlockIds(onlyWindowSize).at(beamIdx), ::testing::ElementsAreArray({0, 5, 6}));
EXPECT_NO_THROW(kvCacheManager.removeSequence(requestId, llmRequest));
EXPECT_NO_THROW(static_cast<void>(kvCacheManager.removeSequence(requestId, llmRequest)));

///////////////////////////////////////////////////////////////////////////
// add a medium request and then remove it
Expand All @@ -3615,7 +3620,7 @@ TEST_F(KVCacheManagerTest, KVCacheManagerMaxAttentionWindowWithReuseTest)
GenerationRequest const& seq2 = kvCacheManager.getSequence(requestId);
EXPECT_EQ(llmRequest->getContextCurrentPosition(), 9);
EXPECT_THAT(seq2.getCacheBlockIds(onlyWindowSize).at(beamIdx), ::testing::ElementsAreArray({0, 1, 7}));
EXPECT_NO_THROW(kvCacheManager.removeSequence(requestId, llmRequest));
EXPECT_NO_THROW(static_cast<void>(kvCacheManager.removeSequence(requestId, llmRequest)));

///////////////////////////////////////////////////////////////////////////
// add a longer request within attention window and try to reuse
Expand All @@ -3637,7 +3642,7 @@ TEST_F(KVCacheManagerTest, KVCacheManagerMaxAttentionWindowWithReuseTest)
llmRequest->addNewToken(1016, beamIdx);
kvCacheManager.addToken(requestId);
EXPECT_THAT(seq3.getCacheBlockIds(onlyWindowSize).at(beamIdx), ::testing::ElementsAreArray({0, 1, 2, 8, 9}));
EXPECT_NO_THROW(kvCacheManager.removeSequence(requestId, llmRequest));
EXPECT_NO_THROW(static_cast<void>(kvCacheManager.removeSequence(requestId, llmRequest)));
}

TEST_F(KVCacheManagerTest, KVCacheManagerSWAInvalidateReuseTest)
Expand Down Expand Up @@ -3715,8 +3720,8 @@ TEST_F(KVCacheManagerTest, KVCacheManagerSWAInvalidateReuseTest)
EXPECT_FALSE(blockManager.isSequenceValidForStoreForReuse(seq0.getRequestId(), onlyWindowSize));
EXPECT_TRUE(blockManager.isSequenceValidForStoreForReuse(seq1.getRequestId(), onlyWindowSize));

EXPECT_NO_THROW(kvCacheManager.removeSequence(seq0.getRequestId(), llmRequest0));
EXPECT_NO_THROW(kvCacheManager.removeSequence(seq1.getRequestId(), llmRequest1));
EXPECT_NO_THROW(static_cast<void>(kvCacheManager.removeSequence(seq0.getRequestId(), llmRequest0)));
EXPECT_NO_THROW(static_cast<void>(kvCacheManager.removeSequence(seq1.getRequestId(), llmRequest1)));
}

TEST_F(KVCacheManagerTest, KVCacheManagerVariableWindowAttentionWithReuseTest)
Expand Down Expand Up @@ -3806,7 +3811,7 @@ TEST_F(KVCacheManagerTest, KVCacheManagerVariableWindowAttentionWithReuseTest)
assertBlocks(seq0, {0, 1, 2}, {0, 1, 2});
auto numAllocatedPrimaryBlocks = blockManager.getNumAllocatedBlocks() - blocksInSecondaryPoolPerWindow * numWindows;

EXPECT_NO_THROW(kvCacheManager.removeSequence(requestId, llmRequest));
EXPECT_NO_THROW(static_cast<void>(kvCacheManager.removeSequence(requestId, llmRequest)));
numAllocatedPrimaryBlocks = blockManager.getNumAllocatedBlocks() - blocksInSecondaryPoolPerWindow * numWindows;
EXPECT_EQ(numAllocatedPrimaryBlocks, 0);
// For both windows, store blocks 0, 1, 2 for reuse ([1000,1001,1002,1003], [1004,1005,1006,1007],
Expand All @@ -3832,7 +3837,7 @@ TEST_F(KVCacheManagerTest, KVCacheManagerVariableWindowAttentionWithReuseTest)
llmRequest->addNewToken(1009, beamIdx);
kvCacheManager.addToken(requestId);
assertBlocks(seq1, {0, 3, 4}, {0, 3, 4});
EXPECT_NO_THROW(kvCacheManager.removeSequence(requestId, llmRequest));
EXPECT_NO_THROW(static_cast<void>(kvCacheManager.removeSequence(requestId, llmRequest)));
}

TEST_F(KVCacheManagerTest, KVCacheManagerEventStreamOverflow)
Expand Down