MCPcopy Create free account

hub / github.com/LMCache/LMCache / functions

Functions13,371 in github.com/LMCache/LMCache

↓ 3 callersFunction_read_hugepage_info
Read hugepage pool stats from sysfs. NOTE: We only use 2 MiB hugepages, so the pool stats are taken from the 2 MiB pool directly rather than
lmcache/v1/memory_management.py:499
↓ 3 callersMethod_recalculate_capacity_locked
(self)
lmcache/v1/distributed/l2_adapters/dax_l2_adapter.py:1548
↓ 3 callersFunction_reconfigure_device_path
(payload: dict[str, object])
lmcache/v1/distributed/l2_adapters/dax_l2_adapter.py:130
↓ 3 callersMethod_register_future_async
( self, op: str, future_id: int, keepalive: Tuple[Any, ...] = () )
lmcache/v1/storage_backend/native_clients/connector_client_base.py:83
↓ 3 callersMethod_register_future_sync
( self, op: str, future_id: int, keepalive: Tuple[Any, ...] = () )
lmcache/v1/storage_backend/native_clients/connector_client_base.py:90
↓ 3 callersMethod_register_prefetch_job
(self, job: _PrefetchJob)
lmcache/v1/multiprocess/modules/lookup.py:489
↓ 3 callersMethod_release_arena_resources
( fd: Optional[int], mmap_obj: Optional[mmap.mmap], arena_view: Optional[memoryview],
lmcache/v1/storage_backend/dax/core.py:1073
↓ 3 callersFunction_remove
Helper: subtract the key's bytes from the usage ledger and drop it from the eviction LRU — the two calls that the production ``/l2/events`` ha
tests/v1/mp_coordinator/test_eviction_manager.py:58
↓ 3 callersMethod_remove_adapter
Drain and remove the peer L2 adapter for ``peer_id``. Args: peer_id: Instance id of the peer whose adapter to remove.
lmcache/v1/multiprocess/modules/p2p_controller.py:544
↓ 3 callersFunction_render_choices
Build display lines for the choice selector.
lmcache/cli/commands/bench/engine_bench/interactive/terminal.py:181
↓ 3 callersMethod_reset_connection_failures
Reset connection failure counter on successful operation.
lmcache/v1/storage_backend/connector/s3_connector.py:692
↓ 3 callersFunction_resolve_salt_from_api_path
Map the ``_default`` sentinel to the empty string.
lmcache/v1/mp_coordinator/http_apis/l2_api.py:34
↓ 3 callersMethod_root_span
Return the finished root span for *sid*, or None.
tests/v1/mp_observability/subscribers/tracing/test_mp_server.py:641
↓ 3 callersFunction_run_load
(adapter: RawBlockL2Adapter, keys, objects)
tests/v1/distributed/test_raw_block_l2_adapter.py:221
↓ 3 callersFunction_run_loop_briefly
Run keep_registered for a few ticks, then cancel it.
tests/v1/mp_coordinator/test_registrar.py:93
↓ 3 callersFunction_send_request
Send a completion request and return wall-clock time in ms.
examples/kv_cache_reuse/remote_backends/valkey/benchmark_l2.py:175
↓ 3 callersMethod_send_sync_batch
Send a batch of keys via PUSH mode
lmcache/v1/cache_controller/full_sync_sender.py:151
↓ 3 callersMethod_send_sync_end
Send FullSyncEndMsg via PUSH mode
lmcache/v1/cache_controller/full_sync_sender.py:163
↓ 3 callersFunction_send_unregister_kv_cache
Deregister a KV cache context from the MP server. The inverse of :func:`_send_register_kv_cache`. Without this call the server keeps the benc
lmcache/cli/commands/bench/server_bench/helpers.py:409
↓ 3 callersMethod_signal_event_fd
(self, event_fd: EventNotifier | None)
lmcache/v1/distributed/l2_adapters/raw_block_l2_adapter.py:766
↓ 3 callersMethod_signal_eventfd
(self, notifier: _EventNotifier)
lmcache/v1/distributed/l2_adapters/dax_l2_adapter.py:1016
↓ 3 callersFunction_skip_if_no_xpu
()
tests/benchmarks/test_xpu_layerwise_connector_benchmark.py:32
↓ 3 callersFunction_skip_if_no_xpu
()
tests/benchmarks/test_xpu_connector_benchmark.py:32
↓ 3 callersMethod_spans_by_name
(self, exporter: InMemorySpanExporter, sid: str)
tests/v1/mp_observability/subscribers/tracing/test_cb_server.py:939
↓ 3 callersMethod_store
(self, adapter: HFBucketL2Adapter, key: ObjectKey)
tests/v1/distributed/test_hfbucket_l2_adapter.py:351
↓ 3 callersMethod_store
(self, adapter, key, obj)
tests/v1/distributed/test_s3_l2_adapter.py:529
↓ 3 callersFunction_strip_warmup
Drop the leading *warmup* rounds from a BenchResult.
lmcache/cli/commands/bench/l2_adapter_bench/command.py:452
↓ 3 callersFunction_stub_dynamic_agent
(monkeypatch)
tests/v1/storage_backend/test_nixl_shared_pool.py:211
↓ 3 callersFunction_unescape_salt
Translate the URL sentinel back to the empty-string salt.
lmcache/v1/multiprocess/http_apis/quota_api.py:36
↓ 3 callersFunction_unique_name
(tag: str)
tests/v1/multiprocess/test_posix_shm.py:24
↓ 3 callersFunction_unique_token_coverage
Total token coverage, merging overlapping ranges (sliding-window probe can return overlaps; naive sum would double-count).
lmcache/v1/multiprocess/modules/blend_v3.py:309
↓ 3 callersFunction_unique_type_name
()
tests/v1/distributed/transfer_channel/test_factory.py:32
↓ 3 callersMethod_update_connection_failures
(self, error_msg: str)
lmcache/v1/storage_backend/connector/s3_connector.py:665
↓ 3 callersMethod_update_lookup_results
Store the prefix-hit count from the lookup phase.
lmcache/v1/distributed/storage_controllers/prefetch_controller.py:1001
↓ 3 callersFunction_usage_manager
Return the shared :class:`L2UsageManager` from ``app.state``.
lmcache/v1/mp_coordinator/http_apis/l2_api.py:52
↓ 3 callersFunction_v2_store_vllm_contract
Mimic vLLM non-layerwise store contract for V2 connector.
tests/benchmarks/test_xpu_connector_benchmark.py:94
↓ 3 callersFunction_validate_device_path
(value: object, field_name: str = "device_path")
lmcache/v1/distributed/l2_adapters/dax_l2_adapter.py:124
↓ 3 callersFunction_validate_extra_count
Validate and clamp extra_count. Args: extra_count: Extra lock count on top of the default 1 lock. Returns: Clamp
lmcache/v1/distributed/l1_manager.py:92
↓ 3 callersFunction_wait_event_fd
Wait for an eventfd to be signaled.
lmcache/v1/check/check_mode_test_l2_adapter.py:87
↓ 3 callersFunction_wait_fd
(fd: int, timeout: float = 30.0)
tests/v1/distributed/test_aerospike_l2_integration.py:74
↓ 3 callersFunction_wait_fd
Wait up to ``timeout`` seconds for ``event_fd`` to signal, draining it. Returns True if the fd became readable before the timeout, False otherwis
tests/v1/distributed/test_fault_inject_l2_adapter.py:63
↓ 3 callersFunction_wait_for_store
(engine, tokens, expected, timeout=60)
tests/benchmarks/test_xpu_layerwise_connector_benchmark.py:55
↓ 3 callersFunction_wait_for_store
( engine, tokens: list[torch.Tensor], expected: int, timeout: int = 60, )
tests/benchmarks/test_xpu_connector_benchmark.py:59
↓ 3 callersMethod_write_header
(self, sm_config_json: str, sm_config_digest: str)
lmcache/v1/mp_observability/trace/recorder.py:167
↓ 3 callersMethod_write_jsonl
(self, lines)
tests/tools/test_cache_simulator.py:149
↓ 3 callersFunction_write_key
(key: ObjectKey, fill_value: int, usage_fraction: float)
tests/v1/distributed/test_dax_l2_adapter.py:893
↓ 3 callersMethodaccess
Mark an existing entry as most-recently used. O(log n).
lmcache/tools/cache_simulator/lru_cache.py:101
↓ 3 callersMethodadd
(self, value)
examples/disagg_prefill/disagg_proxy_server.py:214
↓ 3 callersMethodadd_adapter
Blocking function to add a new adapter into the store controller with the specified adapter ID and descriptor. Args: adap
lmcache/v1/distributed/storage_controllers/store_controller.py:337
↓ 3 callersFunctionadd_coordinator_args
Add MP coordinator registration arguments to an existing parser. Each flag falls back to its ``LMCACHE_COORDINATOR_*`` environment variable s
lmcache/v1/multiprocess/config.py:528
↓ 3 callersMethodadd_handler
Register a handler. Args: handler: The handler to add.
lmcache/cli/metrics/metrics.py:60
↓ 3 callersFunctionadd_l2_adapters_args
Add L2 adapter configuration arguments to an existing parser. Adds a repeatable --l2-adapter <JSON> argument. Each JSON object specifies
lmcache/v1/distributed/l2_adapters/config.py:396
↓ 3 callersMethodadd_on_finished_callback
Register a callback to be invoked when a request finishes.
lmcache/cli/commands/bench/engine_bench/request_sender.py:97
↓ 3 callersMethodadd_request
Add a request to the metadata. Args: req_meta (ReqMeta): the request metadata.
lmcache/integration/vllm/vllm_v1_adapter.py:444
↓ 3 callersMethodadd_request_metadata
(self, request_metadata: LMCacheMPRequestMetadata)
lmcache/integration/vllm/lmcache_mp_connector.py:470
↓ 3 callersMethodadd_request_metadata
(self, request_metadata: LMCacheMPRequestMetadata)
lmcache/integration/vllm/lmcache_mp_connector_0201.py:444
↓ 3 callersMethodadd_request_metadata
(self, request_metadata: LMCacheMPRequestMetadata)
lmcache/integration/vllm/lmcache_mp_connector_0180.py:423
↓ 3 callersMethodallocate
Allocate a single memory object from the PD buffer. For the sender role, this method enforces staging buffer flow control: it blocks
lmcache/v1/storage_backend/pd_backend_async.py:525
↓ 3 callersFunctionapplyBareFlag
applyBareFlag appends a valueless flag (e.g. --no-enable-chunked-prefill) if it is not already present in either bare or --flag=... form. Returns a ne
operator/internal/webhook/cacheblend_inject_builders.go:262
↓ 3 callersFunctionargsContainFlagValue
argsContainFlagValue reports whether the ["--flag", "value", ...] slice contains the given two-token flag/value pair.
operator/internal/controller/cacheblendengine_controller_test.go:44
↓ 3 callersFunctionassertNoArg
(t *testing.T, args []string, flag string)
operator/internal/resources/resources_test.go:1088
↓ 3 callersFunctionassert_is_vllm_flash_attn_or_flash_infer
Ensure that we have an Engine KV Cache Format that is either vLLM's flash attention or flash infer.
lmcache/v1/gpu_connector/utils.py:347
↓ 3 callersMethodasync_lookup_and_prefetch
An async version of lookup + prefetch. There are three categories of backends: (1) sync lookup + sync retrieval (e.g., cpu)
lmcache/v1/cache_engine.py:1317
↓ 3 callersMethodasync_put_and_wait_msg
Send a message to the controller and wait for the response. This method handles different types of WorkerReqMsg using appropriate so
lmcache/v1/cache_controller/worker.py:257
↓ 3 callersMethodattach_storage_config
Write the header populated from the StorageManagerConfig. Must be called before any records are written; the server lifecycle does th
lmcache/v1/mp_observability/trace/recorder.py:108
↓ 3 callersMethodbatch_exists
Check existence of multiple keys in a single batch operation.
lmcache/v1/storage_backend/resp_client.py:201
↓ 3 callersMethodbatch_exists
(self, keys: list[str])
lmcache/v1/storage_backend/native_clients/connector_client_base.py:121
↓ 3 callersMethodbatch_exists_sync
(self, keys: list[str])
lmcache/v1/storage_backend/native_clients/connector_client_base.py:153
↓ 3 callersMethodbatched_submit_put_task
Synchronously put the MemoryObjs into the local cpu backend. :param on_complete_callback: Optional callback invoked once per key
lmcache/v1/storage_backend/local_cpu_backend.py:186
↓ 3 callersMethodbatched_submit_retrieve_requests
Submit a batched retrieve request to LMCache Args: request_ids: The IDs of the requests ops: The LoadStoreOp
lmcache/integration/vllm/vllm_multi_process_adapter.py:1510
↓ 3 callersMethodbatched_submit_store_requests
Submit a batched store request to LMCache Args: request_ids: The IDs of the requests ops: The LoadStoreOps d
lmcache/integration/vllm/vllm_multi_process_adapter.py:1484
↓ 3 callersMethodbatched_to_gpu
This function is a generator that moves the KV cache from the memory objects to paged GPU memory. The first iteration will prepare so
lmcache/v1/gpu_connector/gpu_connectors.py:1181
↓ 3 callersMethodbatched_to_gpu
This function is a generator that moves the KV cache from the memory objects to paged GPU memory. The first iteration will prepare so
lmcache/v1/gpu_connector/xpu_connectors.py:991
↓ 3 callersMethodbatched_unpin
Batch-unpin keys via single RPC. Decrements server-side pin_count for each key. When pin_count reaches 0, the entry becomes eligible
lmcache/v1/storage_backend/maru_backend.py:688
↓ 3 callersMethodbuild
Build ROCm/HIP extensions via hipcc.
setup_extensions/build_profiles/rocm.py:80
↓ 3 callersMethodbytes_get_ptr
(self, mv: Union[bytearray, memoryview, bytes])
lmcache/v1/storage_backend/connector/eic_connector.py:434
↓ 3 callersFunctioncalculate_draft_layers
Calculate the number of draft layers for speculative decoding.
lmcache/integration/vllm/utils.py:387
↓ 3 callersMethodcan_be_coalesced
(self, succ: "FreeBlock")
lmcache/v1/memory_management.py:115
↓ 3 callersMethodcan_exit_freeze
Check if the completion threshold is reached and freeze mode can be exited. Args: progress: Pre-computed global progress
lmcache/v1/cache_controller/controllers/full_sync_tracker.py:373
↓ 3 callersMethodcan_parse
(self, url: str)
lmcache/v1/storage_backend/connector/__init__.py:163
↓ 3 callersMethodcan_parse
Match plugin URLs for the built-in ``hfbucket`` connector type.
lmcache/v1/storage_backend/connector/hfbucket_adapter.py:25
↓ 3 callersMethodcan_shrink_to
Check whether every live or borrowed slot fits below a new limit. Args: new_max_slots: Candidate slot count after a shrink.
lmcache/v1/storage_backend/dax/core.py:676
↓ 3 callersMethodcheck_lookup_result
Check the result of a previously submitted lookup request. Sends a QUERY_PREFETCH_STATUS request to the servers and blocks u
lmcache/integration/vllm/vllm_multi_process_adapter.py:816
↓ 3 callersFunctioncheck_sglang_paged_kv_cache_equal
check whether two paged kv caches are the same at slot_mapping Format: List[2] -> List[num_layers] of [page_buffer_size, num_heads, head_siz
tests/v1/utils.py:574
↓ 3 callersMethodcleanup_lookup_result
Clean up lookup state for a finished request to prevent memory leak. Args: request_id: The ID of the finished request.
lmcache/integration/vllm/vllm_multi_process_adapter.py:901
↓ 3 callersMethodcleanup_memory_objs
Cleanup memory objects allocated during prefetch for an aborted lookup. Called by the scheduler when it determines that an aborted l
lmcache/v1/cache_engine.py:1375
↓ 3 callersMethodclose
(self)
tests/v1/lookup_client/test_chunk_statistics_lookup_client.py:32
↓ 3 callersMethodclose
(self)
tests/v1/shm_allocator/shmfile_connector.py:295
↓ 3 callersMethodclose
(self)
examples/lmc_external_l2_adapter/src/lmc_external_l2_adapter/adapter.py:229
↓ 3 callersMethodclose
(self)
lmcache/v1/distributed/l2_adapters/hfbucket_l2_adapter.py:470
↓ 3 callersMethodclose
Close the P2P backend and cleanup resources.
lmcache/v1/storage_backend/p2p_backend.py:711
↓ 3 callersMethodclose
(self)
lmcache/v1/storage_backend/local_disk_backend.py:91
↓ 3 callersMethodclose
Close the DAX arena after active reads and writes finish. The call is idempotent. It marks the core as closing, waits for active oper
lmcache/v1/storage_backend/dax/core.py:776
↓ 3 callersFunctioncompare_results
Compare native transfer against the torch fallback.
benchmarks/musa/bench_inprocess_transfer.py:35
↓ 3 callersMethodcompress
( self, tokens: Union[torch.Tensor, List[int]], method: str, location: str,
lmcache/v1/cache_engine.py:1419
↓ 3 callersFunctioncompute_extra_count
Compute extra count for MLA multi-reader locking. Non-MLA: each TP worker owns a distinct KV shard, so each ObjectKey is retrieved by exact
lmcache/v1/multiprocess/modules/lookup.py:30
↓ 3 callersMethodcontext_entries_snapshot
Return a shallow copy of the registry for iteration or status. Returns: A new dict mapping instance ID to entry; does not refresh
lmcache/v1/multiprocess/modules/lmcache_driven_transfer.py:475
↓ 3 callersFunctioncountLMCacheRetrieved
countLMCacheRetrieved is the RETRIEVE counterpart of countLMCacheStored.
operator/test/e2e/vllm_integration_smoke_test.go:249
↓ 3 callersMethodcreatePool
Create a NIXL descriptor pool with path sharding support. Args: backend: Backend type (e.g., "GDS", "POSIX", "OBJ").
lmcache/v1/storage_backend/nixl_storage_backend.py:1027
↓ 3 callersFunctioncreate_config
backend can be: - cpu - disk - fsconnector
tests/benchmarks/test_benchmark.py:55
← previousnext →1,801–1,900 of 13,371, ranked by callers