move session to python/sglang/srt/session (#23144)

python/sglang/srt/managers/schedule_batch.py

@@ -96,8 +96,8 @@ if TYPE_CHECKING:
 
     from sglang.srt.configs.model_config import ModelConfig
     from sglang.srt.managers.hisparse_coordinator import HiSparseCoordinator
-    from sglang.srt.managers.session_controller import Session
     from sglang.srt.observability.scheduler_metrics_mixin import PrefillStats
+    from sglang.srt.session.session_controller import Session
     from sglang.srt.speculative.eagle_info import EagleDraftInput
     from sglang.srt.speculative.spec_info import SpecInput, SpeculativeAlgorithm
 

python/sglang/srt/managers/scheduler.py

@@ -183,12 +183,10 @@ from sglang.srt.managers.scheduler_runtime_checker_mixin import (
 from sglang.srt.managers.scheduler_update_weights_mixin import (
     SchedulerUpdateWeightsMixin,
 )
-from sglang.srt.managers.session_controller import SessionController
 from sglang.srt.managers.utils import GenerationBatchResult, validate_input_length
 from sglang.srt.mem_cache.cache_init_params import CacheInitParams
 from sglang.srt.mem_cache.common import release_kv_cache
 from sglang.srt.mem_cache.radix_cache import RadixCache
-from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache
 from sglang.srt.model_executor.forward_batch_info import ForwardMode, PPProxyTensors
 from sglang.srt.model_loader.utils import get_resolved_model_impl
 from sglang.srt.multiplex.multiplexing_mixin import SchedulerMultiplexMixin
@@ -207,6 +205,8 @@ from sglang.srt.parser.reasoning_parser import ReasoningParser
 from sglang.srt.plugins import load_plugins
 from sglang.srt.sampling.sampling_batch_info import SamplingBatchInfo
 from sglang.srt.server_args import PortArgs, ServerArgs, get_global_server_args
+from sglang.srt.session.session_aware_cache import SessionAwareCache
+from sglang.srt.session.session_controller import SessionController
 from sglang.srt.speculative.spec_info import SpeculativeAlgorithm
 from sglang.srt.utils import (
     DynamicGradMode,

python/sglang/srt/managers/scheduler_runtime_checker_mixin.py

@@ -8,8 +8,8 @@ from typing import TYPE_CHECKING, List, Optional, Tuple
 
 from sglang.srt.disaggregation.utils import DisaggregationMode
 from sglang.srt.environ import envs
-from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache
 from sglang.srt.observability.metrics_collector import QueueCount
+from sglang.srt.session.session_aware_cache import SessionAwareCache
 from sglang.srt.utils.common import ceil_align, raise_error_or_warn
 from sglang.srt.utils.request_logger import disable_request_logging
 from sglang.srt.utils.watchdog import WatchdogRaw

python/sglang/srt/session/session_controller.py

@@ -24,7 +24,7 @@ from sglang.srt.managers.io_struct import (
     TokenizedGenerateReqInput,
 )
 from sglang.srt.managers.schedule_batch import FINISH_ABORT, Req
-from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache
+from sglang.srt.session.session_aware_cache import SessionAwareCache
 from sglang.srt.utils.common import log_info_on_rank0
 
 if TYPE_CHECKING:

test/registered/unit/mem_cache/test_streaming_session_unit.py

@@ -5,7 +5,7 @@ import torch
 from sglang.srt.managers.schedule_batch import FINISH_ABORT
 from sglang.srt.mem_cache.base_prefix_cache import MatchResult
 from sglang.srt.mem_cache.common import release_kv_cache
-from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache, SessionSlot
+from sglang.srt.session.session_aware_cache import SessionAwareCache, SessionSlot
 from sglang.test.ci.ci_register import register_cpu_ci
 
 register_cpu_ci(est_time=8, suite="stage-a-test-cpu")