digitalkin 1.0.0.dev3__tar.gz → 1.0.0.dev5__tar.gz

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: digitalkin
-Version: 1.0.0.dev3
+Version: 1.0.0.dev5
 Summary: SDK to build kin used in DigitalKin
 Author-email: "DigitalKin.ai" <contact@digitalkin.ai>
 License: Attribution-NonCommercial-ShareAlike 4.0 International
@@ -458,7 +458,7 @@ Requires-Dist: anyio>=4.13.0
 Requires-Dist: grpcio-health-checking==1.80.0
 Requires-Dist: grpcio-reflection==1.80.0
 Requires-Dist: grpcio-status==1.80.0
-Requires-Dist: pydantic>=2.12.5
+Requires-Dist: pydantic>=2.12.4
 Requires-Dist: pydantic-settings>=2.14.0
 Requires-Dist: redis[hiredis]<8,>=7.4.0
 Provides-Extra: performance

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/pyproject.toml

@@ -33,11 +33,11 @@
         "grpcio-health-checking==1.80.0",
         "grpcio-reflection==1.80.0",
         "grpcio-status==1.80.0",
-        "pydantic>=2.12.5",
+        "pydantic>=2.12.4",
         "pydantic-settings>=2.14.0",
         "redis[hiredis]>=7.4.0,<8",
     ]
-    version = "1.0.0.dev3"
+    version = "1.0.0.dev5"
 
     [project.optional-dependencies]
         performance = [ "uvloop>=0.21" ]
@@ -59,8 +59,8 @@
     dev = [
         "build>=1.5.0",
         "bump-my-version>=1.3.0",
-        "cryptography>=47.0.0",
-        "mypy>=1.20.2",
+        "cryptography>=48.0.0",
+        "mypy>=2.0.0",
         "pre-commit>=4.6.0",
         "pyright>=1.1.409",
         "ruff>=0.15.12",
@@ -68,8 +68,8 @@
         "types-grpcio-health-checking>=1.0.0.20260408",
         "types-grpcio-reflection>=1.0.0.20260408",
         "types-grpcio>=1.0.0.20260408",
-        "types-protobuf>=7.34.1.20260408",
-        "typos>=1.45.2",
+        "types-protobuf>=7.34.1.20260503",
+        "typos>=1.46.0",
     ]
     docs = [
         "griffe-inherited-docstrings>=1.1.3",

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin/__version__.py

@@ -5,4 +5,4 @@ from importlib.metadata import PackageNotFoundError, version
 try:
     __version__ = version("digitalkin")
 except PackageNotFoundError:
-    __version__ = "1.0.0.dev3"
+    __version__ = "1.0.0.dev5"

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin/core/job_manager/single_job_manager.py

@@ -83,6 +83,13 @@ class SingleJobManager(BaseJobManager[InputModelT, OutputModelT, SetupModelT]):
         self._redis_client = redis_client
         self._stream_writers: dict[str, RedisStreamWriter] = {}
 
+        # Pool one RedisTaskManager across all preload_instance calls.
+        # The class is task-id-stateless; its `_listener` is already a
+        # process-wide singleton via SharedRedisListener.get_or_create.
+        from digitalkin.services.task_manager.redis_task_manager import RedisTaskManager
+
+        self._task_manager_strategy = RedisTaskManager(self._redis_client)
+
     async def start(self) -> None:
         """Start manager (no-op, no external connections needed)."""
 
@@ -273,7 +280,6 @@ class SingleJobManager(BaseJobManager[InputModelT, OutputModelT, SetupModelT]):
             ``(module, job_id, callback)`` — pass to ``run_preloaded``.
         """
         from digitalkin.core.profiling.step_timer import StepTimer
-        from digitalkin.services.task_manager.redis_task_manager import RedisTaskManager
 
         timer = StepTimer()
         job_id = job_id or str(uuid.uuid4())
@@ -288,7 +294,8 @@ class SingleJobManager(BaseJobManager[InputModelT, OutputModelT, SetupModelT]):
         )
         timer.mark("factory_create")
 
-        module.context.task_manager = RedisTaskManager(self._redis_client)
+        # Reuse the pooled RedisTaskManager — task-id-stateless, safe to share.
+        module.context.task_manager = self._task_manager_strategy
         timer.mark("redis_task_manager")
 
         if callback is None:

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin/core/task_manager/module_runner.py

@@ -115,8 +115,16 @@ class ModuleRunner:
                 nonlocal first_logged
                 data = output_data.model_dump(mode="json")
                 if data.get("root", {}).get("protocol") == "stream.end":
+                    t_eos_write_start = time.perf_counter_ns()
                     await self._redis_client.xadd(stream_key, {"eos": b"true"})
                     await self._redis_client.expire(stream_key, 60)
+                    t_eos_write_end = time.perf_counter_ns()
+                    logger.info(
+                        "[close-debug] producer_eos_write: xadd_expire=%.2fms t_done_ns=%d task_id=%s",
+                        (t_eos_write_end - t_eos_write_start) / 1e6,
+                        t_eos_write_end,
+                        task_id,
+                    )
                     return
                 s = struct_pb2.Struct()
                 s.update(data)

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin/core/task_manager/redis/proto_streams.py

@@ -369,11 +369,13 @@ class ProtoStreamReader:
         """
         entries_since_save = 0
         while True:
+            t_xread_start = time.perf_counter_ns()
             result = await self._redis_client.xread(
                 {self._stream_key: self._last_id},
                 count=count,
                 block=block_ms,
             )
+            t_xread_end = time.perf_counter_ns()
             if not result:
                 continue
 
@@ -383,6 +385,13 @@ class ProtoStreamReader:
 
                     eos = fields.get(b"eos", b"")
                     if eos == b"true":
+                        logger.info(
+                            "[close-debug] reader_saw_eos: last_xread_block=%.2fms "
+                            "t_seen_ns=%d task_id=%s",
+                            (t_xread_end - t_xread_start) / 1e6,
+                            t_xread_end,
+                            self._task_id,
+                        )
                         await self._save_cursor()
                         return
 

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin/grpc_servers/gateway_constants.py

@@ -19,7 +19,6 @@ _prof = ProfilingSettings()
 # Redis Key Patterns (not configurable — structural)
 # ══════════════════════════════════════════════════════════════════
 
-REDIS_KEY_SESSION = "gateway:session:{task_id}"
 REDIS_KEY_STREAM = "task:{task_id}:stream"
 REDIS_KEY_INPUT_STREAM = "task:{task_id}:input"
 REDIS_KEY_CURSOR = "task:{task_id}:cursor"
@@ -31,7 +30,6 @@ REDIS_KEY_SIGNAL_CHANNEL = "signal_ch:{task_id}"
 
 MAX_STREAMS = _gw.max_streams
 MAX_LOCAL_CACHE = _gw.max_local_cache
-SESSION_STATE_TTL_S = _gw.session_state_ttl
 DIAL_BACK_BIDI_TIMEOUT_S = _gw.dial_back_bidi_timeout_s
 
 # ══════════════════════════════════════════════════════════════════

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin/grpc_servers/gateway_servicer.py

@@ -555,8 +555,17 @@ class GatewayServicer:
         # consumed silently by ProtoStreamReader). Emit an explicit stream.end
         # sentinel so the wire contract is uniform: every stream ends with
         # exactly one stream.end entry, regardless of how it concluded.
+        t_after_reader = time.perf_counter_ns()
         seq += 1
         yield self._sentinel(seq, task_id, "stream.end")
+        t_after_yield = time.perf_counter_ns()
+        logger.info(
+            "[close-debug] gateway_stream_end: reader_to_yield=%.2fms "
+            "t_yielded_ns=%d task_id=%s",
+            (t_after_yield - t_after_reader) / 1e6,
+            t_after_yield,
+            task_id,
+        )
 
     async def _dial_consumer(  # noqa: C901, PLR0912, PLR0915
         self,

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin/grpc_servers/stream_registry.py

@@ -6,10 +6,9 @@ bound to the dial-back asyncio task: the task's ``finally`` calls
 ``BaseException`` propagated past finally), the task done-callback
 force-unregisters as a backstop.
 
-Optional Redis session-state mirror (``gateway:session:{task_id}`` HSET)
-is written on register and deleted on unregister for cross-instance
-observability. Heartbeat-based zombie reaping is gone — long streams
-were getting cut off when their heartbeat zset entry went stale.
+No Redis I/O on register/unregister — the gateway is fully local for
+session lifecycle. The previous Redis session-state mirror had no readers
+once the heartbeat reaper was retired.
 """
 
 from __future__ import annotations
@@ -18,13 +17,10 @@ import asyncio
 from collections import OrderedDict
 from typing import TYPE_CHECKING, Any
 
-from redis.exceptions import RedisError
-
 from digitalkin.core.resilience.task_supervisor import log_unhandled
 from digitalkin.grpc_servers.gateway_constants import (
     MAX_LOCAL_CACHE,
     MAX_STREAMS,
-    SESSION_STATE_TTL_S,
 )
 from digitalkin.logger import logger
 
@@ -32,62 +28,38 @@ if TYPE_CHECKING:
     from digitalkin.core.task_manager.redis.redis_client import RedisClient
     from digitalkin.grpc_servers.stream_session import StreamSession
 
-# Lua: atomic session-state write on register.
-# KEYS: [1]=session_key (empty string to skip)
-# ARGV: [1]=setup_id, [2]=mission_id, [3]=session_ttl
-# Returns 1 unconditionally — capacity is enforced process-locally now.
-_LUA_REGISTER = """
-local session_key = KEYS[1]
-if session_key ~= '' then
-    redis.call('HSET', session_key, 'status', 'starting', 'setup_id', ARGV[1], 'mission_id', ARGV[2])
-    redis.call('EXPIRE', session_key, tonumber(ARGV[3]))
-end
-return 1
-"""
-
 
 class StreamRegistry:
     """Tracks active stream sessions per-instance + supervises spawned tasks.
 
     Local dict is a bounded LRU cache of sessions with active BiDi
     connections on this gateway instance. Capacity is enforced
-    process-locally against ``max_streams``. Optional Redis HSET
-    mirrors session metadata for observability.
+    process-locally against ``max_streams``.
     """
 
     _local_cache: OrderedDict[str, StreamSession]
     _max_local: int
     _max_streams: int
     _monitored_tasks: set[asyncio.Task[Any]]
-    _redis_client: RedisClient
-
-    @staticmethod
-    def session_key(task_id: str) -> str:
-        """Redis hash key for session metadata.
-
-        Returns:
-            Key in the format ``gateway:session:{task_id}``.
-        """
-        return f"gateway:session:{task_id}"
 
     def __init__(
         self,
-        redis_client: RedisClient,
+        redis_client: RedisClient | None = None,  # noqa: ARG002 — kept for back-compat with callers
         max_streams: int = MAX_STREAMS,
         max_local: int = MAX_LOCAL_CACHE,
     ) -> None:
         """Initialize the stream registry.
 
         Args:
+            redis_client: Unused (kept for back-compat); the registry no
+                longer touches Redis on register/unregister.
             max_streams: Maximum concurrent streams on this instance.
             max_local: Maximum sessions cached locally on this instance.
-            redis_client: Redis client for the optional session-state mirror.
         """
         self._local_cache = OrderedDict()
         self._max_local = max_local
         self._max_streams = max_streams
         self._monitored_tasks = set()
-        self._redis_client = redis_client
 
     @property
     def active_count(self) -> int:
@@ -97,15 +69,15 @@ class StreamRegistry:
     async def register(
         self,
         session: StreamSession,
-        setup_id: str = "",
-        mission_id: str = "",
+        setup_id: str = "",  # noqa: ARG002 — accepted for back-compat with callers
+        mission_id: str = "",  # noqa: ARG002 — accepted for back-compat with callers
     ) -> bool:
         """Register a new session. Capacity is enforced process-locally.
 
         Args:
             session: The stream session to register.
-            setup_id: Setup ID to store in session state (optional).
-            mission_id: Mission ID to store in session state (optional).
+            setup_id: Accepted for back-compat; no longer persisted to Redis.
+            mission_id: Accepted for back-compat; no longer persisted to Redis.
 
         Returns:
             True if registered, False if at capacity (this instance).
@@ -114,18 +86,6 @@ class StreamRegistry:
         if len(self._local_cache) >= self._max_streams:
             return False
 
-        # Optional: mirror session metadata to Redis for observability.
-        if setup_id:
-            try:
-                await self._redis_client.eval(
-                    _LUA_REGISTER,
-                    [self.session_key(session.task_id)],
-                    [setup_id, mission_id, str(SESSION_STATE_TTL_S)],
-                )
-            except RedisError:
-                logger.exception("Redis session-state write failed: task_id=%s", session.task_id)
-                # Continue — local registration is enough.
-
         # LRU cache — evict oldest if past max_local (separate from
         # max_streams which gates new admissions above).
         if len(self._local_cache) >= self._max_local:
@@ -151,7 +111,7 @@ class StreamRegistry:
         return session
 
     async def unregister(self, task_id: str) -> StreamSession | None:
-        """Unregister a session and delete its Redis session-state key.
+        """Unregister a session from the local cache.
 
         Args:
             task_id: Session to remove.
@@ -160,12 +120,6 @@ class StreamRegistry:
             The removed session, or None if not found locally.
         """
         session = self._local_cache.pop(task_id, None)
-
-        try:
-            await self._redis_client.delete(self.session_key(task_id))
-        except RedisError:
-            logger.exception("Redis session-state delete failed: task_id=%s", task_id)
-
         if session is not None:
             logger.debug("StreamRegistry.unregister: task_id=%s local=%d", task_id, len(self._local_cache))
         return session

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin/models/settings/gateway.py

@@ -56,7 +56,6 @@ class GatewaySettings(BaseSettings):
 
     max_streams: int = Field(default=20000, description="Max concurrent gateway sessions (per instance)")
     max_local_cache: int = Field(default=5000, description="Max local session cache entries")
-    session_state_ttl: int = Field(default=3600, description="Session metadata TTL in seconds")
     redis_health_timeout: float = Field(default=5.0, description="Redis health check timeout in seconds")
     dial_back_bidi_timeout_s: float = Field(
         default=300.0,

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin/modules/_base_module.py

@@ -695,11 +695,14 @@ class BaseModule(  # Module SDK base class requires many public methods # noqa:
             t3 = _t.perf_counter_ns()
             self._status = ModuleStatus.STOPPED
             logger.info(
-                "module.stop: cleanup=%.1fms flush=%.1fms eos=%.1fms total=%.1fms",
+                "[close-debug] module.stop: cleanup=%.2fms flush=%.2fms eos=%.2fms "
+                "total=%.2fms t_done_ns=%d task_id=%s",
                 (t1 - t0) / 1e6,
                 (t2 - t1) / 1e6,
                 (t3 - t2) / 1e6,
                 (t3 - t0) / 1e6,
+                t3,
+                self.context.session.current_ids().get("task_id", ""),
             )
         except Exception:
             self._status = ModuleStatus.FAILED

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: digitalkin
-Version: 1.0.0.dev3
+Version: 1.0.0.dev5
 Summary: SDK to build kin used in DigitalKin
 Author-email: "DigitalKin.ai" <contact@digitalkin.ai>
 License: Attribution-NonCommercial-ShareAlike 4.0 International
@@ -458,7 +458,7 @@ Requires-Dist: anyio>=4.13.0
 Requires-Dist: grpcio-health-checking==1.80.0
 Requires-Dist: grpcio-reflection==1.80.0
 Requires-Dist: grpcio-status==1.80.0
-Requires-Dist: pydantic>=2.12.5
+Requires-Dist: pydantic>=2.12.4
 Requires-Dist: pydantic-settings>=2.14.0
 Requires-Dist: redis[hiredis]<8,>=7.4.0
 Provides-Extra: performance

{digitalkin-1.0.0.dev3 → digitalkin-1.0.0.dev5}/src/digitalkin.egg-info/requires.txt

@@ -4,7 +4,7 @@ anyio>=4.13.0
 grpcio-health-checking==1.80.0
 grpcio-reflection==1.80.0
 grpcio-status==1.80.0
-pydantic>=2.12.5
+pydantic>=2.12.4
 pydantic-settings>=2.14.0
 redis[hiredis]<8,>=7.4.0