Integrating planner stats db with ConfigeratorStats (#3331)

Summary:
Pull Request resolved: #3331

internal
Context: Planner stats db is introduced in this diff to track metadata and perf metrics associated with sharding plan.

This Diff:
1. Added methods to insert, select and delete planner stats db row.
2. UTs for planner stats db
3. Integration of planner stats db with ConfigeratorStats

Reviewed By: ge0405

Differential Revision: D81216987

fbshipit-source-id: 8591026a3122aab110161eb80aaa51d5abec81b8

Author: aliafzal

torchrec/distributed/planner/planners.py

@@ -42,6 +42,7 @@
     ParameterConstraints,
     Partitioner,
     PerfModel,
+    PlanDebugStats,
     PlannerError,
     PlannerErrorType,
     Proposer,
@@ -528,6 +529,10 @@ def plan(
                     enumerator=self._enumerator,
                     sharders=sharders,
                     debug=self._debug,
+                    debug_stats=PlanDebugStats(
+                        planner_type=self.__class__.__name__,
+                        timeout_seconds=self._timeout_seconds,
+                    ),
                 )
             return sharding_plan
         else:

torchrec/distributed/planner/stats.py

@@ -40,6 +40,7 @@
     Enumerator,
     ParameterConstraints,
     Perf,
+    PlanDebugStats,
     ShardingOption,
     Stats,
     Storage,
@@ -160,6 +161,7 @@ def log(
         sharders: Optional[List[ModuleSharder[nn.Module]]] = None,
         enumerator: Optional[Enumerator] = None,
         debug: bool = True,
+        debug_stats: Optional[PlanDebugStats] = None,
     ) -> None:
         """
         Logs stats for a given sharding plan.
@@ -1138,5 +1140,6 @@ def log(
         sharders: Optional[List[ModuleSharder[nn.Module]]] = None,
         enumerator: Optional[Enumerator] = None,
         debug: bool = True,
+        debug_stats: Optional[PlanDebugStats] = None,
     ) -> None:
         pass

torchrec/distributed/planner/types.py

@@ -960,6 +960,16 @@ def partition(
         ...
 
 
+@dataclass
+class PlanDebugStats:
+    """
+    Representation of debug stats associated with a sharding plan, used for logging.
+    """
+
+    planner_type: str
+    timeout_seconds: Optional[int]
+
+
 class Stats(abc.ABC):
     """
     Logs statistics related to the sharding plan.
@@ -980,6 +990,7 @@ def log(
         sharders: Optional[List[ModuleSharder[nn.Module]]] = None,
         enumerator: Optional[Enumerator] = None,
         debug: bool = False,
+        debug_stats: Optional[PlanDebugStats] = None,
     ) -> None:
         """
         See class description
@@ -1007,6 +1018,16 @@ def hash_sha256_to_int(hashable_list: List[Any]) -> int:  # pyre-ignore
     return int(hash_digest, 16)
 
 
+def hash_sha256_str(hashable_list: List[Any]) -> str:  # pyre-ignore
+    """
+    Hashes the given data using SHA256 and returns the hash as an string
+    """
+    serialized_list = str(hashable_list).encode("utf-8")
+    hash_object = hashlib.sha256(serialized_list)
+    hash_digest = hash_object.hexdigest()
+    return hash_digest
+
+
 def hash_planner_context_inputs(
     topology: Topology,
     batch_size: int,
@@ -1047,3 +1068,45 @@ def hash_planner_context_inputs(
         constraints.items() if constraints else None,
     ]
     return hash_function(hashable_list)
+
+
+def hash_planner_context_inputs_str(
+    topology: Topology,
+    batch_size: int,
+    enumerator: Enumerator,
+    storage_reservation: StorageReservation,
+    constraints: Optional[Dict[str, ParameterConstraints]],
+    # pyre-ignore
+    hash_function: Callable[[List[Any]], str] = hash_sha256_str,
+) -> str:
+    assert hasattr(
+        enumerator, "last_stored_search_space"
+    ), "This enumerator is not compatible with hashing"
+    assert (
+        enumerator.last_stored_search_space is not None  # pyre-ignore
+    ), "Unable to hash planner context without an enumerator that has a precomputed search space"
+    search_space = enumerator.last_stored_search_space
+    storage_reservation_policy = type(storage_reservation).__name__
+
+    assert (
+        storage_reservation._last_reserved_topology is not None  # pyre-ignore
+    ), "Unable to hash planner context without a storage reservation that has a precomputed topology"
+
+    hashable_list = [
+        topology,
+        batch_size,
+        [
+            [
+                shard_option.fqn,
+                shard_option.sharding_type,
+                shard_option.compute_kernel,
+                tuple(shard_option.shards),
+                shard_option.cache_params,
+            ]
+            for shard_option in search_space
+        ],
+        storage_reservation_policy,
+        storage_reservation._last_reserved_topology,
+        constraints.items() if constraints else None,
+    ]
+    return hash_function(hashable_list)