Refactored

src/AdoNet/Orleans.GrainDirectory.AdoNet/PostgreSQL-GrainDirectory.sql

@@ -1,19 +1,19 @@
 /*
 Orleans Grain Directory.
 
-This tables stores the location of all grains in the cluster.
+This table stores the location of all grains in the cluster.
 
-The rationale for this table is as follows:
+The demands for this table are as follows:
 
 1. The table will see rows inserted individually, as new grains are added.
 2. The table will see rows deleted at random as grains are deactivated, without regard for order.
 3. Insert/Delete churn is expected to be very high.
 4. The GrainId is too large to be indexed by SQL Server.
 
-Given the above, the table cannot be a clustered index.
-Not only is the GrainId too large to index directly, the expected insert/delete churn would cause fragmentation to the point of rendering the directory unusable.
-Therefore the design choice is to use a heap table with a non-clustered index on the stable hash of the grain key.
+To address these demands, this table is implemented as a non-unique table with a hash index.
+This index covers the hash of the grain id to allow for fast lookups.
 Uniqueness is then guaranteed by careful use of locks on the hash index.
+
 */
 CREATE TABLE OrleansGrainDirectory
 (
@@ -39,27 +39,15 @@ CREATE TABLE OrleansGrainDirectory
     CreatedOn TIMESTAMPTZ NOT NULL
 );
 
-/*
-This index is a workaround for the GrainId being too large to index by SQL Server.
-Instead we index a stable hash of the GrainId.
-Collisions are possible yet handled by careful use of locks on this index.
-*/
-CREATE INDEX IX_OrleansGrainDirectory_Lookup
+/* This turns the table into a CLUSTERED INDEX that allows duplication on the hash. */
+CREATE INDEX IX_OrleansGrainDirectory
 ON OrleansGrainDirectory
 (
-    ClusterId ASC,
-    ProviderId ASC,
-    GrainIdHash ASC
-)
-INCLUDE
-(
-    GrainId,
-    SiloAddress,
-    ActivationId,
-    CreatedOn
+    ClusterId,
+    ProviderId,
+    GrainIdHash
 );
 
-/* Registers a new grain activation */
 /* Registers a new grain activation */
 CREATE OR REPLACE FUNCTION RegisterGrainActivation(
     _ClusterId VARCHAR(150),
@@ -87,47 +75,53 @@ BEGIN
 -- this is required to prevent both duplication and deadlocks
 LOCK TABLE OrleansGrainDirectory IN EXCLUSIVE MODE;
 
-MERGE INTO OrleansGrainDirectory AS Target
-USING (SELECT _ClusterId, _ProviderId, _GrainIdHash, _GrainId, _SiloAddress, _ActivationId, _Now) AS Source
-ON
-    Target.ClusterId = Source._ClusterId
-    AND Target.ProviderId = Source._ProviderId
-    AND Target.GrainIdHash = Source._GrainIdHash
-    AND Target.GrainId = Source._GrainId
-WHEN NOT MATCHED THEN
-INSERT
-(
-    ClusterId,
-    ProviderId,
-    GrainIdHash,
-    GrainId,
-    SiloAddress,
-    ActivationId,
-    CreatedOn
-)
-VALUES
-(
-    Source._ClusterId,
-    Source._ProviderId,
-    Source._GrainIdHash,
-    Source._GrainId,
-    Source._SiloAddress,
-    Source._ActivationId,
-    Source._Now
-);
-
 RETURN QUERY
 SELECT
     ClusterId,
     ProviderId,
     GrainId,
     SiloAddress,
     ActivationId
-FROM OrleansGrainDirectory
-WHERE ClusterId = _ClusterId
+FROM
+    OrleansGrainDirectory
+WHERE
+    ClusterId = _ClusterId
     AND ProviderId = _ProviderId
     AND GrainIdHash = _GrainIdHash
-    AND GrainId = _GrainId;
+    AND GrainId = _GrainId
+FOR UPDATE;
+
+IF NOT FOUND THEN
+
+    RETURN QUERY
+    INSERT INTO OrleansGrainDirectory
+    (
+        ClusterId,
+        ProviderId,
+        GrainIdHash,
+        GrainId,
+        SiloAddress,
+        ActivationId,
+        CreatedOn
+    )
+    VALUES
+    (
+        _ClusterId,
+        _ProviderId,
+        _GrainIdHash,
+        _GrainId,
+        _SiloAddress,
+        _ActivationId,
+        _Now
+    )
+    RETURNING
+        ClusterId,
+        ProviderId,
+        GrainId,
+        SiloAddress,
+        ActivationId;
+
+END IF;
 
 END;
 $$;