refinement; add to changelog

CHANGELOG.md

@@ -59,6 +59,10 @@ The index had low cardinality and workflow pickup is faster without it. Migratio
 
 The `IX_METADATA_ENTRY_WEU_MK` index is added to `METADATA_ENTRY`. In pre-release testing, the migration proceeded at about 3 million rows per minute. Please plan downtime accordingly.
 
+#### Unique constraint addition
+
+The `UC_GROUP_METRICS_ENTRY_GI` constraint has been added to column `GROUP_ID` in `GROUP_METRICS_ENTRY` table. This additionally enforces at database level that the `GROUP_ID` column always contains unique values.
+
 ### Reduce errors from boot disk filling up on Google Lifesciences API
 
  * If Cromwell can't determine the size of the user command Docker image, it will increase Lifesciences API boot disk size by 30GB rather than 0. This should reduce incidence of tasks failing due to boot disk filling up.

database/sql/src/main/scala/cromwell/database/slick/GroupMetricsSlickDatabase.scala

@@ -25,17 +25,19 @@ trait GroupMetricsSlickDatabase extends GroupMetricsSqlDatabase {
       runTransaction(updateAction, TransactionIsolation.ReadCommitted)
     }
 
-    /* The approach here is to try and insert the record into database in 1 transaction and if that fails, because
-       a record with that group_id already exists, then it will update that record with new quota_exhaustion_detected
-       timestamp in a separate transaction. The reason for 2 separate transactions are because when manual upsert or
-       Slick's insertOrUpdate is performed in a single transaction, and there are threads trying to update the table
-       concurrently it results in a deadlock (even with stricter transaction isolation level Serializable). This
-       happens because it gets a gap lock on index IX_GROUP_METRICS_ENTRY_GI which locks the gaps between index records
-       within the page and as a result it runs into a deadlock when there are transactions trying to insert/update for
-       same group ID or for group IDs that exist in the locked records.
-       See https://broadworkbench.atlassian.net/browse/AN-286 and https://broadworkbench.atlassian.net/browse/WX-1847.
+    /*
+      The approach here is to try and insert the record into the table and if that fails because a record with that
+      group_id already exists, then it will update that record with new quota_exhaustion_detected timestamp.
+      The insert and update happen in 2 different transactions. Using 2 separate transactions avoids deadlocks that
+      occur during upserts or when using Slick's insertOrUpdate in a single transaction. Deadlocks occur in scenarios
+      with concurrent threads trying to update the table, even under the stricter Serializable transaction isolation
+      level. This is caused by a gap lock on the IX_GROUP_METRICS_ENTRY_GI index, which locks gaps between index
+      records on a page, leading to deadlocks for transactions involving the same or nearby group_id values.
+      See https://broadworkbench.atlassian.net/browse/AN-286 and https://broadworkbench.atlassian.net/browse/WX-1847.
 
-       Note: a unique constraint on group_id also exists in database.
+      This approach should have minimal performance impact since the table has only 3 columns and low cardinality.
+
+      Note: a unique constraint on group_id also exists in database.
      */
     val insertAction = for {
       _ <- dataAccess.groupMetricsEntryIdsAutoInc += groupMetricsEntry