From 3edbf60e9c1d3070ffe6e1e70bea39c688f39a21 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Fri, 18 Sep 2026 14:58:34 -0700
Subject: [PATCH] fix(proxy): requeue the daily tag rollup on commit failure
 without the Redis buffer

---
 litellm/proxy/db/db_spend_update_writer.py    | 20 +++++--------
 .../proxy/db/test_db_spend_update_writer.py   | 29 +++++++++++++++++++
 2 files changed, 37 insertions(+), 12 deletions(-)

diff --git a/litellm/proxy/db/db_spend_update_writer.py b/litellm/proxy/db/db_spend_update_writer.py
index a4f1713d61d..b2e6f9dc54d 100644
--- a/litellm/proxy/db/db_spend_update_writer.py
+++ b/litellm/proxy/db/db_spend_update_writer.py
@@ -1305,7 +1305,7 @@ class DBSpendUpdateWriter:
     async def _flush_daily_spend_queue(
         self,
         queue: DailySpendUpdateQueue,
-        entity_type: Literal["user", "team", "org", "end_user", "agent"],
+        entity_type: Literal["user", "team", "org", "tag", "end_user", "agent"],
         commit: _DailySpendCommit[_DailySpendTransactionT],
         n_retry_times: int,
         prisma_client: PrismaClient,
@@ -1447,19 +1447,15 @@ class DBSpendUpdateWriter:
         Commit only tag spend updates to database.
         This is called by a separate scheduler job at a longer interval.
         """
-        daily_tag_spend_update_transactions: Final = cast(
-            dict[str, DailyTagSpendTransaction],
-            await self.daily_tag_spend_update_queue.flush_and_get_aggregated_daily_spend_update_transactions(),
+        await self._flush_daily_spend_queue(
+            queue=self.daily_tag_spend_update_queue,
+            entity_type="tag",
+            commit=DBSpendUpdateWriter.update_daily_tag_spend,
+            n_retry_times=n_retry_times,
+            prisma_client=prisma_client,
+            proxy_logging_obj=proxy_logging_obj,
         )
 
-        if daily_tag_spend_update_transactions:
-            await DBSpendUpdateWriter.update_daily_tag_spend(
-                n_retry_times=n_retry_times,
-                prisma_client=prisma_client,
-                proxy_logging_obj=proxy_logging_obj,
-                daily_spend_transactions=daily_tag_spend_update_transactions,
-            )
-
     async def _commit_daily_tag_spend_to_db_with_redis(
         self,
         prisma_client: PrismaClient,
diff --git a/tests/test_litellm/proxy/db/test_db_spend_update_writer.py b/tests/test_litellm/proxy/db/test_db_spend_update_writer.py
index 3feb117edb7..b27b838133b 100644
--- a/tests/test_litellm/proxy/db/test_db_spend_update_writer.py
+++ b/tests/test_litellm/proxy/db/test_db_spend_update_writer.py
@@ -2864,6 +2864,35 @@ async def test_failed_daily_spend_commit_requeues_the_rows_and_flushes_the_other
     assert db_writer.daily_spend_update_queue.update_queue.empty()
 
 
+@pytest.mark.asyncio
+async def test_failed_daily_tag_spend_commit_requeues_the_rows():
+    """The tag rollup drains on its own scheduler job with the same no-Redis drop:
+    a failed LiteLLM_DailyTagSpend commit has to put the rows back for the next tick."""
+    db_writer = DBSpendUpdateWriter()
+    tag_txn = {key: value for key, value in _daily_txn().items() if key != "user_id"} | {"tag": "tag-1"}
+    await db_writer.daily_tag_spend_update_queue.add_update({"tag-key": tag_txn})
+    db = _DailySpendFakeDB(failing_table="LiteLLM_DailyTagSpend")
+    proxy_logging_obj = MagicMock()
+    proxy_logging_obj.failure_handler = AsyncMock()
+
+    await db_writer._commit_daily_tag_spend_to_db(
+        prisma_client=_WindowSpendFakePrisma(db), n_retry_times=0, proxy_logging_obj=proxy_logging_obj
+    )
+
+    assert _daily_upserts(db, "LiteLLM_DailyTagSpend") == []
+    assert not db_writer.daily_tag_spend_update_queue.update_queue.empty()
+
+    db.failing_table = None
+    await db_writer._commit_daily_tag_spend_to_db(
+        prisma_client=_WindowSpendFakePrisma(db), n_retry_times=0, proxy_logging_obj=proxy_logging_obj
+    )
+
+    (tag_upsert,) = _daily_upserts(db, "LiteLLM_DailyTagSpend")
+    assert _row_values(tag_upsert, "tag") == ["tag-1"]
+    assert _row_values(tag_upsert, "spend") == [0.1]
+    assert db_writer.daily_tag_spend_update_queue.update_queue.empty()
+
+
 @pytest.mark.asyncio
 async def test_failed_window_spend_commit_from_redis_is_restored_to_redis():
     """The Redis drain is destructive, so a failed window commit has to push