Skip to content

:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

fix(notebooks): reject logs-cluster tables in frame materialization (#118099)

posthog[bot]1c82ed6cb878
diff --git a/products/notebooks/backend/temporal/frame_materialize.py b/products/notebooks/backend/temporal/frame_materialize.py
index 9eb383f9cadd..5bec8408a0d3 100644
--- a/products/notebooks/backend/temporal/frame_materialize.py
+++ b/products/notebooks/backend/temporal/frame_materialize.py
@@ -53,7 +53,7 @@ from posthog.schema import QueryStatus
from posthog.hogql import ast
from posthog.hogql.constants import HogQLGlobalSettings, LimitContext
from posthog.hogql.context import HogQLContext
-from posthog.hogql.errors import ExposedHogQLError
+from posthog.hogql.errors import ExposedHogQLError, QueryError
from posthog.hogql.parser import parse_select
from posthog.hogql.query import HogQLQueryExecutor
from posthog.hogql.visitor import clone_expr
@@ -157,6 +157,9 @@ _RESULT_SIZE_MESSAGE = (
"The materialized result is too large (over the frame size budget). "
"Select fewer columns or aggregate before materializing."
)
+_LOGS_CLUSTER_MESSAGE = (
+ "Notebooks can't load logs, trace spans, or metrics yet. Run this query in the SQL editor instead."
+)
_MID_STREAM_ERROR_MESSAGE = "The query failed while its result was streaming. Adjust it and re-run."
# ClickHouse exception codes worth a specific user-facing message when a query dies
# mid-stream: 158 TOO_MANY_ROWS, 241 MEMORY_LIMIT_EXCEEDED, 307 TOO_MANY_BYTES,
@@ -432,6 +435,10 @@ def _generate_sql(
context.values.clear()
sql, resolved = executor.generate_clickhouse_sql()
seconds = time.perf_counter() - started
+ # Both write paths connect to the main cluster, and logs-cluster tables exist only on the
+ # logs cluster. Fail before the DESCRIBE so the user sees why, not an unknown-table error.
+ if resolved.workload == Workload.LOGS:
+ raise QueryError(_LOGS_CLUSTER_MESSAGE)
# The executor keeps the database it built on its own `dataclasses.replace` copy, so
# without this hand-back the shared context still looks empty and the next pass builds a
# second one. It resets the per-execution state it shares by reference, so a database
diff --git a/products/notebooks/backend/test/test_frame_materialize.py b/products/notebooks/backend/test/test_frame_materialize.py
index 9e7b338ffbb2..cae85c1391e3 100644
--- a/products/notebooks/backend/test/test_frame_materialize.py
+++ b/products/notebooks/backend/test/test_frame_materialize.py
@@ -150,6 +150,35 @@ class TestFrameMaterializeEnqueue(APIBaseTest):
self.assertTrue(status.complete and status.error)
self.assertIn(expected_message, status.error_message or "")
+ @parameterized.expand(
+ [
+ ("trace_spans_streaming", "select 1 from posthog.trace_spans", False),
+ ("logs_ch_writes", "select 1 from logs", True),
+ ("metrics_streaming", "select 1 from posthog.metrics", False),
+ ]
+ )
+ def test_logs_cluster_table_is_rejected_before_clickhouse(self, _name, query, ch_writes):
+ # These tables exist only on the logs cluster, and neither write path connects there.
+ # Without the early rejection the DESCRIBE fails with an unknown-table error.
+ inputs, manager = _registered_inputs(
+ self.team.id, self.notebook.short_id, self.user.id, query=query, ch_writes=ch_writes
+ )
+
+ with (
+ patch.object(frame_materialize, "_materialize_slots"),
+ patch.object(frame_materialize, "_describe_columns") as describe,
+ patch.object(frame_materialize, "_describe_columns_pooled") as describe_pooled,
+ ):
+ with self.assertRaises(exceptions.ApplicationError) as caught:
+ frame_materialize.materialize_frame(inputs)
+
+ describe.assert_not_called()
+ describe_pooled.assert_not_called()
+ self.assertTrue(caught.exception.non_retryable)
+ status = manager.get_query_status()
+ self.assertTrue(status.complete and status.error)
+ self.assertIn("SQL editor", status.error_message or "")
+
def test_mid_stream_failure_removes_the_corrupt_object_and_surfaces_the_real_error(self):
# ClickHouse streams 200 before execution finishes; a mid-stream failure can close
# the body cleanly with exception text appended instead of tearing the read. Without