fix(webapp): restore Postgres fallback for non-ClickHouse OTLP spans

.server-changes/otlp-postgres-store-fallback.md

@@ -0,0 +1,6 @@
+---
+area: webapp
+type: fix
+---
+
+Fixes OTLP ingest endpoints returning HTTP 500 for runs on environments that use a Postgres-backed task event store. This caused the OpenTelemetry collector to drop entire span batches as non-retryable, resulting in real span loss.

apps/webapp/app/v3/otlpExporter.server.ts

@@ -24,6 +24,7 @@ import type { ClickhouseFactory } from "~/services/clickhouse/clickhouseFactory.
 import { clickhouseFactory } from "~/services/clickhouse/clickhouseFactoryInstance.server";
 
 import { generateSpanId } from "./eventRepository/common.server";
+import { eventRepository } from "./eventRepository/eventRepository.server";
 import type {
   CreatableEventKind,
   CreatableEventStatus,
@@ -120,10 +121,18 @@ class OTLPExporter {
         const routeKey = `${event.organizationId}\0${taskEventStore}`;
         let resolved = routeCache.get(routeKey);
         if (!resolved) {
-          resolved = this._clickhouseFactory.getEventRepositoryForOrganizationSync(
-            taskEventStore,
-            event.organizationId
-          );
+          // Non-ClickHouse stores (taskEvent / taskEventPartitioned) are Postgres-backed.
+          // The ClickHouse factory only handles clickhouse/clickhouse_v2 and throws otherwise.
+          if (taskEventStore !== "clickhouse" && taskEventStore !== "clickhouse_v2") {
+            // Non-ClickHouse stores (taskEvent / taskEventPartitioned) are Postgres-backed.
+            // The ClickHouse factory only handles clickhouse/clickhouse_v2 and throws otherwise.
+            resolved = { key: "postgres:default", repository: eventRepository };
+          } else {
+            resolved = this._clickhouseFactory.getEventRepositoryForOrganizationSync(
+              taskEventStore,
+              event.organizationId
+            );
+          }
           routeCache.set(routeKey, resolved);
         }