perf(webapp,clickhouse): move runs empty-state check to ClickHouse

.server-changes/runs-empty-state-clickhouse.md

@@ -0,0 +1,6 @@
+---
+area: webapp
+type: improvement
+---
+
+Reduce primary database load on the runs page by serving its empty-state check from ClickHouse instead of Postgres.

apps/webapp/app/env.server.ts

@@ -330,6 +330,13 @@ const EnvironmentSchema = z
       .string()
       .default(process.env.REDIS_TLS_DISABLED ?? "false"),
     TASK_META_CACHE_CURRENT_ENV_TTL_SECONDS: z.coerce.number().default(86400),
+
+    // Runs-list empty-state check: how far back the ClickHouse "does this env have any run"
+    // probe looks. Bounds the prove-absence partition scan. 0 = unbounded ("any run ever").
+    RUN_LIST_HAS_RUNS_LOOKBACK_DAYS: z.coerce.number().default(30),
+    // SWR TTLs for the empty-state has-runs cache (memory + Redis).
+    RUN_LIST_HAS_RUNS_CACHE_FRESH_MS: z.coerce.number().default(86_400_000),
+    RUN_LIST_HAS_RUNS_CACHE_STALE_MS: z.coerce.number().default(604_800_000),
     TASK_META_CACHE_BY_WORKER_TTL_SECONDS: z.coerce.number().default(2592000),
 
     REALTIME_STREAMS_REDIS_HOST: z

apps/webapp/app/presenters/v3/NextRunListPresenter.server.ts

@@ -11,11 +11,47 @@ import { timeFilters } from "~/components/runs/v3/SharedFilters";
 import { findDisplayableEnvironment } from "~/models/runtimeEnvironment.server";
 import { getTaskIdentifiers } from "~/models/task.server";
 import { RunsRepository } from "~/services/runsRepository/runsRepository.server";
+import { env } from "~/env.server";
+import {
+  createCache,
+  createLRUMemoryStore,
+  DefaultStatefulContext,
+  Namespace,
+} from "@internal/cache";
+import { RedisCacheStore } from "~/services/unkey/redisCacheStore.server";
+import { singleton } from "~/utils/singleton";
 import { regionForDisplay } from "~/runEngine/concerns/workerQueueSplit.server";
 import { machinePresetFromRun } from "~/v3/machinePresets.server";
 import { ServiceValidationError } from "~/v3/services/baseService.server";
 import { isCancellableRunStatus, isFinalRunStatus, isPendingRunStatus } from "~/v3/taskStatus";
 
+// Positive-only cache: only envs known to have runs are stored (empty envs are re-checked),
+// so "has runs" is monotonic and the TTL can be very long. Tiered memory + Redis.
+const runsExistCache = singleton("runsExistCache", () => {
+  const ctx = new DefaultStatefulContext();
+  const memory = createLRUMemoryStore(5000, "runs-has-runs-cache");
+  const redis = new RedisCacheStore({
+    name: "runs-has-runs",
+    connection: {
+      keyPrefix: "tr:cache:runs-has-runs",
+      port: env.CACHE_REDIS_PORT,
+      host: env.CACHE_REDIS_HOST,
+      username: env.CACHE_REDIS_USERNAME,
+      password: env.CACHE_REDIS_PASSWORD,
+      tlsDisabled: env.CACHE_REDIS_TLS_DISABLED === "true",
+      clusterMode: env.CACHE_REDIS_CLUSTER_MODE_ENABLED === "1",
+    },
+  });
+
+  return createCache({
+    hasRuns: new Namespace<boolean>(ctx, {
+      stores: [memory, redis],
+      fresh: env.RUN_LIST_HAS_RUNS_CACHE_FRESH_MS,
+      stale: env.RUN_LIST_HAS_RUNS_CACHE_STALE_MS,
+    }),
+  });
+});
+
 export type RunListOptions = {
   userId?: string;
   projectId: string;
@@ -42,6 +78,8 @@ export type RunListOptions = {
   direction?: Direction;
   cursor?: string;
   pageSize?: number;
+  // Run the empty-state "has any run ever" probe. Only the runs list consumes it.
+  includeHasAnyRuns?: boolean;
 };
 
 const DEFAULT_PAGE_SIZE = 25;
@@ -65,37 +103,31 @@ export class NextRunListPresenter {
     }
   ) {}
 
-  // Existence probe for the empty-state. run-ops (TaskRun) read.
-  // Split off / single-DB: one plain findFirst on `replica` (passthrough).
-  // Split on: true if a row exists in the NEW run-ops DB OR the LEGACY run-ops
-  // read replica only (never the legacy writer). New first to avoid touching
-  // legacy whenever the new DB already answers.
-  async #anyRunExistsInEnv(environmentId: string): Promise<boolean> {
-    const splitEnabled = this.readThroughDeps?.splitEnabled ?? false;
+  // Empty-state existence probe, served from ClickHouse (same connection as the runs
+  // list) so it no longer scans TaskRun in Postgres. SWR-cached to spare ClickHouse;
+  // RUN_LIST_HAS_RUNS_LOOKBACK_DAYS bounds the prove-absence partition scan.
+  async #anyRunExistsInEnv(
+    runsRepository: RunsRepository,
+    organizationId: string,
+    projectId: string,
+    environmentId: string
+  ): Promise<boolean> {
+    const lookbackDays = env.RUN_LIST_HAS_RUNS_LOOKBACK_DAYS;
+    const createdAtLowerBoundMs =
+      lookbackDays > 0 ? Date.now() - lookbackDays * 24 * 60 * 60 * 1000 : undefined;
 
-    if (!splitEnabled) {
-      const firstRun = await this.replica.taskRun.findFirst({
-        where: { runtimeEnvironmentId: environmentId },
-        select: { id: true },
+    const result = await runsExistCache.hasRuns.swr(environmentId, async () => {
+      const exists = await runsRepository.runExistsInEnvironment({
+        organizationId,
+        projectId,
+        environmentId,
+        createdAtLowerBoundMs,
       });
-      return firstRun !== null;
-    }
-
-    const newClient = this.readThroughDeps?.newClient ?? this.replica;
-    const newRun = await newClient.taskRun.findFirst({
-      where: { runtimeEnvironmentId: environmentId },
-      select: { id: true },
+      // undefined (not false) so swr does NOT cache the empty result — re-check until a run exists.
+      return exists ? true : undefined;
     });
-    if (newRun) {
-      return true;
-    }
 
-    const legacyReplica = this.readThroughDeps?.legacyReplica ?? this.replica;
-    const legacyRun = await legacyReplica.taskRun.findFirst({
-      where: { runtimeEnvironmentId: environmentId },
-      select: { id: true },
-    });
-    return legacyRun !== null;
+    return result.val ?? false;
   }
 
   public async call(
@@ -125,6 +157,7 @@ export class NextRunListPresenter {
       direction = "forward",
       cursor,
       pageSize = DEFAULT_PAGE_SIZE,
+      includeHasAnyRuns = false,
     }: RunListOptions
   ) {
     //get the time values from the raw values (including a default period)
@@ -252,8 +285,13 @@ export class NextRunListPresenter {
 
     let hasAnyRuns = runs.length > 0;
 
-    if (!hasAnyRuns) {
-      hasAnyRuns = await this.#anyRunExistsInEnv(environmentId);
+    if (!hasAnyRuns && includeHasAnyRuns) {
+      hasAnyRuns = await this.#anyRunExistsInEnv(
+        runsRepository,
+        organizationId,
+        projectId,
+        environmentId
+      );
     }
 
     return {

apps/webapp/app/routes/_app.orgs.$organizationSlug.projects.$projectParam.env.$envParam.runs._index/route.tsx

@@ -111,6 +111,7 @@ export const loader = async ({ request, params }: LoaderFunctionArgs) => {
     userId,
     projectId: project.id,
     ...filters,
+    includeHasAnyRuns: true,
   });
 
   // Only persist rootOnly when no tasks are filtered. While a task filter is active,

apps/webapp/app/services/runsRepository/clickhouseRunsRepository.server.ts

@@ -35,6 +35,38 @@ export class ClickHouseRunsRepository implements IRunsRepository {
     return "clickhouse";
   }
 
+  async runExistsInEnvironment(options: {
+    organizationId: string;
+    projectId: string;
+    environmentId: string;
+    createdAtLowerBoundMs?: number;
+  }): Promise<boolean> {
+    const queryBuilder = this.options.clickhouse.taskRuns.existsQueryBuilder();
+
+    queryBuilder
+      .where("organization_id = {organizationId: String}", {
+        organizationId: options.organizationId,
+      })
+      .where("project_id = {projectId: String}", { projectId: options.projectId })
+      .where("environment_id = {environmentId: String}", { environmentId: options.environmentId });
+
+    if (typeof options.createdAtLowerBoundMs === "number") {
+      queryBuilder.where("created_at >= fromUnixTimestamp64Milli({createdAtLowerBound: Int64})", {
+        createdAtLowerBound: options.createdAtLowerBoundMs,
+      });
+    }
+
+    queryBuilder.limit(1);
+
+    const [queryError, result] = await queryBuilder.execute();
+
+    if (queryError) {
+      throw queryError;
+    }
+
+    return (result?.length ?? 0) > 0;
+  }
+
   /**
    * Runs the keyset-paginated query and returns `{ runId, createdAt }` rows
    * (one extra beyond `page.size` to signal "has more"). The ordering is always

apps/webapp/app/services/runsRepository/runsRepository.server.ts

@@ -176,6 +176,12 @@ export interface IRunsRepository {
   }>;
   countRuns(options: RunListInputOptions): Promise<number>;
   listTags(options: TagListOptions): Promise<TagList>;
+  runExistsInEnvironment(options: {
+    organizationId: string;
+    projectId: string;
+    environmentId: string;
+    createdAtLowerBoundMs?: number;
+  }): Promise<boolean>;
 }
 
 export class RunsRepository implements IRunsRepository {
@@ -189,6 +195,26 @@ export class RunsRepository implements IRunsRepository {
     return "runsRepository";
   }
 
+  async runExistsInEnvironment(options: {
+    organizationId: string;
+    projectId: string;
+    environmentId: string;
+    createdAtLowerBoundMs?: number;
+  }): Promise<boolean> {
+    return startActiveSpan(
+      "runsRepository.runExistsInEnvironment",
+      async () => this.clickHouseRunsRepository.runExistsInEnvironment(options),
+      {
+        attributes: {
+          "repository.name": "clickhouse",
+          organizationId: options.organizationId,
+          projectId: options.projectId,
+          environmentId: options.environmentId,
+        },
+      }
+    );
+  }
+
   async listRunIds(options: ListRunsOptions): Promise<RunIdsPage> {
     return startActiveSpan(
       "runsRepository.listRunIds",