feat(query-ochestrator): Reduce number of cache set for used flag (#9828)

Right now, we use both flags: used and touch. We already had LRU cache for a long time with the touch flag to reduce the number of updates in the Cache Store, but we didn't do it for the used.

This leads to a situation where, for a schema with 100 pre-aggs, we will do 200 cache sets per 1 minute because the default refresh interval is set to 30 seconds.

Author: ovr

packages/cubejs-backend-shared/src/env.ts

@@ -649,6 +649,13 @@ const variables: Record<string, (...args: any) => any> = {
     .default(8192)
     .asInt(),
 
+  /**
+   * Max number of elements
+   */
+  usedPreAggregationCacheMaxCount: (): number => get('CUBEJS_USED_PRE_AGG_CACHE_MAX_COUNT')
+    .default(8192)
+    .asInt(),
+
   /**
    * Max number of elements
    */

packages/cubejs-query-orchestrator/src/orchestrator/PreAggregations.ts

@@ -242,11 +242,11 @@ type PreAggregationQueryBody = QueryBody & {
 };
 
 export class PreAggregations {
-  public options: PreAggregationsOptions;
+  public readonly options: PreAggregationsOptions;
 
-  public externalDriverFactory: DriverFactory;
+  public readonly externalDriverFactory: DriverFactory;
 
-  public structureVersionPersistTime: any;
+  public readonly structureVersionPersistTime: any;
 
   private readonly touchTablePersistTime: number;
 
@@ -260,6 +260,8 @@ export class PreAggregations {
 
   private readonly queue: Record<string, QueryQueue> = {};
 
+  private readonly usedCache: LRUCache<string, true>;
+
   private readonly touchCache: LRUCache<string, true>;
 
   public constructor(
@@ -277,6 +279,25 @@ export class PreAggregations {
     this.dropPreAggregationsWithoutTouch = options.dropPreAggregationsWithoutTouch || getEnv('dropPreAggregationsWithoutTouch');
     this.usedTablePersistTime = options.usedTablePersistTime || getEnv('dbQueryTimeout');
     this.externalRefresh = options.externalRefresh;
+
+    /**
+     * Comment for both caches: used and touch:
+     *
+     * Memory usage: By default, it defines max as 8096 keys,
+     * let's assume that avg key size is 64 symbols, it's around 100 bytes
+     * with V8's internal stuff:
+     *
+     * 100 x 8096 = 809 bytes, max 1Mb in memory.
+     *
+     * However, this is a theoretical limit. In practice, even a couple of thousand
+     * is a very large number
+     */
+    this.usedCache = new LRUCache({
+      max: getEnv('usedPreAggregationCacheMaxCount'),
+      ttl: Math.round(this.usedTablePersistTime / 2) * 1000,
+      allowStale: false,
+      updateAgeOnGet: false
+    });
     this.touchCache = new LRUCache({
       max: getEnv('touchPreAggregationCacheMaxCount'),
       ttl: getEnv('touchPreAggregationCacheMaxAge') * 1000,
@@ -301,11 +322,23 @@ export class PreAggregations {
   }
 
   public async addTableUsed(tableName: string): Promise<void> {
-    await this.queryCache.getCacheDriver().set(
-      this.tablesUsedRedisKey(tableName),
-      true,
-      this.usedTablePersistTime
-    );
+    if (this.usedCache.has(tableName)) {
+      return;
+    }
+
+    try {
+      this.usedCache.set(tableName, true);
+
+      await this.queryCache.getCacheDriver().set(
+        this.tablesUsedRedisKey(tableName),
+        true,
+        this.usedTablePersistTime
+      );
+    } catch (e: unknown) {
+      this.usedCache.delete(tableName);
+
+      throw e;
+    }
   }
 
   public async tablesUsed() {

packages/cubejs-query-orchestrator/src/orchestrator/QueryCache.ts

@@ -130,7 +130,7 @@ export class QueryCache {
   protected memoryCache: LRUCache<string, CacheEntry>;
 
   public constructor(
-    protected readonly redisPrefix: string,
+    protected readonly cachePrefix: string,
     protected readonly driverFactory: DriverFactoryByDataSource,
     protected readonly logger: any,
     public readonly options: QueryCacheOptions
@@ -165,11 +165,7 @@ export class QueryCache {
   }
 
   public getKey(catalog: string, key: string): string {
-    if (this.cacheDriver instanceof CubeStoreCacheDriver) {
-      return `${this.redisPrefix}#${catalog}:${key}`;
-    } else {
-      return `${catalog}_${this.redisPrefix}_${key}`;
-    }
+    return `${this.cachePrefix}#${catalog}:${key}`;
   }
 
   /**
@@ -469,7 +465,7 @@ export class QueryCache {
       const queueOptions = await this.options.queueOptions(dataSource);
       if (!this.queue[dataSource]) {
         this.queue[dataSource] = QueryCache.createQueue(
-          `SQL_QUERY_${this.redisPrefix}_${dataSource}`,
+          `SQL_QUERY_${this.cachePrefix}_${dataSource}`,
           () => this.driverFactory(dataSource),
           (client, req) => {
             this.logger('Executing SQL', { ...req });
@@ -537,7 +533,7 @@ export class QueryCache {
   public getExternalQueue() {
     if (!this.externalQueue) {
       this.externalQueue = QueryCache.createQueue(
-        `SQL_QUERY_EXT_${this.redisPrefix}`,
+        `SQL_QUERY_EXT_${this.cachePrefix}`,
         this.options.externalDriverFactory,
         (client, q) => {
           this.logger('Executing SQL', {