Snowflake Query Acceleration Service (QAS) Complete Guide: SCALE_FACTOR Design and Cost Optimization

Query Acceleration Service (QAS) is a Snowflake feature that automatically allocates additional serverless compute resources to queries which scan large volumes of data and then drastically reduce the output through filtering or aggregation — cutting execution time as a result. QAS is enabled per warehouse and requires Enterprise Edition or higher.

How QAS Works

When a query runs on a QAS-enabled warehouse, Snowflake analyzes the query plan and detects the scan-then-filter-then-aggregate pattern. If the query is deemed eligible, Snowflake temporarily allocates serverless compute nodes — separate from the warehouse's own cluster — to parallelize the scan phase.

QAS doesn't change how the query executes — it adds resources to the scan phase to break through that bottleneck. As a result, queries that already scan very little data, or queries whose output row count is close to the input, see no benefit.

Enabling QAS via SQL

QAS is enabled on a per-warehouse basis. Set ENABLE_QUERY_ACCELERATION to TRUE and use QUERY_ACCELERATION_MAX_SCALE_FACTOR to cap the resource ceiling.

-- QASを有効化（SCALE_FACTOR = 8）
ALTER WAREHOUSE analytics_wh SET
  ENABLE_QUERY_ACCELERATION = TRUE
  QUERY_ACCELERATION_MAX_SCALE_FACTOR = 8;

-- SCALE_FACTORを変更
ALTER WAREHOUSE analytics_wh SET
  QUERY_ACCELERATION_MAX_SCALE_FACTOR = 4;

-- QASを無効化
ALTER WAREHOUSE analytics_wh SET
  ENABLE_QUERY_ACCELERATION = FALSE;

-- ウェアハウスの設定確認
SHOW WAREHOUSES LIKE 'ANALYTICS_WH';

Designing SCALE_FACTOR

QUERY_ACCELERATION_MAX_SCALE_FACTOR controls the upper limit on the serverless resources QAS can use. It allows resources up to N times the warehouse size.

Identifying QAS-Eligible Queries

QAS is not applied to every query. Understanding which queries benefit and which don't is essential.

Estimating the Acceleration Up Front

To estimate the QAS benefit for a specific query ID in advance, use the SYSTEM$ESTIMATE_QUERY_ACCELERATION function.

-- 特定クエリの加速効果を見積もり
SELECT PARSE_JSON(
  SYSTEM$ESTIMATE_QUERY_ACCELERATION('01abc123-0001-0002-0003-00000000abcd')
) AS estimate;

-- 結果例:
-- {
--   "estimatedQueryTimes": {
--     "1":  120,   -- SCALE_FACTOR=1 での推定時間(秒)
--     "2":  75,
--     "4":  50,
--     "8":  35,
--     "16": 30
--   },
--   "originalQueryTime": 180,
--   "status": "eligible",
--   "upperLimitScaleFactor": 16
-- }

-- QASが適用されたクエリの履歴
SELECT
  QUERY_ID,
  QUERY_ACCELERATION_BYTES_SCANNED,
  QUERY_ACCELERATION_PARTITIONS_SCANNED,
  QUERY_ACCELERATION_UPPER_LIMIT_SCALE_FACTOR
FROM SNOWFLAKE.ACCOUNT_USAGE.QUERY_HISTORY
WHERE QUERY_ACCELERATION_BYTES_SCANNED > 0
  AND START_TIME >= DATEADD('DAY', -7, CURRENT_TIMESTAMP())
ORDER BY QUERY_ACCELERATION_BYTES_SCANNED DESC;

Cost Model

QAS is billed as serverless credits. The cost is separate from the warehouse's normal credit consumption and is only incurred for the queries to which QAS actually allocated resources.

-- QASのクレジット消費履歴
SELECT
  START_TIME,
  END_TIME,
  WAREHOUSE_NAME,
  CREDITS_USED,
  NUM_FILES_SCANNED,
  NUM_BYTES_SCANNED
FROM SNOWFLAKE.ACCOUNT_USAGE.QUERY_ACCELERATION_HISTORY
WHERE START_TIME >= DATEADD('DAY', -30, CURRENT_TIMESTAMP())
ORDER BY CREDITS_USED DESC;

-- ウェアハウス別のQASコスト集計
SELECT
  WAREHOUSE_NAME,
  SUM(CREDITS_USED) AS total_qas_credits,
  COUNT(*) AS accelerated_queries
FROM SNOWFLAKE.ACCOUNT_USAGE.QUERY_ACCELERATION_HISTORY
WHERE START_TIME >= DATEADD('DAY', -30, CURRENT_TIMESTAMP())
GROUP BY WAREHOUSE_NAME
ORDER BY total_qas_credits DESC;

QAS vs. Multi-Cluster Warehouses

QAS and multi-cluster warehouses solve different problems.

• QAS: parallelizes the scan portion of a single heavy query to make it faster. It addresses the depth of a query.
• Multi-cluster: solves the problem of many concurrent queries piling up in the queue. It addresses the width of the workload.

If QUEUED_OVERLOAD_TIME is high, look into a multi-cluster warehouse. If individual queries have a long TOTAL_ELAPSED_TIME and scan large amounts of data, consider QAS.

Exam-Relevant Takeaways

• QAS is enabled per warehouse via ENABLE_QUERY_ACCELERATION = TRUE
• The default SCALE_FACTOR is 8; 0 means no upper limit
• Eligible queries: the large-scan → small-output (aggregation/filter) pattern
• Ineligible queries: SELECT *, DML, and queries against small data sets
• SYSTEM$ESTIMATE_QUERY_ACCELERATION lets you estimate the benefit in advance
• Requires Enterprise Edition or higher

Check Your Understanding

Frequently Asked Questions