Compartilhar via

Consultas de exemplo para monitorar a atividade do SQL Warehouse

Use estes exemplos de consultas SQL com tabelas do sistema para monitorar o desempenho, o uso e os custos do SQL Warehouse. Modifique as consultas para atender às necessidades da sua organização. Adicione alertas para ser notificado sobre valores inesperados.

Requisitos

  • Você deve ter acesso a tabelas do sistema. Consulte a referência de tabelas do sistema para obter requisitos.
  • A maioria das tabelas do sistema exige que a conta tenha o Catálogo do Unity habilitado.

Tabelas para monitoramento do SQL Warehouse

Tabela do sistema Descrição
system.compute.warehouse_events Rastreia eventos de início, interrupção, ampliação e diminuição do armazém.
system.compute.warehouses Contém instantâneos de configurações de armazém.
system.query.history Registra detalhes sobre cada consulta executada em sql warehouses.
system.billing.usage Contém registros de cobrança para todo o uso Azure Databricks.

Exemplo: uso do armazém

Use as consultas a seguir para entender como seu warehouse está sendo usado, incluindo quais consultas, usuários e aplicativos impulsionam a maior parte da atividade.

Localizar as consultas mais lentas em um armazém

SELECT
  statement_id,
  executed_by,
  statement_type,
  execution_status,
  total_duration_ms,
  execution_duration_ms,
  compilation_duration_ms,
  waiting_at_capacity_duration_ms,
  read_rows,
  produced_rows,
  start_time,
  statement_text
FROM
  system.query.history
WHERE
  compute.warehouse_id = '<warehouse-id>'
  AND start_time >= NOW() - INTERVAL 1 DAY
ORDER BY
  total_duration_ms DESC
LIMIT 50

SELECT
  DATE(start_time) AS query_date,
  COUNT(*) AS total_queries,
  COUNT(CASE WHEN execution_status = 'FINISHED' THEN 1 END) AS successful_queries,
  COUNT(CASE WHEN execution_status = 'FAILED' THEN 1 END) AS failed_queries,
  ROUND(AVG(total_duration_ms), 0) AS avg_duration_ms,
  ROUND(PERCENTILE(total_duration_ms, 0.5), 0) AS p50_duration_ms,
  ROUND(PERCENTILE(total_duration_ms, 0.95), 0) AS p95_duration_ms,
  ROUND(AVG(waiting_at_capacity_duration_ms), 0) AS avg_queue_wait_ms
FROM
  system.query.history
WHERE
  compute.warehouse_id = '<warehouse-id>'
  AND start_time >= NOW() - INTERVAL 30 DAY
GROUP BY
  DATE(start_time)
ORDER BY
  query_date DESC

Localizar os usuários mais ativos em um warehouse

SELECT
  executed_by,
  COUNT(*) AS query_count,
  ROUND(SUM(total_duration_ms) / 1000 / 60, 2) AS total_duration_minutes,
  ROUND(AVG(total_duration_ms), 0) AS avg_duration_ms
FROM
  system.query.history
WHERE
  compute.warehouse_id = '<warehouse-id>'
  AND start_time >= NOW() - INTERVAL 7 DAY
GROUP BY
  executed_by
ORDER BY
  query_count DESC

Localizar os principais aplicativos cliente

SELECT
  client_application,
  CASE
    WHEN query_source.job_info.job_id IS NOT NULL THEN 'Job'
    WHEN query_source.dashboard_id IS NOT NULL THEN 'Dashboard'
    WHEN query_source.alert_id IS NOT NULL THEN 'Alert'
    WHEN query_source.notebook_id IS NOT NULL THEN 'Notebook'
    WHEN query_source.genie_space_id IS NOT NULL THEN 'Genie Space'
    WHEN query_source.sql_query_id IS NOT NULL THEN 'SQL Editor'
    ELSE 'Other'
  END AS source_type,
  COUNT(*) AS query_count,
  ROUND(AVG(total_duration_ms), 0) AS avg_duration_ms
FROM
  system.query.history
WHERE
  compute.warehouse_id = '<warehouse-id>'
  AND start_time >= NOW() - INTERVAL 7 DAY
GROUP BY
  client_application,
  source_type
ORDER BY
  query_count DESC

Monitorar interrogações falhas

SELECT
  DATE(start_time) AS failure_date,
  execution_status,
  error_message,
  COUNT(*) AS failure_count,
  COLLECT_SET(executed_by) AS affected_users
FROM
  system.query.history
WHERE
  compute.warehouse_id = '<warehouse-id>'
  AND execution_status IN ('FAILED', 'CANCELED')
  AND start_time >= NOW() - INTERVAL 7 DAY
GROUP BY
  DATE(start_time),
  execution_status,
  error_message
ORDER BY
  failure_date DESC,
  failure_count DESC

Exemplo: Dimensionamento do warehouse

Use as consultas a seguir para determinar se o armazém está dimensionado corretamente. Consultas aguardando devido à capacidade sugerem que você precisa aumentar max_clusters. Consultas que causam derramamento excessivo de disco sugerem que você precisa aumentar o tamanho do armazém de dados.

Identificar consultas aguardando devido à capacidade máxima

Consultas com valores altos waiting_at_capacity_duration_ms estão gastando tempo na fila em vez de serem executadas. Considere aumentar a configuração do armazém max_clusters para promover a escalabilidade do armazém.

SELECT
  statement_id,
  executed_by,
  total_duration_ms,
  waiting_at_capacity_duration_ms,
  execution_duration_ms,
  start_time,
  statement_text
FROM
  system.query.history
WHERE
  compute.warehouse_id = '<warehouse-id>'
  AND start_time >= NOW() - INTERVAL 7 DAY
  AND waiting_at_capacity_duration_ms > 0
ORDER BY
  waiting_at_capacity_duration_ms DESC
LIMIT 50

Identificar consultas com derramamento excessivo de disco

O vazamento de disco ocorre quando uma consulta requer mais memória do que está disponível. Considere ampliar o espaço do armazém para fornecer mais capacidade de memória às consultas. O derramamento excessivo normalmente significa que as consultas precisam de otimização ou que o tamanho do data warehouse é muito pequeno para o volume de trabalho.

SELECT
  statement_id,
  executed_by,
  spilled_local_bytes / (1024 * 1024) AS spilled_mb,
  read_bytes / (1024 * 1024) AS read_mb,
  total_duration_ms,
  start_time,
  statement_text
FROM
  system.query.history
WHERE
  compute.warehouse_id = '<warehouse-id>'
  AND start_time >= NOW() - INTERVAL 7 DAY
  AND spilled_local_bytes > 0
ORDER BY
  spilled_local_bytes DESC
LIMIT 50

Exemplo: custos do warehouse

Use as consultas a seguir para entender e acompanhar os custos associados aos seus sql warehouses.

Monitorar o custo do armazém por dia

SELECT
  usage_date,
  sku_name,
  ROUND(SUM(usage_quantity), 2) AS total_dbus,
  ROUND(SUM(usage_quantity * list_prices.pricing.default), 2) AS estimated_list_cost
FROM
  system.billing.usage
  LEFT JOIN system.billing.list_prices ON usage.sku_name = list_prices.sku_name
    AND price_end_time IS NULL
WHERE
  usage_metadata.warehouse_id = '<warehouse-id>'
  AND usage_date >= NOW() - INTERVAL 30 DAY
GROUP BY
  usage_date,
  sku_name
ORDER BY
  usage_date DESC

Correlacionar eventos do armazém com o volume de consultas

Essa consulta ajuda você a entender a relação entre eventos de dimensionamento de warehouse e atividade de consulta para identificar oportunidades de otimização de custos.

WITH hourly_events AS (
  SELECT
    DATE_TRUNC('hour', event_time) AS event_hour,
    warehouse_id,
    MAX(cluster_count) AS max_clusters,
    COLLECT_SET(event_type) AS event_types
  FROM
    system.compute.warehouse_events
  WHERE
    warehouse_id = '<warehouse-id>'
    AND event_time >= NOW() - INTERVAL 7 DAY
  GROUP BY
    DATE_TRUNC('hour', event_time),
    warehouse_id
),
hourly_queries AS (
  SELECT
    DATE_TRUNC('hour', start_time) AS query_hour,
    COUNT(*) AS query_count,
    ROUND(AVG(total_duration_ms), 0) AS avg_duration_ms,
    ROUND(AVG(waiting_at_capacity_duration_ms), 0) AS avg_queue_wait_ms
  FROM
    system.query.history
  WHERE
    compute.warehouse_id = '<warehouse-id>'
    AND start_time >= NOW() - INTERVAL 7 DAY
  GROUP BY
    DATE_TRUNC('hour', start_time)
)
SELECT
  COALESCE(e.event_hour, q.query_hour) AS hour,
  q.query_count,
  q.avg_duration_ms,
  q.avg_queue_wait_ms,
  e.max_clusters,
  e.event_types
FROM
  hourly_events e
  FULL OUTER JOIN hourly_queries q ON e.event_hour = q.query_hour
ORDER BY
  hour DESC
  • Last updated on