Skip to main content

Pillar 1 — Visibility

Principle: You cannot reduce what you cannot see. Before any optimization, establish a per-query, per-warehouse, per-workload baseline of credit spend.

Visibility is the foundation pillar of the Snowflake Cost Optimization Framework. It is where most teams are strongest — and still incomplete, because account-level dashboards answer how much but not which query or whose.

Why it matters​

Snowflake bills credits for warehouse time, not query work. A single poorly-written query that resumes an XL warehouse for its 60-second minimum can cost more than thousands of small queries. Without per-query visibility, that query is invisible inside a monthly warehouse total — so it never gets fixed.

What to measure​

SignalSourceTells you
Credits per warehouse over timeWAREHOUSE_METERING_HISTORYwhere the money goes
Credits attributable per queryQUERY_HISTORY + execution timewhich queries are expensive
Idle vs active warehouse timeWAREHOUSE_EVENTS_HISTORYsuspend opportunity
Repeated identical queriesquery fingerprintcache opportunity
Spend trend per workloadtagged QUERY_HISTORYwhere growth is coming from

Copy-paste recipe: most expensive queries last 30 days​

-- Approximate credit cost per query family, last 30 days.
-- Attributes warehouse credits to queries by share of execution time.
with q as (
  select
    query_id,
    query_text,
    warehouse_name,
    warehouse_size,
    total_elapsed_time / 1000.0 as elapsed_s
  from snowflake.account_usage.query_history
  where start_time > dateadd('day', -30, current_timestamp())
    and execution_status = 'SUCCESS'
)
select
  warehouse_name,
  warehouse_size,
  any_value(left(query_text, 120)) as sample_sql,
  count(*)                          as runs,
  round(sum(elapsed_s), 1)          as total_seconds,
  round(sum(elapsed_s) / 3600, 2)   as warehouse_hours
from q
group by warehouse_name, warehouse_size, hash(regexp_replace(upper(query_text), '\\s+', ' '))
order by total_seconds desc
limit 25;

This surfaces the queries worth attention. chukei takes the next step automatically: it computes a hard and soft fingerprint for every query at the proxy, so repeated and semantically-equivalent queries are grouped without you normalizing SQL by hand.

How chukei enforces it​

The attribute plugin records every query's fingerprint, warehouse, client app, and the cache/rewrite/suspend decision applied — exported as Prometheus metrics and OpenTelemetry spans. You get per-query visibility live, not after the monthly close.

Visibility checklist​

  • A baseline of credits per warehouse for the last 90 days exists.
  • You can list the top 25 most expensive query families on demand.
  • Idle warehouse time is measured, not assumed.
  • Repeated/duplicate queries are identified by fingerprint.
  • Cost signals are refreshed continuously, not only at month end.

Get the baseline in minutes. The replay simulator ingests your QUERY_HISTORY and reports cacheable volume and idle time with no traffic change.