slashpeak

Blog

Guides and deep dives on cutting warehouse spend, from the SlashPeak team.

Databricks system table catalog path into ranked cost findings

Diagnosing Databricks costs with system tables

system.query.history and system.compute.warehouses rank full scans, idle warehouses, and shuffle-heavy jobs. SlashPeak never needs SELECT on your tables.

13 min read
Shuffle exchange where every map node sends rows to every reduce node

Databricks Photon shuffle costs and join key layout

Photon speeds up a shuffle join; it does not shrink it. Adaptive execution helps at the edges, but an unclustered join key still burns DBUs every run.

10 min read
Z-order curve versus liquid clustering over Delta files

Z-ordering vs. liquid clustering in Databricks

ZORDER and liquid clustering write min/max stats so Photon can skip files. The wrong columns, or a stale table, means it still reads everything.

10 min read
Databricks SQL warehouse DBU chart with idle auto-stop tail

Databricks SQL warehouses: auto-stop and idle DBUs

A serverless warehouse with a 10-minute auto-stop bills 10 minutes after the last query. Oversizing is a separate leak, and system tables show both.

13 min read
Delta table where every Parquet file is read because the filter misses the file index

Databricks Delta full scans and missed file pruning

A Hive-style partition column only helps once the predicate reaches the file index. WHERE DATE(event_ts) still opens every Parquet file in the table.

16 min read
Snowflake ACCOUNT_USAGE catalog path into ranked cost findings

Diagnosing Snowflake costs with ACCOUNT_USAGE

QUERY_HISTORY, WAREHOUSE_METERING_HISTORY, and clustering views rank prune misses, idle warehouses, and Time Travel without SELECT on a single user table.

11 min read
Snowflake active, Time Travel, and Fail-safe storage stacked

Snowflake Time Travel and Fail-safe storage costs

DATA_RETENTION_TIME_IN_DAYS plus Fail-safe outlives the query it protected. Transient tables skip Fail-safe. TABLE_STORAGE_METRICS splits the bill.

10 min read
Snowflake clustering depth next to Search Optimization Service credits

Snowflake clustering vs. Search Optimization

Automatic Clustering rewrites partitions for range pruning. Search Optimization is a separate path with its own credits. The wrong table pays for both.

10 min read
Snowflake warehouse credit chart with idle auto-suspend tail

Snowflake idle warehouses: auto-suspend and sizing

Auto-suspend at 10 minutes bills for 10 minutes after the last query. WAREHOUSE_METERING_HISTORY shows the idle tail and the oversized warehouse behind it.

12 min read
Snowflake micro-partitions with clustering metadata skipping most of a table

Snowflake clustering keys and micro-partition pruning

A CLUSTER BY key isn't a partition. Snowflake skips micro-partitions by min/max metadata; a wrapped predicate like DATE(order_ts) can't use it.

15 min read
Banner reading Why dbt MERGE scans it all, a bulldozer pushing a whole table to add new rows

Why dbt MERGE jobs scan your whole BigQuery table

A MERGE touching only yesterday can still rewrite the whole target. SlashPeak catches the missing partition filter from the plan, not the YAML.

17 min read
Banner reading Partitioning + clustering, a date partition drawer with one block hit and the rest skipped

How BigQuery clustering complements partitioning

A DATE partition still reads every block in that day without CLUSTER BY on account_id. Clustering makes the partitions you paid for skip blocks too.

14 min read
Banner reading Ad-hoc queries vs your reservation, ad-hoc cars jamming scheduled trucks at a tunnel

Ad-hoc queries and BigQuery reservation contention

INTERACTIVE is a latency SLA for one workload, not a default. Ad-hoc humans hogging a shared reservation queue the finance dashboard behind console SQL.

13 min read
Banner reading CTEs that scan twice, a WITH clause reading the same table two times

BigQuery CTEs that scan the same table twice

BigQuery does not always materialize a CTE. Two references can mean two reads of the same 4 TB table. The SQL looks DRY; the execution plan is the invoice.

14 min read
Banner reading BigQuery slots explained, a slot board with a contention gauge and a query queue

BigQuery slots explained: usage, contention, and cost

A slot is compute, a slot-hour is a billing line, and contention is demand sitting on the ceiling. Here is how BigQuery capacity works.

20 min read
Banner reading Hidden scan costs in Looker and Connected Sheets, a small dashboard on a huge scan iceberg

Hidden scan costs in Looker and Connected Sheets

A dashboard can return 8 columns and 500 rows while BigQuery bills 1.2 TB from a SELECT * view underneath. Bytes follow the scan, not the screenshot.

13 min read
Banner reading Stop full table scans in BigQuery, with a WHERE date funnel lighting one partition

BigQuery full table scans: fixing partition pruning

A partitioned table still bills as a full scan when the filter never reaches the storage layer. Here is how to tell, and how to fix it.

22 min read
Banner reading The real cost of SELECT * in BigQuery, with burning cash and only two needed columns

The real cost of SELECT * in BigQuery

Columnar storage only saves money when a query names its columns. SELECT * can turn a small dashboard into a terabyte-scale job.

17 min read
Banner reading On-demand vs Editions, a scale weighing per-TB pricing against slot-hours

BigQuery on-demand vs. Editions: choosing a pricing model

BigQuery pricing is not fixed. A bursty on-demand project can overpay a slot baseline, and an oversized reservation wastes every idle hour.

15 min read
Banner reading Logical vs physical storage, a bloated drive next to a compressed one

BigQuery storage costs: logical vs. physical billing

Storage is a second bill most teams never open: wrong billing models, tables nobody queries, and layouts with no partitioning at all.

14 min read
Banner reading Finding repeated and redundant queries in BigQuery, a magnifier over tagged query cards

Finding repeated and redundant queries in BigQuery

A query that is fine once becomes a budget line at 600 runs a month. Repeated jobs, re-read CTEs, and wholesale MERGEs are the usual cause.

16 min read

Take control of your warehouse bill

Understand, reduce and govern your cloud data spend with SlashPeak — starting today.

Start for freeGet a demo
  • 30-day free trial
  • No credit card
  • Read-only access
  • Set up in minutes