Connect to sources and choose a storage mode
8 min read
Lesson objectives
- Compare Import, DirectQuery, and Direct Lake against latency, freshness, feature, and governance constraints.
- Explain a practical trade-off in connect to sources and choose a storage mode and verify its result.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
Connect to sources and choose a storage mode
Compare Import, DirectQuery, and Direct Lake against latency, freshness, feature, and governance constraints. Import stores a model copy, DirectQuery sends queries to the source, and Direct Lake reads eligible OneLake tables with fallback behavior where applicable.
A sales model needs near-current data but its source cannot sustain interactive query load.
Choose Import with frequent refresh if it meets the SLA; do not choose DirectQuery merely because the scenario says current before measuring source load. This works because the product behaves as follows: Import stores a model copy, DirectQuery sends queries to the source, and Direct Lake reads eligible OneLake tables with fallback behavior where applicable.
Measure refresh duration, data age, and visual latency, then test the required feature in the selected mode.
Worked example and decision boundary
To understand Connect to sources and choose a storage mode, connect the expression to behavior rather than button location: `Import / DirectQuery / Direct Lake` performs the change, while the test proves that meaning remains correct.
Inspect the mechanism first: Import stores a model copy, DirectQuery sends queries to the source, and Direct Lake reads eligible OneLake tables with fallback behavior where applicable. Then perform the stated success check: Measure refresh duration, data age, and visual latency, then test the required feature in the selected mode.
Imagine a sales manager asking for orders placed since the start of today. The data resides in an operational database and fifty employees use the report. The word current is not enough to select a design. Ask for the maximum acceptable delay: one minute, fifteen minutes, or a day. Then ask the source owner how much query load is permitted. Import keeps a compressed model copy and serves users from it; new transactions appear after that copy refreshes. DirectQuery sends requests to the source during interaction, so source speed and concurrency matter. Direct Lake requires appropriate analytics infrastructure and eligible tables; it is not another name for reading any storage file. Begin with data location and environment capability, then choose an available path. A model can combine modes, but relationships, calculations, and refresh become more involved.
Distinguish semantic-model refresh from report-page refresh. Redrawing a page exposes whatever its current mode can access. In an imported model, the data remains the loaded copy until the model refreshes. In DirectQuery, another query can retrieve newer source values, subject to caching and environment configuration. This distinction matters in operating scenarios. Read newest source data and newest imported copy as different requirements. Also separate visual latency from data freshness. A page can open in one second while showing yesterday's figures, or display current figures after a long wait. Design two tests: response time and data age. Collapsing both into the word performance hides the cause and encourages a feature that addresses the wrong problem.
Apply the comparison to a financial table updated nightly, an active order database, and analytical tables already held in the organization's lake. A fast morning import may suit the financial table. For orders, inspect freshness needs and source load before choosing DirectQuery. For lake data, verify table eligibility, capacity, and restrictions against current documentation. Build a measurement page showing latest data time and latest refresh time. Keep one visual unchanged across experiments. Record behavior during a source outage too: can a user read the last saved copy, or does the report return a query failure? That is an important operational consequence. Explain why the chosen design fits and exactly which changed requirement would trigger reassessment. A useful storage decision has explicit freshness, response-time, and source-capacity boundaries.
Lesson summary
The lesson explains how connect to sources and choose a storage mode works through a solved decision and a result check.