Tableau
Tableau end to end: connecting data, building worksheets from dimensions and measures, calculations, dashboards, and publishing to Server. Interviewers use it to check you think in terms of the underlying data model, not just chart choices.
on this pageshowhide
guide
overview
~1 minTableau is a visual analytics tool: you drag fields onto shelves, Tableau writes the query, and the result comes back as marks on a chart. Interviews use it to test whether you understand the query being written on your behalf. A candidate who can only talk about chart choices stalls quickly; one who can say what grain a view is at, which rows a filter removed and when a calculation ran can reason through almost any surprise the tool produces. The hub follows a workbook from source to audience. [Data connections](/topics/data-tableau-data-connections) decide where the data is read from and how tables are combined. [Worksheets and marks](/topics/data-tableau-worksheets-marks) is the building block: fields, shelves and the level of detail of a view. [Calculated fields and LOD](/topics/data-tableau-calculated-fields-lod) and [filters and parameters](/topics/data-tableau-filters-parameters) hold most of the senior questions, because both turn on the order in which Tableau evaluates things. [Dashboards and actions](/topics/data-tableau-dashboards-actions) assemble sheets into one interactive surface, and [publishing and Tableau Server](/topics/data-tableau-publishing-server) covers permissions, refreshes and embedding once other people depend on the work. Junior rounds check the vocabulary: dimension against measure, live against extract, what a parameter is for. Senior and principal rounds become debugging and governance conversations: why a number changed, why a dashboard is slow, who may see which rows. Learn worksheets and marks first, then data connections, then the order of operations that calculations and filters share.
primer
### The view sets the grain Every worksheet has a level of detail: the set of dimensions it contains, wherever they sit, on Rows, Columns or the Marks card. Tableau groups the data to that grain and aggregates measures inside each group. Most surprising numbers, such as a total that does not add up, are grain questions in disguise. ### Two independent switches on every field Dimension or measure decides whether a field slices the view or is aggregated within it. Discrete or continuous decides whether it produces headers or an axis. They are separate choices, so a date or a number can play either part. ### Calculations run at different moments A row-level calculation works on source rows; an aggregate calculation works per mark; a table calculation works on the result set already returned to Tableau; an LOD expression sets its own grain, either fixed or relative to the view. Knowing *when* each one runs explains what it can see, which filters reach it, and why two formulas that look equivalent can disagree. ### Order of operations is the debugging tool Filters are not interchangeable. Extract, data source, context, dimension and measure filters are evaluated in a fixed sequence, with LOD expressions and table calculations slotted in between. Most "the filter is ignored" or "the top N looks wrong" questions are settled by placing each piece on that line. ### Where the data lives decides freshness and speed A live connection queries the source on every render; an extract is a refreshed snapshot held by Tableau. Tables can be joined, related in the logical layer, or blended per sheet, and each option preserves or loses the grain of its inputs differently. The choice trades freshness against query cost and aggregate correctness.
- Dimension
- A field that slices the view into groups and so contributes to its level of detail. Usually categorical, though any field can be given the role.
- Measure
- A field aggregated within each group the dimensions define. Its default aggregation is set per field and shown on the pill.
- Discrete and continuous
- A field's second role: discrete fields split the view into labelled headers, continuous fields produce an axis with a range. Pills show them blue and green.
- Mark
- One visual item in a view, such as a bar, point or line segment, standing for one group of rows at the view's level of detail.
- Level of detail
- The grain of a view, set by the dimensions it contains. An LOD expression computes at a grain it names instead of the view's own.
- Table calculation
- A calculation run inside Tableau over the aggregated result already in the view, such as a running total or rank, scoped by addressing and partitioning.
- Context filter
- A dimension filter promoted to run earlier, so later filters, top-N conditions and FIXED expressions see only the rows it keeps.
- Parameter
- A workbook variable holding one user-changeable value. It affects a view only through the calculations, filters or reference lines that use it.
- Extract
- A snapshot of source data stored in Tableau's Hyper format and refreshed fully or incrementally on a schedule, instead of querying the source live.
- Relationship
- A link between logical tables that keeps each at its native grain; Tableau works out the joins per query rather than flattening the tables up front.
- Data blending
- Combining a primary and a secondary data source per worksheet on linking fields, with the secondary aggregated before the two meet.
- Site role
- The licence-level ceiling on what a user can do on a site, such as Viewer, Explorer or Creator. Content permissions cannot grant beyond it.
Trace one dashboard click back to the database. Someone clicks a mark, and an action changes the filters on its target sheets. For each target, Tableau builds a query from the shelves: dimensions become the grouping, measures the aggregates, and filters the conditions, each at its own stage. Data source and context filters shape the rows first, FIXED expressions are computed against what is left, and aggregate filters trim the grouped result. The query goes to the source database on a live connection or to the Hyper engine on an extract; with relationships, Tableau may send several queries at different grains and combine them. What returns is one row per mark, and only then do table calculations and totals run, before anything is drawn. Each section owns one stage of that path: - **Data connections** decide what the query is sent to and how tables combine before it runs. - **Worksheets and marks** decide the grain of the query and how each returned row becomes a mark. - **Calculated fields and LOD** and **filters and parameters** decide what happens to rows and aggregates along the way, so they share one order of operations. - **Dashboards and actions** turn a selection on one sheet into new queries on others, so every action adds work to each click. - **Publishing and Server** wrap the whole path in permissions, schedules and stored credentials. Problems are traced the same way: a slow click usually means many queries, a wrong number a grain or filter-order mistake, stale data a refresh that did less than expected.
- Worksheets and Marks →
Dimensions, measures, discrete and continuous fields, and the grain of a view: the model every other section assumes.
- Data Connections →
Live against extract and joins against relationships against blends, which decide freshness, speed and whether aggregates double-count.
- Calculated Fields and LOD →
Row-level, aggregate, table and LOD calculations, and the moment each one is evaluated.
- Filters and Parameters →
The full order of operations, context filters, sets, and why a parameter changes nothing until something references it.
- Dashboards and Actions →
Combining sheets into guided analysis, and the action settings that decide both behaviour and speed.
- Publishing and Tableau Server →
Permissions, refresh credentials, subscriptions and embedding: what changes once other people rely on the workbook.
Reading blue and green pills as dimension and measure: the colour shows discrete or continuous, which is a separate choice from the field's role.
Expecting a dimension filter to narrow a FIXED expression; it is applied too late, so promote it to a context filter or the LOD keeps computing over everything.
Calling a grand total wrong because it differs from the visible rows; totals re-aggregate underlying data, so distinct counts, averages and ratios are not row sums.
Assuming a parameter filters anything on its own; it only changes what references it, such as a calculated field, a top-N condition or a reference line.
Joining tables of different grain and then summing the coarser table's measure; the join repeats its values, the problem relationships were introduced to avoid.
Relying on a data source filter for row-level security; anyone allowed to edit or download the content can remove it - see the data source filter question.
Embedding one person's database credentials in a scheduled refresh; when that password rotates or the person leaves, every dashboard on the source stops refreshing.
Setting every sheet to filter every other; each click then re-queries the whole dashboard, and the slowness gets blamed on the database.
This guide assumes a current Tableau release on the year-numbered versioning, on Desktop, Server or Cloud. A few earlier changes still shape interview answers, because many Server installations and tutorials predate them: - **10.5** replaced the older extract format with the Hyper engine, so extract questions today mean `.hyper` files. - **2018.3** added set actions and **2019.2** parameter actions; before them a filter action was the main way for one sheet to drive another. - **2020.2** introduced relationships and the logical layer. Earlier, combining tables meant a join or a blend, and much advice about duplicated measures dates from then. - **2022** renamed Tableau Online to Tableau Cloud. When an answer depends on one of these, such as how tables are combined, how a selection drives a calculation, or how embedding authenticates, say which era you are describing.
Tableau, now part of Salesforce, is most often weighed against Microsoft Power BI, Looker and Qlik. The comparison an interviewer expects is about fit rather than features: Tableau is strongest at visual, exploratory analysis by analysts; Power BI sits closest to the Microsoft stack; Looker centres a semantic model defined in code. None wins in general, and a defensible answer names the team, the data platform and who will author content. Tableau Prep shapes data before analysis, and the data usually comes from a cloud warehouse, which raises a recurring design question: should business logic live in the warehouse, a published data source or a workbook calculation? The further upstream a definition sits, the more workbooks share it.
explore
- Q-07549In Tableau, what is the difference between a dimension and a measure?level: juniorimportance: must knowfreq 85%
- Data Connections6 questions
- Worksheets and Marks6 questions
- Calculated Fields and LOD6 questions
- Filters and Parameters6 questions
- Dashboards and Actions6 questions
- Publishing and Tableau Server6 questions
questions
page 2 of 2In Tableau, what does the Show Me panel do and why do analysts bypass it?
basics
~20 sShow Me recommends Tableau chart types for the fields you have selected, greys out charts whose field requirements are unmet, and rebuilds the shelves when you pick one. It is a starting point that guesses intent from field types alone.
In Tableau, where does a cross-database join between a warehouse table and a CSV execute?
basics
~20 sIn Tableau's own engine. Tableau's federated query layer sends a separate query to each connection, brings the results back, and performs the join locally — on Desktop, or on the Tableau Server or Cloud node running the query.
In Tableau, what is the difference between a fixed set and a computed set?
basics
~20 sA fixed set stores a hand-picked list of members in the workbook and never changes as data arrives. A computed set stores a condition or a top-N rule and re-evaluates it each time the workbook queries, so its membership tracks the data.
In Tableau, how does TOTAL(COUNTD([Customer])) differ from WINDOW_SUM(COUNTD([Customer]))?
basics
~20 sWINDOW_SUM adds up the distinct counts already computed for each mark, so a customer active in several marks is counted several times. TOTAL re-aggregates the partition's underlying rows with the same aggregation, de-duplicating customers across it.
In Tableau, what can set actions and parameter actions do that a filter action cannot?
basics
~20 sThey write the clicked value into an object calculations can read: a parameter holds one value, a set holds membership. A filter action only removes rows, so it cannot drive a calculation, keep the unselected rows visible, or compare selection against the whole.
How do you embed a Tableau Cloud view in an app without making users sign in separately?
basics
~20 sYour application authenticates the user itself, then mints a short-lived signed token that asserts that user's Tableau identity to the site — a connected-app JWT on current deployments, trusted tickets on older Tableau Server. The embedded view then loads as that Tableau user.
showing 31–36 of 36