Auditing a Tableau Cloud site: what you can measure without a repository

Every guide in this series so far leans on the same instrument: the repository, queried read-only, with thirty days of history and a statement timeout. On Tableau Cloud that instrument does not exist. There is no workgroup database to reach — not hidden, not license-gated; it is not part of the product. An administrator arriving from Server, saved queries in hand, has to start over.

This guide is the honest inventory of what replaces it: what Admin Insights actually gives you, what the Metadata API and the Activity Log add, what is genuinely out of reach compared with Server — and how to build a picture worth acting on anyway. The honesty is the point. A Cloud audit that pretends to be a Server audit produces confident numbers with quiet gaps; knowing exactly where the gaps are is what makes the rest trustworthy.

What replaced the repository

The supported window into a Cloud site is the Admin Insights project: curated data sources, maintained by Tableau, that you connect to like any published data source and explore in Desktop or a workbook. Nine of them, per the documentation. The ones an audit leans on hardest:

TS Events — in the documentation's words, it "functions as a primary audit data source," recording sign-ins, publishing, and view access. This is the Cloud counterpart of the historical_events star: who did what, when.

Job Performance — "events and runtime information for background jobs on the site, such as extract refreshes and flow runs." Queue and runtime analysis lives here: the wait-versus-runtime split, the pile-up clusters, the repeat offenders.

Site Content — the inventory: projects, workbooks, flows, views, and the governance fields around them. Paired with TS Events, it answers the stale-content question.

Viz Load Times — per-view load performance, the curated stand-in for what http_requests answers on Server.

The remaining five — TS Users, Groups, Permissions, Subscriptions, Tokens — cover licensing, membership, effective permissions, subscription delivery, and open authentication tokens respectively.

The constraint to internalize before building anything: retention is fixed and short. Admin Insights captures up to 90 days of data by default — up to 365 with an Advanced Management license, per the documentation. There is no MIN(created_at) to check and no retention setting to extend; the window is the window. On Server you discover your history; on Cloud you are issued it.

Re-running the series on Cloud

The analyses from the earlier guides translate more directly than the missing database suggests. The wait-versus-runtime split from diagnosing backgrounder delay comes straight out of Job Performance's timing fields; so does the repeat-offender failure analysis from why extract refreshes fail — the five failure modes are identical, only the query surface changes. What does not translate is the fix ladder's last rung: on Cloud you cannot add backgrounders, because Tableau manages the pool. Demand-side fixes — de-anchoring schedules, trimming fat refreshes, retiring unread content — are the whole toolkit, which is less of a loss than it sounds: on Server, they were most of the fix anyway.

One Cloud-specific blind spot from that same analysis: refreshes against private-network sources run through Tableau Bridge, on machines you provision. Bridge capacity problems look like platform slowness while living entirely on your side of the fence — measure the Bridge pool separately before blaming the site.

What the Metadata API adds

The Metadata API is a GraphQL interface over your site's content and its relationships: which workbooks sit on which data sources, which fields they use, what sits upstream and downstream. For audit purposes it answers the dependency questions the inventory alone cannot: before retiring a data source, who actually consumes it; before trusting a dashboard, where its numbers come from. It is content lineage, not operational history — no job timings, no access events — which makes it the complement to Admin Insights rather than an alternative.

What the Activity Log adds

The Activity Log is the detailed event stream — site events like permission changes and content interactions, tenant events like role assignments — designed, per the documentation, to enhance Admin Insights rather than replace it, and shaped for shipping into external tools. The documentation is explicit that license edition determines the access path, update frequency, and retention: on a default license, events come through the Tableau Cloud Manager REST API, updated daily, with 14 days of retention; Advanced Management, Enterprise, and Tableau+ editions add Amazon S3 delivery, updates within minutes, and 365-day retention. Check what your own license includes before designing an audit process around it — the difference between 14 days pulled daily and a year streamed to S3 is the difference between a spot check and a monitoring pipeline.

What stays out of reach

Compared with a Server repository session, four things are genuinely gone, not merely relocated.

Arbitrary SQL. Admin Insights is curated data sources, not tables. Questions the curation anticipated are easy; questions it did not — a join nobody shipped, a field nobody exposed — have no workaround. On Server, the odd question costs a query; on Cloud, it may simply have no answer.

History you control. Every retention window above is Tableau's decision. A Server admin who needs two years of job history configures it; a Cloud admin works with 90 or 365 days, full stop.

Request-level traffic. Viz Load Times covers view performance; the raw request log underneath — every URL, every session, every latency — has no Cloud equivalent at that granularity.

The platform's own knobs. Process counts, pool sizes, node topology: on Cloud these are Tableau's to manage, which removes both the burden and the lever.

A meaningful picture anyway

The practice that works, and this is ours rather than anything documented: treat the fixed windows as a reason to snapshot, not a reason to shrug. Connect to the Admin Insights sources in Desktop, build the handful of views this series has been building all along — wait percentiles by hour, failure streaks by workbook, last-access by content age — and export the aggregates on a monthly cadence. Twelve monthly snapshots reconstruct exactly the long-baseline view the retention window refuses to give you, at the cost of one calendar reminder. The audit discipline is the same one the repository guide teaches for Server — aggregates out, sensitive rows never — applied to a politer data source. And where the inventory needs an outside eye across all of it at once, a fixed-scope audit of your environment covers Cloud sites with the same rule library, scoped to what Cloud actually exposes.

Honest limits, measured anyway

A Cloud site cannot be audited the way a Server can. It can be audited well: jobs, failures, load times, content, access, lineage — each from a named, supported source, each with its window stated. The discipline this series keeps returning to matters more here, not less: say what was measured, say over what period, and say what was not measurable at all. On Cloud, that last sentence is doing real work.