data_engineer role, or a role that inherits it, to install or remove. Any member from
viewer up can browse listings and see what is installed.
Install
1
Open the listing
Go to Marketplace, find the listing on the Browse tab, and select it. Read Includes and
the access line: every listing today says it Requests no access to your data.
2
Select Install
The install dialog opens on its first step, Connect.A semantic-model listing asks for two values, which the drafts are bound to:
Both are identifiers only: the name of a catalog and the name of a schema, not a connection or a
credential. If your columns are named differently from the listing’s, change the field
expressions on the drafts afterwards.An ontology listing has no values to enter, so Connect has nothing to fill in.
3
Review and confirm
The Review step restates what will happen: “Install copies v1.0.0 into this workspace as
drafts you own. Updates never apply on their own.” Confirm to start the copy.
4
Open the drafts
The install appears on the Installed tab, briefly as Installing, then as Installed.
Select Open on the record to go to the Semantic layer tab the drafts are in: Models for a
semantic-model listing, Ontologies for an ontology. Each draft is in the state Draft.
- A semantic-model listing creates one draft model per dataset it includes, with the listing’s
fields, relationships and metrics, bound to
<catalog>.<schema>.<table>. Sales orders and customers creates two models,ordersandcustomers, carrying the metricstotal_revenue,order_countandcustomer_count. - An ontology listing creates one draft concept per concept it includes, in the ontology the
listing names (
commercefor Commerce vocabulary). The ontology is created if your tenant does not have one by that name. No concept is bound to a model: which model realizes “Customer” in your tenant is a claim only you can make, on the draft.
Add a pipeline
A pipeline listing is an image, not a definition you copy. Add to workspace puts it in your tenant’s pipeline registry as you and stops there: the pipeline does not run, read data or write anything. The listing is a black box until you run it; there are no bindings to fill in, and its configuration is yours to set afterwards.1
Open the listing
Go to Marketplace, filter the Type facet to Pipeline, and select the listing. The page
shows the image the listing adds, pinned by digest, and a sentence on what is in it.
2
Select Add to workspace
The dialog has one step. Confirm, and the pipeline appears in your registry and as a row on the
Installed tab. Adding needs
create on both Marketplace installs and Pipelines; the
built-in data_engineer role holds both.3
Open it in Pipelines
The listing page links to the pipeline’s own page. Set its configuration and start a run from there;
who may run it is decided by the Pipelines page, not by the Marketplace.
pipelines entitlement. In a tenant without it the page says
“Adding it needs Pipelines, which this workspace does not include.” and the button is disabled, because the registry
would refuse the add. A pipeline you already hold under the same id, or a tenant with more than one
query engine where the pipeline has to be pinned to one, is refused by the pipelines service: the
record is kept as Failed with that service’s sentence. Choosing Install through another client
on a pipeline listing is refused with a message telling you to use Add to workspace.
Removing the record from Installed forgets the record only; the pipeline stays in your registry.
Add an agent
An agent listing is added to your agent registry, like a pipeline, and is handed to Agents for the step that needs consent.1
Select Add to workspace
The dialog restates what happens: the agent goes into your registry, and it does not run, read
data or open the network. Confirm to add it. You need Create on Marketplace installs and on
Agent registry (
data_engineer holds both).2
Hand off to Agents
The record on Installed shows the agent, and the listing page shows Open in Agents. The
install itself, where the access the agent asks for is reviewed and consented to, is on the
agent’s page and needs
data_admin. A data_engineer can add an agent that only an admin can
install.agents, the add ends as
Failed with the agents service’s own reason. Remove forgets the marketplace record only: the
agent stays in your registry. A newer catalog version does not update an agent you already added.
Add an app
An app listing is added to your app registry the same way an agent is, and is handed to Apps for the step that needs consent.1
Select Add to workspace
The dialog restates what happens: the app goes into your registry with no pull credential, and it
does not run, read data or open the network. You need Create on Marketplace installs and on
Data apps (
data_engineer holds both).2
Hand off to Apps
The record on Installed shows the app, and the listing page shows Open in Apps. The install
itself, where the access the app asks for is reviewed and consented to, is on the app’s page and
needs
data_admin.marketplace. If your tenant already holds an app with that
name the add ends as Failed with the apps service’s own reason. Remove forgets the marketplace
record only: the app stays in your registry.
The Installed tab
Every install your tenant has made is a row on Marketplace → Installed:Versions
A listing installs once per version in a tenant. Selecting Install on a version you already hold is refused with a message of the form “Sales orders and customers v1.0.0 is already installed in this workspace; the drafts it created are in your semantic layer”. When the catalog carries a newer version than the one you hold, the listing shows v1.1.0 available (with the new version number). Installing it creates a new set of drafts; the drafts from the earlier install are left exactly as they are. Updates never apply on their own, so a definition you have edited or certified is never overwritten by the Marketplace.Remove a record
Select Remove on an Installed row. The dialog says what it does: “Your drafts stay; this only removes the marketplace record.” Removing forgets that the install happened, and lets you install that version again; it never deletes a model or concept. To get rid of the drafts themselves, delete them on the Semantic layer page as you would any draft.If an install fails
A Failed row carries the reason, in the semantic layer’s own words, and the install itself ends with “the semantic layer refused the copy:” followed by that reason. The two you are likely to meet:
Remove the failed record before you try again. While it is there, Install on that version is
refused with “Sales orders and customers v1.0.0 has a failed install record; remove it from
Installed, then install again” (naming the listing you chose).
Who can do what
The full row is Marketplace installs in the roles matrix. Installing
also creates drafts on your behalf, so it needs three permissions together: Create on Marketplace
installs, Create on Semantic models and Create on Concepts.
data_engineer holds all
three. A custom role missing any one of them cannot install, whichever listing you pick. Certifying
what you installed is a separate privilege, held by data_admin, as for any draft.
Related
Marketplace overview
What a listing shows, the publisher tiers, and the add-on entitlement.
Models and metrics
Edit, propose and certify the semantic-model drafts an install created.
Ontology concepts
Bind the concept drafts to the models that realize them.
Roles and permissions matrix
The Marketplace installs row.