T3 Groups and entitlements¶
Description¶
Which entitlements does each client see, does the node's policy work through the chain, and can people outside a group still log in? Uses the tokens of T1 and T2, in the same session. Needs the group release set up, and otherwise shows what happens without it.
Runs against: LS Login. The mock has no entitlements per client, so there it only shows that the commands work. It does show a node's 403, under Checking the setup.
From the LS Login team, 2026-11-03: entitlements are computed for the client calling the exchange (A5), and they come from assigning a group to a resource on the service's facility (A9).
Expectations¶
- Entitlements follow the client that a token was issued to, not its audience (A5). So
to-orch, issued to the client, shows the group, andgw-o, issued to the orchestrator, which has no release, doesn't. - For the same reason, the node's token, issued to the gateway, probably shows the gateway's release, not the node's (C4).
- Account A through the chain: 200, and
policy: allowedat the node. - Account B can log in and exchange, the gateway passes the request on, and the node answers 403 with
refused: entitlement, not 401, so that a portal can offer a way to get access.
Steps¶
Release per client¶
tb> userinfo user
tb> userinfo to-gw
tb> exchange to-node --as gateway --from to-gw --audience @node
tb> userinfo to-node
tb> userinfo to-orch
tb> exchange gw-o --as orchestrator --from to-orch --audience @gateway
tb> userinfo gw-o
tb> introspect to-node --as node
tb> introspect to-node --as gateway
tb> exchange to-node-s --as gateway --from to-gw --audience @node --scope "openid entitlements"
tb> userinfo to-node-s
Account A through the chain¶
A 403 here, for a member of the group, means that the release is missing on the gateway. A 502 with refused: userinfo means that user info gave no answer. If the previous step needed the scope, set TESTBED_EXCHANGE_SCOPE and restart the services.
Account B must not be locked out¶
Paste the URL into a private window, so that the browser's session for A isn't used.
tb> login b --as client --no-browser
tb> exchange b-gw --as client --from b --audience @gateway
tb> userinfo b
tb> call gateway --token b-gw
If LS Login refuses the login or the exchange, B is locked out: raise it before going on.
Questions it answers¶
Which group release does a node see, and per which client?¶
Group release as such is confirmed (A9), but what reaches a node isn't known.
- The exchanging client's, as proTES #391 assumes: each gateway's facility needs the group of everything downstream. That goes into #391's registration notes and the operator docs.
- If
introspect --as nodegives the node's own view, and we want that: FOCA needs introspection, as it only has user info today. - Does the exchange need
scopefor this? Then FOCA's exchange gets a scope setting per target.
How can a group be released without locking out users outside it?¶
From the LS Login team, 2026-11-03: create a second resource for the Lifescience VO and assign its members group (A10). What's still open:
- Is the second resource needed on every client in the chain, or only on those that log in users? In other words: can a gateway exchange for a user who isn't in any group assigned to its facility?
- Only those that log in users: registering a gateway stays simple.
- Every client: each one needs that resource, which goes into the registration notes.
- Are users outside a group ever stopped at LS Login in the first place, and by what setting?
- No: node policy decides, with a 403, and Krini shows "request compute access".
- Yes: find out which setting caused it before registering Krini.
- Does the members group add an entitlement that nodes see for all users? Node policies must not match it by accident.
Which role is needed to assign groups to a resource?¶
Answered. This project's maintainer could open a facility and create a resource, but in "Assign groups" the Next button stays greyed out and no groups can be added (seen by 2026-11-03).
The LS AAI team confirmed that TrustedFacilityAdmin within the relevant VO enables this assignment. The VO admins have chosen not to grant that role for this setup; maintainers create resources and ask the VO admins to assign groups, as described in Client registration.
What it changes: operators should plan for VO-admin involvement when setting up group release, rather than trying to obtain self-service assignment permissions.
Group entitlements or resource capabilities?¶
The operator's Entitlements page recommends resource capabilities, as group entitlements change when a group is renamed or moved (see What public documents say). Are they available on LS Login, for a client like these?
- Yes: node policies can match a capability rather than a group's name.
TESTBED_REQUIRED_ENTITLEMENTtakes either, so the testbed can try it. - No: nodes match group entitlements and break if groups are renamed, which the operator docs should say.
What is the exact entitlement value?¶
We derived the example from a known entitlement of the same form (A12). If it differs, nodes get another value in their settings, and nothing else changes.
What to record¶
| Step | Record | Result |
|---|---|---|
| Release per client | The entitlements for each line. Does introspect --as node differ? Does the exchange need the scope? |
|
| Exact value | The entitlement in user info, compared with TESTBED_REQUIRED_ENTITLEMENT |
|
| Account A | Status, policy at the node |
|
| Account B | Login, exchange, status and refused at the node |
|
| Group release | Which role it took, and who granted it | TrustedFacilityAdmin within the relevant VO; not granted for this setup, so the VO admins assign groups on request (Client registration) |