With Hibernate ORM 6, pagination started to be unreliable: When
setting the max results only if the first row was 0 has randomly
affected other threads where first row was greater than 0. The
latter thread sometimes produced query which did *not* account
for the offset (cf. threads `-t1` and `-t2` below, while `-t2`
missed the `offset ? rows` part whic `-t3` has).
This has been fixed by setting the first row offset unconditionally.
Closes: #20202Closes: #16570
```
2023-06-02 10:19:03.855000 TRACE [org.keycloak.models.sessions.infinispan.initializer.SessionInitializerWorker] (blocking-thread-node-2-p8-t1) Running computation for segment 0 with worker 0
2023-06-02 10:19:03.856000 TRACE [org.keycloak.models.sessions.infinispan.initializer.OfflinePersistentUserSessionLoader] (blocking-thread-node-2-p8-t1) Loading sessions for segment=0 lastSessionId=00000000-0000-0000-0000-000000000000 first=0
2023-06-02 10:19:03.856000 DEBUG [org.keycloak.models.jpa.PaginationUtils] (blocking-thread-node-2-p8-t1) Set max to 64 in org.hibernate.query.sqm.internal.QuerySqmImpl@2fb60f8b
2023-06-02 10:19:03.856000 DEBUG [org.keycloak.models.jpa.PaginationUtils] (blocking-thread-node-2-p8-t1) After pagination: 0, 64
2023-06-02 10:19:03.857000 TRACE [org.keycloak.models.sessions.infinispan.initializer.SessionInitializerWorker] (blocking-thread-node-2-p8-t2) Running computation for segment 1 with worker 1
2023-06-02 10:19:03.857000 TRACE [org.keycloak.models.sessions.infinispan.initializer.OfflinePersistentUserSessionLoader] (blocking-thread-node-2-p8-t2) Loading sessions for segment=1 lastSessionId=00000000-0000-0000-0000-000000000000 first=64
2023-06-02 10:19:03.857000 TRACE [org.keycloak.models.sessions.infinispan.initializer.SessionInitializerWorker] (blocking-thread-node-2-p8-t3) Running computation for segment 2 with worker 2
2023-06-02 10:19:03.857000 DEBUG [org.keycloak.models.jpa.PaginationUtils] (blocking-thread-node-2-p8-t2) Set first to 64 in org.hibernate.query.sqm.internal.QuerySqmImpl@71464e9f
2023-06-02 10:19:03.857000 DEBUG [org.keycloak.models.jpa.PaginationUtils] (blocking-thread-node-2-p8-t2) Set max to 64 in org.hibernate.query.sqm.internal.QuerySqmImpl@71464e9f
2023-06-02 10:19:03.857000 DEBUG [org.keycloak.models.jpa.PaginationUtils] (blocking-thread-node-2-p8-t2) After pagination: 64, 64
2023-06-02 10:19:03.857000 TRACE [org.keycloak.models.sessions.infinispan.initializer.OfflinePersistentUserSessionLoader] (blocking-thread-node-2-p8-t3) Loading sessions for segment=2 lastSessionId=00000000-0000-0000-0000-000000000000 first=128
10:19:03,859 DEBUG [org.hibernate.SQL] (blocking-thread-node-2-p8-t1)
select
p1_0.OFFLINE_FLAG,
p1_0.USER_SESSION_ID,
p1_0.CREATED_ON,
p1_0.DATA,
p1_0.LAST_SESSION_REFRESH,
p1_0.REALM_ID,
p1_0.USER_ID
from
OFFLINE_USER_SESSION p1_0,
REALM r1_0
where
r1_0.ID=p1_0.REALM_ID
and p1_0.OFFLINE_FLAG=?
and p1_0.USER_SESSION_ID>?
order by
p1_0.USER_SESSION_ID fetch first ? rows only
10:19:03,859 DEBUG [org.hibernate.SQL] (blocking-thread-node-2-p8-t2)
select
p1_0.OFFLINE_FLAG,
p1_0.USER_SESSION_ID,
p1_0.CREATED_ON,
p1_0.DATA,
p1_0.LAST_SESSION_REFRESH,
p1_0.REALM_ID,
p1_0.USER_ID
from
OFFLINE_USER_SESSION p1_0,
REALM r1_0
where
r1_0.ID=p1_0.REALM_ID
and p1_0.OFFLINE_FLAG=?
and p1_0.USER_SESSION_ID>?
order by
p1_0.USER_SESSION_ID fetch first ? rows only
2023-06-02 10:19:03.860000 TRACE [org.hibernate.orm.jdbc.bind] (blocking-thread-node-2-p8-t1) binding parameter [1] as [VARCHAR] - [1]
2023-06-02 10:19:03.860000 TRACE [org.hibernate.orm.jdbc.bind] (blocking-thread-node-2-p8-t1) binding parameter [2] as [VARCHAR] - [00000000-0000-0000-0000-000000000000]
2023-06-02 10:19:03.860000 TRACE [org.hibernate.orm.jdbc.bind] (blocking-thread-node-2-p8-t1) binding parameter [3] as [INTEGER] - [64]
10:19:03,860 DEBUG [org.hibernate.SQL] (blocking-thread-node-2-p8-t3)
select
p1_0.OFFLINE_FLAG,
p1_0.USER_SESSION_ID,
p1_0.CREATED_ON,
p1_0.DATA,
p1_0.LAST_SESSION_REFRESH,
p1_0.REALM_ID,
p1_0.USER_ID
from
OFFLINE_USER_SESSION p1_0,
REALM r1_0
where
r1_0.ID=p1_0.REALM_ID
and p1_0.OFFLINE_FLAG=?
and p1_0.USER_SESSION_ID>?
order by
p1_0.USER_SESSION_ID offset ? rows fetch first ? rows only
2023-06-02 10:19:03.861000 TRACE [org.hibernate.orm.jdbc.bind] (blocking-thread-node-2-p8-t3) binding parameter [3] as [INTEGER] - [128]
2023-06-02 10:19:03.861000 TRACE [org.hibernate.orm.jdbc.bind] (blocking-thread-node-2-p8-t3) binding parameter [4] as [INTEGER] - [64]
```
Co-authored-by: mkanis <mkanis@redhat.com>
* Use a valid SemVer format for the SNAPSHOT version
* Update pom.xml
* Update pom.xml
---------
Co-authored-by: Stian Thorgersen <stianst@gmail.com>
Co-authored-by: Stian Thorgersen <stian@redhat.com>
Instead create a new instance of LegacyUserCredentialManager to ensure all calls are routed via the CacheAdapter and its SubjectCredentialManagerCacheAdapter.
Closes#14309
Provide startup time in UserSessionProvider independent of Infinispan,
cleanup code that is not necessary for the map storage as it isn't using Clustering.
Move classes to the legacy module.
Closes#12972
Split PublicKeyStorageProvider
- Extract clearCache() method to separate interface and move it to the legacy module
- Make PublicKeyProvider factories environment dependent
- Simple map storage for public keys that just delegates
Resolves#12763
Co-authored-by: Martin Kanis <mkanis@redhat.com>
- class SingleUserCredentialManager to SingleEntityCredentialManager
- method UserModel.getUserCredentialManager() to credentialManager()
Renaming of API without "get" prefix to make it consistent with other APIs like for example with KeycloakSession
- UserStorageManager now handles authentication for old Kerberos+LDAP style
- new getUserByCredential method in MapUserProvider would eventually do the same.
- Introduction of new AdminRealmResource SPI
- Moving handler of /realm/{realm}/user-storage into model/legacy-service
- session.users() and userStorageManager() moved refers legacy module
IMPORTANT: Broken as UserStorageSyncManager is not yet moved
- Introduces Datastore SPI for isolating data store methods
- Introduces implementation of the datastore for legacy storage
- Updates DefaultKeycloakSession to leverage Datastore SPI instead
of direct creating of area providers by the session
* Allow exposing some initial provider config options via web site
Co-authored-by: Stian Thorgersen <stian@redhat.com>
Closes#10571
* Include type to provider options, and hide build-icon column as it's not relevant
Co-authored-by: stianst <stianst@gmail.com>
Closes: #10447
* Prepare logical layer to distinguish between ResourceServer id and client.id
* Reorder Authz methods: For entities outside of Authz we use RealmModel as first parameter for each method, to be consistent with this we move ResourceServer to the first place for each method in authz
* Prepare Logical (Models/Adapters) layer for returning other models instead of ids
* Replace resourceServerId with resourceServer model in PermissionTicketStore
* Replace resourceServerId with resourceServer model in PolicyStore
* Replace resourceServerId with resourceServer model in ScopeStore
* Replace resourceServerId with resourceServer model in ResourceStore
* Fix PermissionTicketStore bug
* Fix NPEs in caching layer
* Replace primitive int with Integer for pagination parameters
- fix RealmAdapter to create a new map instead of trying to change unmodifiable map
- only provide POST endpoints for creating or updating the texts (to have the endpoints consistent with other Admin API endpoints)
- add tests
- extend RoleMapperModel with method hasDirectRole(RoleModel), which only checks for direct assignment in contrast to the existing method hasRole(RoleModel)
- extend ScopeContainerModel with method hasDirectScope(RoleModel), which only checks for direct scope mapping in contrast to the existing method hasScope(RoleModel)
- use the new hasDirectRole and hasDirectScope methods to check whether a role is in the "available" list and whether it can be assigned (previously, the hasRole method was used for this purpose)
- add hint to UI that available roles contain effectively assigned roles which are not directly assigned
- adjust and extend tests
