[
https://issues.apache.org/jira/browse/UNOMI-1000?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18124446#comment-18124446
]
Serge Huber commented on UNOMI-1000:
------------------------------------
Pull request: https://github.com/apache/unomi/pull/911
Profile purge, segment date recalculation, rule statistics refresh and router
startup now run under each tenant via {{executeAsTenant}} (context restored
afterward). The system persistence filter is unchanged.
Unit tests: {{BackgroundTenantTasksTest}}, {{RouterCamelContextTenantLoadTest}}.
> Background tasks only process the internal "system" tenant
> ----------------------------------------------------------
>
> Key: UNOMI-1000
> URL: https://issues.apache.org/jira/browse/UNOMI-1000
> Project: Apache Unomi
> Issue Type: Bug
> Affects Versions: unomi-4.0.0
> Reporter: Serge Huber
> Priority: Blocker
> Fix For: unomi-4.0.0
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> h3. Summary
> Unomi runs several tasks in the background. In 4.0, some of them only see the
> data of the internal "system" tenant. They skip every real tenant, including
> "default", where migrated data goes:
> * Purge of old profiles, sessions and events. The purge settings have no
> effect, so data keeps growing.
> * Daily update of segments and scorings that use relative dates, such as
> "visited in the last 7 days". Profiles never leave these segments.
> * Recurring imports and exports. After a restart they are not scheduled again
> until someone saves their configuration.
> * Refresh of rule statistics.
> Other tasks, such as the rule refresh and tenant usage, already go through
> every tenant.
> The goal of this ticket is that these tasks run for every tenant.
> h3. Ways to explore
> * Run each task once per tenant, with that tenant's context, as the rule
> refresh already does. Restore the previous context after each tenant so
> nothing carries over.
> * Do not fix this by letting the system context read all tenants at once.
> That would change how data is kept apart for every caller.
> * Recurring imports and exports must run with their own tenant, and keep the
> file location limits added by UNOMI-973.
> h3. How to test
> Manual check: in the "default" tenant, check that a profile older than the
> purge setting is removed, that a profile leaves a "last 7 days" segment once
> it is too old, and that a recurring import runs again after a restart.
> Automated tests to add, each with at least two tenants besides "system". Each
> must fail on the 4.0.0 release candidate and pass once the fix is in:
> * Profile purge: old profiles, sessions and events are removed in every
> tenant, and recent ones are kept.
> * Segment and scoring update: a profile in a relative-date segment of a
> tenant leaves it once the date has passed.
> * Rule statistics: statistics are refreshed for the rules of every tenant.
> * Imports and exports: after a restart of the router, the recurring
> configurations of every tenant are scheduled again. The import writes into
> its own tenant, and the file location limits still apply.
> * Separation: a task run for one tenant never changes another tenant's data,
> and the context is the same before and after the task. This guards against
> data leaking between tenants.
> * A check that the persistence filter for the system context is unchanged, so
> this fix cannot be done by widening it.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
