apache / apache/openserverless

[Feature] Separate user deprovisioning from verified tenant data purge

Offen
#215 0 Kommentare 0 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen
enhancement
Vorherrschende Sprache
Python
Sterne
576
Forks
29
Ø Merge
50 Min.
Gemergte PRs (30 T.)
13

Beschreibung

## Feature request

Define explicit and safe lifecycle semantics for tenant removal. Today, deleting a user mixes identity deprovisioning with partial backend cleanup, while the command name suggests that all tenant data is removed.

## Current behavior

Depending on the task/operator version, `ops admin deleteuser ` deletes the `WhiskUser` and then attempts local CLI config cleanup. The operator cleanup is not an atomic, verified tenant purge:

- the OpenWhisk subject/auth document is removed, while tenant entities such as actions can remain;
- Redis cleanup removes the ACL user, while keys under the tenant prefix can remain;
- cleanup is attempted for other enabled services, but the command does not expose a reliable per-backend postcondition report;
- recreating a tenant with the same namespace reuses the same logical identifiers and can make surviving data visible again.

In one observed environment, after delete and recreate, dozens of OpenWhisk actions and more than 1,000 tenant-prefixed Redis keys were still present. Only aggregate counts were inspected; no tenant data content was read.

This makes it difficult for an administrator to know whether the requested operation means "revoke access but preserve data" or "irreversibly erase the tenant".

## Proposed command model

Introduce two distinct operations:

1. `ops admin deprovisionuser `
- revoke credentials, routes and service identities;
- preserve tenant data for recovery or retention;
- report which resources remain.

2. `ops admin purgeuser --confirm=`
- irreversibly delete all resources owned by the tenant;
- verify every postcondition before reporting success;
- return non-zero when residual resources remain.

For compatibility, `deleteuser` could remain as an alias for deprovisioning with a clear deprecation/message, rather than silently changing its destructive semantics.

The proposed `purgeuser` command does **not** currently exist.

## Purge contract

A purge should cover enabled tenant resources, including:

- OpenWhisk packages, actions, triggers, rules and applicable activation data;
- Redis keys selected by the exact authorized tenant prefix, followed by ACL user removal;
- object storage buckets, objects and service user;
- MongoDB/FerretDB, PostgreSQL and Milvus tenant databases, schemas or collections;
- generated ingress/routes, static hosting resources and operator metadata.

Custom or out-of-band resources should be discovered and reported, but not deleted blindly.

## Safety and operability

- Add `--dry-run` to inventory affected resources without mutation.
- Require an explicit confirmation that includes the namespace name.
- Treat already-missing resources as success so retries are idempotent.
- Produce a structured per-backend summary with deleted, preserved, skipped, failed and residual states.
- Verify postconditions for each enabled backend.
- Avoid logging credentials, secrets or the complete `WhiskUser` specification.
- Optionally support retention/preserve flags and a pre-purge backup hook.

## Acceptance criteria

- Deprovisioning and purge have documented, unambiguous semantics.
- Recreating a deprovisioned namespace can intentionally recover preserved data.
- Recreating a successfully purged namespace starts empty.
- Partial failures return non-zero and identify the affected backend/resources.
- Re-running purge after partial failure completes safely.
- Dry-run and confirmation behavior have automated tests.
- Tests cover OpenWhisk entities, Redis prefixed keys and each optional data service enabled in CI.

## Affected components

This likely spans:

- `apache/openserverless-task` admin task and CLI surface;
- `apache/openserverless-operator` `WhiskUser` reconciliation/finalization and backend cleanup handlers;
- downstream task distributions that expose `ops admin deleteuser`.

## Environment where observed

- ops CLI: `0.9.1-2607121109.dev`
- task branch: `apache/openserverless-task:0.9.1`
- operator image: `apache/openserverless-operator:0.1.0-incubating.2512151437`

Beitragsleitfaden

Beitragsleitfaden öffnen

Rechercherichtung

Beginne bei der Admin-Task- und CLI-Oberfläche von apache/openserverless-task, insbesondere bei deleteuser, und untersuche anschließend die WhiskUser-Reconciliation/Finalisierung sowie die Handler für die Backend-Bereinigung in apache/openserverless-operator. Vergleiche das aktuelle Verhalten mit dem vorgeschlagenen Deprovisioning-/Purge-Vertrag; als abgeschlossen gilt die Arbeit, wenn dokumentierte Semantik sowie automatisierte Tests für Dry-Run, Bestätigung, Wiederholung und Backend-Postcondition vorhanden sind, die die aufgeführten Services abdecken.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Bewertung

Tech-Stack
kubernetes, mongodb, postgresql, python, redis
Bereich
backend, cli, databases, devops
Issue-Typ
Feature
Schwierigkeit
5/5
Geschätzter Aufwand
Über eine Woche
Aktivitätsstatus
Ruhig
Klarheit
Größtenteils klar
Anfängerfreundlichkeit
25/100

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.