akvo / akvo-mis / #1339

90%

push

coveralls-python

[#507] Purge workspaces that never finished signing up (#509)

* [#507] Bind the activation link's lifetime to a new purge window setting

Introduces `TENANT_PURGE_AFTER_HOURS`, defaulting to 48, which the
purge job in the next commit will measure against. The activation
link's lifetime now derives from it instead of being a hardcoded
week.

The trade-off is that activation links get much shorter — a week to
48 hours. That is deliberate rather than incidental: once an
unconfigured workspace is deleted after 48 hours, a link that stayed
valid for five more days would verify an account whose tenant no
longer existed. One setting cannot disagree with itself; two numbers
could.

Validation lives in `utils/purge_window.py` rather than inline in
`settings.py` so that it can be tested. A guard inside `settings.py`
is only reachable by re-importing a module Django has already cached,
which means in practice it could not be tested at all. A non-positive
window is refused at boot because its symptoms — every workspace
purged on sight, every link dead on arrival — would look like a bug
anywhere except at the setting that caused them.

* [#507] Purge workspaces that never finished signing up

Sign-up phase 1 claims a subdomain so that nobody can take the name
during the email round-trip, and nothing ever gave the claim back. An
abandoned sign-up held the name forever, on a free tier where signing
up costs nothing. This gives it back.

"Never finished" reuses `tenant_is_configured` rather than asking the
question again in SQL. That predicate is already shared with the
configuration gate and the bulk-upload gate, and a second spelling of
it inside a destructive job is exactly where two definitions drift
apart.

The safety guard is the database, not a pre-flight check. Every FK to
`Tenant` is PROTECT, so a workspace that still owns anything makes
the delete raise before a row is touched, and the exception names
what held it. An enumerated check would be a second lis... (continued)

8189 of 9303 branches covered (88.03%)

Branch coverage included in aggregate %.

15256 of 16627 relevant lines covered (91.75%)

0.92 hits per line