At the moment this response returns a list of service IDs for hundreds
of organisations.
The admin app doesn’t use this information, but having to wait for it to
be serialized and sent across the network slows it down all the same.
This is changing because we’re going to introduce accepting contracts
and MoUs online.
Previously
---
We had one column for who signed the agreement, which is foreign keyed
to the user table. This is still relevant, because there will always be
a user who is clicking the button.
Now
---
We add two new fields for the name and email address of the person on
whose behalf the agreement is being accepted. This person:
- is different from the one signing the agreement
- won’t necessarily have a Notify account
The admin app now needs to know a few extra things about orgs and
services in order to list them. At the moment it does this by making
multiple API calls.
This commit adds extra fields to the existing response. Once the admin
app is using this fields we’ll be able to remove:
- `reponse['services_without_organisations']`
- `reponse['organisations']['services']`
- Gunicorn can give us some interestings stats about what it is doing
that might prove useful when trying to work out the
gunicorn/eventlet/sqlalchemy kind of bottlenecking issues
- We already ship these env vars to the app so this is a pretty easy
change to make
- Uses statsd which is UDP so it shouldn't have any performance impact
on gunicorn
For a user to be able to be archived, each service that they are a
member of must have at least one other user who is active and who has
the 'manage-settings' permission.
To archive a user we remove them from all their services and
organisations, remove all permissions that they have and change some of
their details:
- email_address will start with '_archived_<date>'
- the current_session_id is changed (to sign them out of their current
session)
- mobile_number is removed (so we also need to switch their auth type to
email_auth)
- password is changed to a random password
- state is changed to 'inactive'
If any of the steps fail, we rollback all changes.
a little complicated because the free_sms_fragment_limit comes from
the annual_billing table. This relies on there always being at least
one row for every service on annual billing - I checked on prod and
that is true.
Join to the annual billing table, then join to a subquery getting the
latest year for that service to extract only the most recent year.
a bit of DRY - use the column definitions to determine what goes into
the dict, and use a `next` iterator rather than a while loop to find
the existing service row. Take advantage of dict mutability to avoid
needing to refer to the list by index.
Also change the tests so if there's an error, the diff is slightly
more readable. But not much
* remove unnecessary parametrisation where we're not testing
type-specific things
* remove some unused fixtures
* replace len(x.all()) with x.count()
* don't deleted notificationhistory now that we know it's not written
* remove some pre-test function asserts
re-order notification dao delete notifications test to move "fixtures"
to the top of the file
changed create_service_data_retention to take an ORM object, not an id.
brings it in line with other db.py test functions
Update subquery to run again but for test keys. Test data is never inserted in Notifications so they need to be deleted separately now given the join to NotificationHistory.
the celery decorator should always be on the outside so that all other
decorators will be captured within the celery task. We had problems
with cronitor not reporting, and only for this task.
- We are running a statsd exporter on tools to collect all our statsd
metrics for scraping by Prometheus
- Update preview to point there instead of at the local one which has
issues with redeployment and DNS changing
- We've been seeing an issue when traffic spikes of the http health
checks taking over 1s and PaaS killing the app
- Port health checks won't care about being stuck in a queue so should
continue to work even at high loads
- We have functional tests to catch if a deployment brings up the app
(and so passes port health check) but then doesn't work
Change method name to be more relevant
Check if verification notification we send is a correct one
pass in notify db session for tests instead of sample_user
Refactor tests by using admin_request instead of client
Refactor all tests for affected reply-to endpoints for good measure
Allow overwriting reply-to address with the same address
Skip checking for duplicates if it's an reply-to email update
Fix refactored tests
Verify duplicates exception not needed
Check for duplicate reply-to email address has been added on:
-verification endpoint, so we do not send the verifying notification
needlessly
- add reply-to email address and update reply-to email address
endpoints, as those can be hit multiple times after the email address
has been verified (so the same email address could end up being added
multiple times). EDIT: this has now been prevented on admin app,
but it's better to retain double-check for safety.
