This runs over all domains that weren't deleted as of September 5. This
will fix most of b/248112997, which is itself caused by b/245940594 --
creating synthetic history objects means that the RDE pipeline will look
at those instead of the potentially-no-longer-valid data in the old
history objects.
Also fixed a bug introduced in #1785 where identity checked were performed instead of equality. This resulted in two sets containing the same elements not being regarded as equal and subsequent DNS updated being unnecessarily enqueued.
The old class is modeled after datastore with some logic jammed in for it to work with SQL as well. As of #1777, the ofy related logic is deleted, however the general structure of the class remained datastore oriented.
This PR refactors the existing class into a ForeignKeyUtils helper class that does away wit the index subclasses and provides static helper methods to do the same, in a SQL-idiomatic fashion.
Some minor changes are made to the EPP resource classes to make it possible to create them in a SQL only environment in tests.
`TransferData` is currently a generic class with a complicated type parameter that designate the `Builder` class of its concrete subclass, on order to facilitate returning the said `Builder` from an instance loosely typed to the superclass (`TransferData`) itself.
While this works, in most all places that a `TransferData` is used, the raw, un-generic type is declared, resulting a lot of warnings, not to mention the fact that type safety not actually checked when raw type is used.
In this PR, we make it so that the concrete `Builder` is returned through a protected abstract method that is implemented by the subclasses. The type information therefore no longer needs to be embedded in the superclass type signature, and reflection is not necessary to create the `Builder` either. Overall, it makes `TransferData` a much cleaner class without the messiness of generics.
This uses the GoogleIdTokenVerifier to verify ID tokens passed in
(presumably from a front end) via cookies. This isn't used anywhere yet
but it will be used for front-end API calls for the new console.
First, we create a sequence of User IDs in Postgres and assign it to the
User ID field, meaning that Hibernate can autogenerate IDs.
Next, add an update timestamp.
Next, add a constraint that we can't have multiple Users with the same
email address.
Finally, create a DAO since we'll usually want to query by that email
address (at least for now).
FKI used to be persisted in datastore to help speed up loading by foreign key.
Now it is just a helper class to do the same thing in SQL because
indexing is natively supported in SQL.
- Create a sequence to generate IDs for the user (this allows us to have
Long ID types so that Hibernate can autogenerate IDs)
- Add an update timestamp column so we can extend BackupGroupRoot
- Add a restriction that there can't be multiple users with the same
email address
* Rename ContactResource -> Contact
This is a follow-up to PR #1725 and #1733. Now all EPP resource entity class
names have been rationalized to match with their SQL table names.
This means that LegacyAuthenticationMechanism or a to-be-created
OAuth2AuthenticationMechanism) can return a UserAuthInfo object that
contains either the GAE User or the console User as appropriate. The
goal is that the non-auth flows shouldn't have to know about which user
type it is. Note: the registry lock flow (for now) needs to know about
the separate types of auth because it is a separate level of auth from
the standard AuthenticatedRegistrarAccessor.
The AuthenticatedRegistrarAccessor code is a bit odd because the new
role system doesn't quite fit neatly into the old registrar ->
OWNER,ADMIN system but this is a fine approximation. Basically, any
new registrar role will map to the old OWNER role.
* Add the PackagePromotion table
* Add long id
* Add NOT NULL
* fix formatting
* make package price non null
* Add not nulls to java file
* Fix broken tests from merge conflicts
Also deletes the autorenew poll message history revision id field in
Domain, which is only needed to recreate the ofy key for the poll
message. The column already contains null values in it, making it
impossible to depend on it. The column itself will be deleted from the
schema after this PR is deployed.
The logic to update autorenew recurrence end time is changed
accordingly: When a poll message already exists, we simply update the
endtime, but when it no longer exists, i. e. when it's deleted
speculatively after a transfer request, we recreate one using the
history entry id that resulted in its creation (e. g. cancelled or rejected
transfer).
This should fix b/240984498. Though the exact reason for that bug is
still unclear to me. Namely, it throws an NPE at this line during an
explicit domain transfer approval:
https://cs.opensource.google/nomulus/nomulus/+/master:core/src/main/java/google/registry/flows/domain/DomainFlowUtils.java;l=603;bpv=1;bpt=0;drc=ede919d7dcdb7f209b074563b3d449ebee19118a
The domain in question has a null autorenewPollMessageHistoryId, but
that in itself should not have caused an NPE because we are not
operating on the null pointer. On that line the only possible way to
throw an NPE is for the domain itself to be null, but if that were the
case, the NPE would have been thrown at line 599 where we called a
method on the domain object.
Regardless of the cause, with this PR we are using an explicitly
provided history id and checking for its nullness before using it. If a
similar issue arises again, we should have a better idea why.
Lastly, the way poll message id is constructed is largely simplified in
PollMessageExternalKeyConverter as a result of the removal ofy parent
keys in PollMessage. This does present a possibility of failure when
immediately before deployment, a registrar requests a poll message and
received the old id, but by the time the registrar acks the id, the new
version is deployed and therefore does not recognize the old key. The
likelihood of this happening should be slim, and we could have prevented
it by letting the converter recognize both the old and the new key.
However, we would like to eventually phase out the old key, and in
theory a registrar could ack a poll message at any time after it was
requested. So, there is not a safe time by which all the old ids are
acked, lest we develop some elaborate scheme to keep track of which
messages were sent with an old id when requested and which of these old
ids are acked. Only then can we be truly safe to phase out the old id.
The benefit does not seem to warrant the effort. If a registrar does
encounter a situation like this, they could open a support bug to have
us manually ack the poll message for them.
