This should prevent having issues with hot key paths on entities that
experience a heavy WHOIS volume (e.g. contacts that registrars reuse on
many domains).
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=191506124
After investigating common domain create/update command usage
patterns by registrars, we noticed that it is frequent for a
given registrar to reuse both hosts (using a standardized set of
nameservers) as well as contacts (e.g. for privacy/proxy
services). With these usage patterns, potential per-registrar
throughput during high volume scenarios (i.e. first moments of
General Availability) suffers from hitting hot keys in Datastore.
The solution, implemented in this CL, is to add short-term
in-memory caching for contacts and hosts, analogous to how we are
already caching Registry and Registrar entities. These new
cached paths are only used inside domain flows to determine
existence and deleted/pending delete status of contacts and
hosts. This is a potential loss of transactional consistency, but
in practice it's hard to imagine this having negative effects, as
contacts or hosts that are in use cannot be deleted, and caching
would primarily affect widely used contacts and hosts.
Note that this caching can be turned on or off through a
configuration option, and by default would be off. We'd only want
it on when we really needed it, i.e. during a big launch.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=187093378
We can easily end up enlisting too many entity groups (separate
DomainApplications) in a TransactionalFlow when loading all applications
tracked by the DomainApplicationIndex. This makes the load operation
transactionless, to avoid overenlisting.
Potential problems:
1. We fail to prevent landrush applications, if a single sunrise application
exists. This is likely fine, except for a brief moment in Sunrush when a
sunrise application is made immediately prior to a landrush application. The
result is we accept an invalid application- which can be mediated manually.
2. We fail to prevent a domain create for a domain with an open application.
This is a little more sinister, but also unlikely unless someone submits an
application immediately before someone tries to create the same domain (sans
application?)
3. We return an invalid DomainCheck response (instead of 'pending allocation').
Not the worst outcome.
4. We reduce the AuctionStatusCommand and GetApplicationIdsCommand to
eventual consistency. Since they're internal tools, that's not too big a deal.
A better solution:
DomainApplications really should just be normalized under a virtualEntityGroup
by fullyQualifiedDomainName, or a hash-bucket like EppResources are. The
DomainApplication -> DomainBase -> EppResource hierarchy seems to be purely for
code reuse, at the cost of Datastore consistency. This would, however, require
quite some refactoring, and a custom resave operation across all
DomainApplications.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=169395586
Memcache is already off but now it's not in the code anymore.
This includes removing domain creation failfast, since that is actually
slower now than just running the flow - all you gain is a non-transactional
read over a transactional read, but the cost is that you always pay that
read, which is going to drive up latency.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=158183506
TESTED=For all tests, I added @Cache to DomainBase because otherwise the tests will
fail. We aren't ready to do this in prod yet, which is why the tests are still
marked @Ignore. The new tests fail if you change line 134 in Ofy to not use memcache
and either use the unchanged original DomainCreateFlow code, or use the new
inlined code and change loadWithMemcache() to load(). They pass with the new
inlined code that calls loadWithMemcache(), as long as the @Cache is added to
DomainResource.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=154224748
DomainApplications have nothing to project, so it's a mistake to call their cloneProjectedAtTime() method. Marking it @Deprecated helps prevent such inadvertent use.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=146716189
The next step will be to get rid of RegistryConfig descendants and RegistryConfigLoader entirely.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=143812815
With some code cleanup/refactoring/formatting by Ben McIlwain.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141199855
1) Prefer .getConstructor().newInstance() over .newInstance()
because otherwise checked exceptions can be propagated from
the constructor even though they aren't declared.
2) Use the type T in the parameters to instantiate().
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138874730
Very few flows actually check the phase. Push the checks down to the leaf
flows so that we can remove the inherited code from ResourceFlow and replace
it with utility methods. In the process, document and test two places that
throw the exception but did not previously test it.
This introduces a temporary hack in BaseDomainCreateFlow that does something
specific for DomainApplicationCreateFlow. It will go away literally tomorrow
when I flatten that flow.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=135480538
This change replaces all Ref objects in the code with Key objects. These are
stored in datastore as the same object (raw datastore keys), so this is not
a model change.
Our best practices doc says to use Keys not Refs because:
* The .get() method obscures what's actually going on
- Much harder to visually audit the code for datastore loads
- Hard to distinguish Ref<T> get()'s from Optional get()'s and Supplier get()'s
* Implicit ofy().load() offers much less control
- Antipattern for ultimate goal of making Ofy injectable
- Can't control cache use or batch loading without making ofy() explicit anyway
* Serialization behavior is surprising and could be quite dangerous/incorrect
- Can lead to serialization errors. If it actually worked "as intended",
it would lead to a Ref<> on a serialized object being replaced upon
deserialization with a stale copy of the old value, which could potentially
break all kinds of transactional expectations
* Having both Ref<T> and Key<T> introduces extra boilerplate everywhere
- E.g. helper methods all need to have Ref and Key overloads, or you need to
call .key() to get the Key<T> for every Ref<T> you want to pass in
- Creating a Ref<T> is more cumbersome, since it doesn't have all the create()
overloads that Key<T> has, only create(Key<T>) and create(Entity) - no way to
create directly from kind+ID/name, raw Key, websafe key string, etc.
(Note that Refs are treated specially by Objectify's @Load method and Keys are not;
we don't use that feature, but it is the one advantage Refs have over Keys.)
The direct impetus for this change is that I am trying to audit our use of memcache,
and the implicit .get() calls to datastore were making that very hard.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131965491
The dark lord Gosling designed the Java package naming system so that
ownership flows from the DNS system. Since we own the domain name
registry.google, it seems only appropriate that we should use
google.registry as our package name.
This change renames directories in preparation for the great package
rename. The repository is now in a broken state because the code
itself hasn't been updated. However this should ensure that git
correctly preserves history for each file.
