This parameter is misleading and does not do what it purports to do.
Namely, it does not impact the level of parallelism. Given the input n for this
parameter, and m for the batch size, the elements are divided (keyed) into n
groups, each of which are then spread evenly across all threads, which
are eventually in turn batched into batches with size m:
https://github.com/apache/beam/blob/master/sdks/java/core/src/main/java/org/apache/beam/sdk/transforms/GroupIntoBatches.java#L227
This is also evident in the implementation itself, where the ShardedKey
is determined by the unique number for a worker/thread combo and the
original key:
https://github.com/apache/beam/blob/master/sdks/java/core/src/main/java/org/apache/beam/sdk/transforms/GroupIntoBatches.java#L268
Using a more concrete example, suppose we have 100 elements and 10
worker threads, with a target batch size of 5. If the "shard" number is set to
1, we first spread the 100 elements across 10 threads, resulting in 10
elements per thread, each thread then batches the elements into 2
batches of size 5.
If the "shard" number is set to 2, the 100 elements are first divided into 2
"shards" of 50 each. Each "shard" is then distributed within the 10
threads, resulting in 5 elements per "shard" per thread. They then get
turned into 1 batch per "shard" per thread. In the end, each thread
still processes 2 batches, even though they are from 2 different "shards".
Therefore this "shard" number does not perform horizontal partitioning
that one normally associates with sharding, and provides no
performance benefits but rather confuses the user.
It is also suggested that using withShardedKey() alone is already
sufficient to achieve auto-sharding within the keyed group. There is no
need to manually divide the input by keying them differently based on
the "shard" number specified:
https://youtu.be/jses0W4Zalc?t=967
Some retriers are no longer needed because transactions are
automatically retried by the JPA transaction manager when there's a
transient exception.
<!-- Reviewable:start -->
- - -
This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/google/nomulus/1874)
<!-- Reviewable:end -->
We no longer use App Engine Remote API as of #1858.
The pipeline endpoint is only for GAE mapreduce, which we stopped doing
for a while.
TESTED=deployed to alpha and used nomulus tool built from master to
connect to the tools service.
<!-- Reviewable:start -->
- - -
This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/google/nomulus/1873)
<!-- Reviewable:end -->
* Add defaultPromoTokens to Registry
* Remove flyway files from this PR
* Fix merge conflicts
* Add back flyway file
* Add more info to error messages
* Change to a list
* Fix javadoc
* Change error message
* Add note to field declaration
These were always properly reflected in the actual creations, but when
running a check flow it would still show a non-zero cost even when using
an ANCHOR_TENANT allocation token. This changes it so that we accurately
show the $0.00 cost.
This is only used for contacting Datastore. With the removal of:
1. All standard usages of Datastore
2. Usage of Datastore for allocation of object IDs
3. Usage of Datastore for GAE user IDs
we can remove the remote API without affecting functionality.
This also allows us to just use SQL for every command since it's lazily
supplied. This simplifies the SQL setup and means that we remove a
possible situation where we forget the SQL setup.
So long, farewell, adios, ciao, sayonara, 再见!
TESTED=deployed to alpha and used `nomulus list_tlds` to confirm that the web app can receive and serve requests.
<!-- Reviewable:start -->
- - -
This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/google/nomulus/1863)
<!-- Reviewable:end -->
Switch to using the login email address instead of GAE user ID to
identify console users. The primary use cases are:
1) When the user logged in the registrar console, need to figure out
which registrars they have access to (in
AuthenticatedReigstrarAccess).
2) When a user tries to apply a registry lock, needs to know if they
can (in RegistryLockGetAction).
Both cases are tested in alpha with a personal email address to ensure
it does not get the permission due to being a GAE admin account.
Also verified that the soy templates includes the hidden login email
form field instead of GAE user ID when registrars are displayed on the
console; and consequently when a contact update is posted to the server,
the login email is part of the JSON payload. Even though it does not
look like it is used in any way by RegistrarSettingsAction, which
receives the POST request. Like GAE user ID, the field is hidden, so
cannot be changed by the user from the console, it is also not used to
identify the RegistryPoc entity, whose composite keys are the contact
email and the registrar ID associated with it.
The login email address is backfilled for all RegistrarPocs that have a
non-null GAE user ID. The backfilled addresses converted to the same ID
as stored in the database.
It doesn't actually use any App Engine libraries or code -- it's just a
generic connection with authentication to a service. This also involves
changing that block of config to be "gcpProject" instead of "appEngine"
since it's more generic.
Note: this will require an internal PR as well to change the
corresponding private config block
* Remove package token on manual transfer approval
* remove extra variables
* Add back white space
* Don't overwrite existingDomain
* Format fixes, use available helper variables
* Use PACKAGE allocation tokens in tests
* Refactor
* Fix merge conflicts
* Dont overwrite existingRecurring
* Remove names from packages on automatic transfers
* Add more tests
* Remove unneccesary local variable
* Eliminate unnecessary api call
* Reformat if blocks
* Don't overwrite existingRecurring
This PR fixes the issue where the onetime billing event for an autorenew
is not correctly created if the recurrence of the autorenew is closed
during the autorenew grace period, such as the case if a manual renew
happens during the same grace period.
The detailed analysis of the issue is captured in b/258822640. Note that
this is a quick and dirty fix to make ongoing billing event expanse work
correctly in the future. It does not fix the missing events in the past,
nor can it be used to reconstruct the missing ones (by providing a
different cursor time), due to timeout when triggering the action from
nomulus curl.
Per Weimin, the recurrences that fits the new condition along, based on
the current cursor, would increase from 382k to 430k, a 12% increase. I
checked last nights cron job run, which starts on 22:00 EST and seemed
to finish at 22:15 EST (when the last log for this request was
recorded), so it should definitely still finish in time for the nightly
runs with the new condition.
This is not a complete removal of ofy as we still a dependency on it
(GaeUserIdConverter). But this PR removed it from a lot of places where
it's no longer needed.
* Fix Gradle dependency version pinning
In Gradle 7, version labels require '!!' at the end to be free from
any forced upgrade.
Hibernate min version needs to be advanced past 5.6.12, which is buggy.
Upgraded most dependencies to the latest version.
Therefore this PR also removed several classes and related tests that
support the setup and verification of Ofy entities.
In addition, support for creating a VKey from a string is limited to
VKey<? extends EppResource> only because it is the only use case (to
pass a key to an EPP resource in a web safe way to facilitate resave),
and we do not want to keep an extra simple name to class mapping, in
addition to what persistence.xml contains. I looked into using
PersistenceXmlUtility to obtain the mapping, but the xml file contains
classes with the same simple name (namely OneTime from both PollMessage
and BillingEvent). It doesn't seem like a worthwhile investment to write
more code to deal with that, when the fact is that we only need to
consider EppResource.
* Remove Cloud KMS from Nomulus Server
Removed Cloud KMS from the Nomulus (:core) since it is no longer used.
Renamed remaining classes to reflect their use of the SecretManager.
Updated the config instructions to use a new codename for the keyring:
KMS to CSM. This PR works with both codenames. Will drop 'KMS' after
the internal repo is updated.
Add a implementation of Delegated credential without using downloaded private key.
This is a stop-gap implementation while waiting for a solution from the Java auth library.
Also added a verifier action to test the new credential in production. Testing is helpful because:
Configuration is per-environment, therefore, success in alpha does not fully validate prod.
The relevant use case is triggered by low-frequency activities. Problem may not pop out for hours or longer.
This PR removes all Ofy related cruft around `HistoryEntry` and its three subclasses in order to support dual-write to datastore and SQL. The class structure was refactored to take advantage of inheritance to reduce code duplication and improve clarity.
Note that for the embedded EPP resources, either their columns are all empty (for pre-3.0 entities imported into SQL), including their unique foreign key (domain name, host name, contact id) and the update timestamp; or they are filled as expected (for entities that were written since dual writing was implemented).
Therefore the check for foreign key column nullness in the various `@PostLoad` methods in the original code is an no-op as the EPP resource would have been loaded as null. In another word, there is no case where the update timestamp is null but other columns are not.
See the following query for the most recent entries in each table where the foreign key column or the update timestamp are null -- they are the same.
```
[I]postgres=> select MAX(history_modification_time) from "DomainHistory" where update_timestamp is null;
max
----------------------------
2021-09-27 15:56:52.502+00
(1 row)
[I]postgres=> select MAX(history_modification_time) from "DomainHistory" where domain_name is null;
max
----------------------------
2021-09-27 15:56:52.502+00
(1 row)
[I]postgres=> select MAX(history_modification_time) from "ContactHistory" where update_timestamp is null;
max
----------------------------
2021-09-27 15:56:04.311+00
(1 row)
[I]postgres=> select MAX(history_modification_time) from "ContactHistory" where contact_id is null;
max
----------------------------
2021-09-27 15:56:04.311+00
(1 row)
[I]postgres=> select MAX(history_modification_time) from "HostHistory" where update_timestamp is null;
max
----------------------------
2021-09-27 15:52:16.517+00
(1 row)
[I]postgres=> select MAX(history_modification_time) from "HostHistory" where host_name is null;
max
----------------------------
2021-09-27 15:52:16.517+00
(1 row)
```
These alternative ORMs are introduced as a way to make querying large number of
domains and domain histories more efficient through bulk loading from several
to-be-joined tables separately, then in-memory re-assembly of the final entity,
bypassing the need to query multiple tables each time an entity is queried.
Their primary use case is loading these entities for comparison between
datastore and SQL during the migration, which has been completed. The
code remain unused as of now and their existence makes refactoring and
general maintenance more complicated than necessary due to the need to keep
them up to date.
Therefore we remove the related code.
<!-- Reviewable:start -->
- - -
This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/google/nomulus/1834)
<!-- Reviewable:end -->
Also removed the ability to disable update timestamp auto update as it
was only needed during the migration.
Lastly, rectified the use of raw Coder in RegistryJpaIO.
Passing in an already-existing instance is an antipattern because it can
lead to race conditions where something else modified the object in
between when the pipeline loaded it and when you're saving it. The Write
action should only be writing new entities.
We cannot check IDs for the objects (some IDs are not autogenerated so
they might exist already). We also cannot call `insert` on the objects
because the underlying JPA `persist` call adds the input object to the
persistence context, meaning that any modifications (e.g.
updateTimestamp) are reflected in the input object. Beam doesn't allow
modification of input objects.
* Does not self allocate IDs in Beam by default.
Per b/250948425, it is dangerous to implicitly allow all Beam pipelines
to create buildables by self allocating the IDs. This change makes it so
that one has to explicitly request self allocation in Beam.
A boolean is added to the pipeline option so that it can be passed to
the beam worker initializer that controls the behavior of the JVM on
each worker. Note that we did not add the option in the metadata.json file
because we did not want people to use the override at run time when launching
a pipeline, due to the risk. As shown in RdePipeline.java, we instead
explicitly hard-code the option in the pipeline. There is nothing that
stops one to supply that option when launching the pipeline, but it's
not advised.
Tested=deployed the pipeline alpha and ran it.
* Properly create and use default credential
This PR consists of the following changes:
- Stopped adding scopes to the default credential when using it to access other
non-workspace GCP APIs. Scopes are not needed here.
- Started applying scopes to the default credential when using to access
Drive and Sheets APIs.
- Upgraded Drive access from the deprecated credential lib to the
up-to-date one
- Switched Sheet access from the exported json credential to the
scoped default credential.
This PR requires that the affected files be writable to the default
service account (project-name@appspot.gserviceaccount.com) of the
project.
- This is already the case for exported files (premium terms, reserved
terms, and domain list).
- The registrar sync sheets in alpha, sandbox, and production have been
updated with the new permissions.
All impacted operations have been tested in alpha.
* Properly create and use default credential
This PR consists of the following changes:
- Added a new method to generate scope-less default credential when using it to
access other non-workspace GCP APIs. Scopes are not needed here.
- Started to use the new credential in the SecreteManager.
- Will migrate other usages to this new credential gradually.
- Marked the old DefaultCredential as deprecated.
- Started applying scopes to the default credential when using to access Drive
and Sheets APIs.
- Upgraded Drive access from the deprecated credentials lib
- Switched Sheet access from the exported json credential to the scoped
default credential.
This PR requires that the affected files be writable to the default service
account (project-name@appspot.gserviceaccount.com) of the project.
- This is already the case for exported files (premium terms, reserved terms,
and domain list).
- The registrar sync sheets in alpha, sandbox, and production have been
updated with the new permissions.
All impacted operations have been tested in alpha.
This reverts commit 1ab077d267.
Apparently the new version of Spinnaker that is compatible with this doesn't
work for our release, so we need to roll this back for now. (Again!)
