Login failures will happen any time that we aren't coming from a
whitelisted IP for that particular TLD. Since whitelists are out of date
(and we don't whitelist IPs for every TLD anyway) those failures aren't
interesting. Store and fully-log the interesting failures if one
happened.
Using the new GoogleCredentials to access Drive API caused 403 forbidden
exception. So, this PR brought back the old GoogleCredential to
temporarily resolve the production issue while we are figuring out the
long term fix.
TESTED=Deployed to alpha and verified exportPremiumTerms succeeded, see
https://paste.googleplex.com/6153215760400384.
* Fail gracefully when copying detailed reports
When the detailed reports are copied from GCS to registrars' Drive
folders, do not fail the entire copy operation when a single registrar
fails. Instead, send an alert email about the failure, and continue to copy the
rest of the reports.
Also, instead of creating duplicates, overwrite the existing files on
Drive.
BUG=127690361
* Build docker image of nomulus tool
In the course of "gradle build", build a docker image of nomulus tool so that
users can run this to allow us to bundle the java version with the image.
* Don't extend expiration times for deleted domains
* Flip order and add a comment
* oops forgot a period
* Use END_OF_TIME
* Add tests for expiration times of domains with pending transfers
* Add test for transfer during autorenew and clean up other tests
* Clarify tests
* Add domain expiration check in EppLifecycleDomainTest
* Add a comment and format test files
* Add a metric for EPP processing time regardless of ID/TLD
* Change name to request_time
* Record EPP processing time by traffic type
* grammar
* request type
* semicolon
* Remove the maybeRuntime configuration
It contains dependencies present in the bazel
build but not needed for compile. We now know
they are not needed in runtime either.
Sometimes, the webdriver tests get stuck forever for no reason. It could
be some issue in the test container but it is hard to root cause it. So,
adding a 30s timeout can either trigger the retry earlier or let the
test just fail.
This PR prevents Gradle from copying the golden images
to build/resources/test, so the screenshot test would
read golden images from src/test/resources directly and
display the path in test log if the test fails. Because
the path pointing to the actual file in src/ folder,
the engineer can easily find it.
We got non-serialization object error when deploying the invoicing
pipeline. It turns out that Beam requires every field in the pipeline
object is serilizable. However, it is non-trivial to make
GoogleCredentialsBundle serilizable because almost all of its
dependency are not serilizable and not contraled by us. Also,
it is non-necessary to inject the credential as the spec11
pipeline also writes output to GCS without having injected
credential. So, removing the injected variable can solve the
problem.
TESTED=First reproduced the problem locally by deploying the invoicing pipeline with the previous code; applied this change and successfully deploy the pipeline without having any issue.
* Attempt login to MosAPI via all available TLDs
There's no reason why we should need a TLD as input here because it
doesn't actually matter which one we use (they all have the same
password).
* Refactor the TLD loop and change cron jobs
* Re-throw the last exception if one exists
* Fix tests and exception
* Remove alpha cron job
* Move test resource files into src/test/resources
* fix a test
* Remove references to javatests/ in Java files
* fix import order
* fix semantic merge conflict
* Throw a more useful error message on attempted domain restore reports
Per DomainRestoreRequestFlow's Javadoc, we automatically approve and instantly
enact all domain restore requests, thus we don't use or support restore
reports. This improves the registrar-visible error message to help make this
more clear.
Replace deprecated GoogleCredential with new lib
This PR also introduced a CredentialsBundle class to carry
HttpTransport and JsonFactory object which are needed by
most of the GCP library to instantiate client.
* Refactor the license check to be a generic presubmit check
This includes all of the old presubmits that were in our METADATA file
that are still valid.
* Responses to CR
Added docstrings, refactored the file finder, changed variable names,
and formatted the file
* More respones to CR
We found that some webdriver tests failed because RegistryEnvironment
was set to 'production' by other test and was carried over to the
webdriver test, and it was nontrivial to fix them because the instance
of RegistryEnvironment was injected into the testing web server.
Also, to eliminate this problem from potential victims, we stopped
provisioning RegistryEnvironment from Dagger. Instead, we just use
RegistryEnvironment.get() function to get the instance, which indeed
retrives the value from system property every time. If any test case
needs to run the test with other environment, it can just simply use
the existing SystemPropertyRule and RegistryEnvironment.${ENV}.setup
to change it.
Our CI (Travis & Kokoro) runs "gradle build", so we need to make
sure that all tasks used in the release process are called during
the build so that breakage can be caught earlier.
In order to stage the GAE folder we need gcloud to be present.
Therefore the Travis config is changed to install gcloud.
See: https://gist.github.com/mjackson/5887963e7d8b8fb0615416c510ae8857
IntelliJ is complaining when this annotation is used on a base class that has no
actual runnable tests itself. The solution is different for different classes
depending on the existing pattern of where the @RunWith annotation is; for some,
it's simplest to move the annotation down to the few extending classes that are
missing it, whereas in others it's easiest just to annotate the base class as
abstract.
Fix or suppress deprecation warnings except those about GoogleCredential,
which is being handled separately.
The @SuppressWarnings("deprecation") annotation does not cover imports
even when it is at the class level. We removed imports of deprecated
classes and use their fully qualified names in class body.
* Consolidate testcontainer used by WebDriver test
Previously, we hard coded the version of the docker image for
provisioning Chrome browser and WebDriver server because the
version of the Chrome browser has to match the version of
the webdriver client, otherwise the screenshot test will fail.
Changing to use BrowserWebDriverContainer can delegate the match
to the library itself because it chooses the correct docker image
version based on the WebDriver version on our classpath.
* Increase maxColorDiff to 20
This is to supress the test flakness after switching to use
BrowserWebDriverContainer to provision browser and webdriver
service.
