The test is still flakey currently. A few attempts have
been made to fix it, but it seems that the bigtable API
does not consistently return the resource after is has
been created.
While exploring ways to make it more robust, disabling the
test to keep the CI tests from being unreliable to the point
of useless.
The client's List continues to be flakey. disabling until
the endpoint is stable.
Switching tests to use ansible 2.14, as 2.13 has issues with
spawning a docker container.
Upgrading the base python version as a consequence.
Removing the need for docker from unit tests, since that was failing
and also non-critical to run unit tests.
Some test cases are unsupported due to environmental or
blocking code restrictions (e.g. unimplemented dependent resources).
As these will likely not be fixed before the first release certified
for Ansible 2.11 and beyond, documenting them instead.
The spanner List API is returning intermittent 404s, which
is resulting in the spanner_database test failing every other
time.
Disabling the API until the test is more stable.
gcp_spanner_database erroneously used LRO handling on delete,
rather than a standard synchronous function. One side effect
was not catching 404s as a not found error code.
skip_initial_version_creation previously defaulted to None,
which results in an invalid value in the create call to
gcp_kms_crypto_key.
Defaulting to False fixes the issue.
Incorporating a fix for GKE 1.19+ (See #444).
Inlined:
Google has removed basic-auth method from within GKE starting version 1.19
This lead the output response of the backend API not to provide basic-auth
data (username and password) anymore.
The current implementation of gcp_container_cluster when generating the kubectl
config file, always set basic-auth data w/o checking if there actually are
available or explicitly provided even when the value are not set/provided
from gcp_container_cluster.
In addition, re-enabling some tests that #444 fixed.
Co-authored-by: Xavier Lamien <laxathom@lxtnow.net>
fixing a flakey service account test. Sometimes the service account
generated can be less than 6 characters (e.g. sa-57), which is an
invalid length.
This ensures that the name is at least 6 characters.
- vpn_gateway required minimal changes (stop using legacy load
balancers)
- compute_vpn_tunnel requires a valid vpn configuration
(several dependent forwarding rules and configuration to set up a
proper tunnel).
the following tests can be enabled without any changes:
- gcp_compute_target_http_proxy
- gcp_compute_target_https_proxy
- gcp_compute_target_ssl_proxy
- gcp_compute_target_tcp_proxy
The backend_service resource was not regional, which is in invalid
configuration for a regional proxy to depend on.
Fixing up cleanup-script to better handle region/ global resources.
compute selflinks return back `wwww.googleapis.com` as the domain,
which was causing a perma-diff. Fixing google.cloud's normalization
fixes that.
Fixing the following tests as well, since creating an instance
now works:
- gcp_compute_instance
- gcp_compute_instance_group
- gcp_compute_instance_group_manager
- gcp_compute_instance_template
- gcp_compute_region_autoscaler
- gcp_compute_region_instance_group_manager
- gcp_compute_target_instance
- gcp_compute_target_pool
with #393 merged, it is now possible to enable gcp_compute_backend_test.
Adding a temporary fix for server-normalized value of capacity_scaler,
which is resolving into perma-diff due to "1.0" != "1".
- gcp_tpu_node was using an old TensorFlow version, as well as using
a popular TPU variant (3-8) in a zone that had a capacity limit.
Both have been updated to working values.
- gcp_spanner_instance works, enabling for CI.
- gcp_resourcemanager_project is running into quota issues for
listing the API to verify state. Adding a to-do as well as
removing some people-specific names.
- bootstrap code was not correctly instantiating the cloud function.
- compute_instance has a typo from a previous commit.
- re-disabling forwarding rule as it still has the backend_service
fingerprint requirement
- remove the global health check (not necessary to create a backend
service)
- configure backend service to external, as internal backend services
cannot use the HTTP protocol.
The test does not yet pass due to a discovered change, but
some progress was made.
- updating disk image to an existing, non-EOL image.
- updating subnetwork configuration.
