Automatic merge from submit-queue (batch tested with PRs 48292, 48121)
Add Google cloudkms dependency, add cloudkms service to GCE cloud provider
Required to introduce a Google KMS based envelope encryption, which shall allow encrypting secrets at rest using KEK-DEK scheme.
The above requires KMS API to create/delete KeyRings and CryptoKeys, and Encrypt/Decrypt data.
Should target release 1.8
@jcbsmpsn
Update: It appears that Godep only allows dependencies which are in use. We may have to modify this PR to include some Google KMS code.
Progresses #48522
Automatic merge from submit-queue (batch tested with PRs 46383, 45645, 45923, 44884, 46294)
Created unit tests for GCE cloud provider storage interface.
- Currently covers CreateDisk and DeleteDisk, GetAutoLabelsForPD
- Created ServiceManager interface in gce.go to facilitate mocking in tests.
**What this PR does / why we need it**:
Increasing test coverage for GCE Persistent Disk.
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes#44573
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 45573, 46354, 46376, 46162, 46366)
GCE - Retrieve subnetwork name/url from gce.conf
**What this PR does / why we need it**:
Features like ILB require specifying the subnetwork if the network is type manual.
**Notes:**
The network URL can be [constructed](68e7e18698/pkg/cloudprovider/providers/gce/gce.go (L211-L217)) by fetching instance metadata; however, the subnetwork is not provided through this feature. Users must specify the subnetwork name/url through the gce.conf.
Although multiple subnets can exist in the same region for a network, the cloud provider will only use one subnet url for creating LBs.
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue
Initialize cloud providers with a K8s clientBuilder
**What this PR does / why we need it**:
This PR provides each cloud provider the ability to generate kubernetes clients. Either the full access or service account client builder is passed from the controller manager. Cloud providers could need to retrieve information from the cluster that isn't provided through defined interfaces, and this seems more preferable to adding parameters.
Please leave your thoughts/comments.
**Release note**:
```release-note
NONE
```
The new metrics is:
cloudprovider_gce_api_request_duration_seconds{request, region, zone}
cloudprovider_gce_api_request_errors{request, region, zone}
`request` is the specific function that is used.
`region` is the target region (Will be "<n/a>" if not applicable)
`zone` is the target zone (Will be "<n/a>" if not applicable)
Note: this fixes some issues with the previous implementation of
metrics for disks:
- Time duration tracked was of the initial API call, not the entire
operation.
- Metrics label tuple would have resulted in many independent
histograms stored, one for each disk. (Did not aggregate well).
The cloudprovider is being refactored out of kubernetes core. This is being
done by moving all the cloud-specific calls from kube-apiserver, kubelet and
kube-controller-manager into a separately maintained binary(by vendors) called
cloud-controller-manager. The Kubelet relies on the cloudprovider to detect information
about the node that it is running on. Some of the cloudproviders worked by
querying local information to obtain this information. In the new world of things,
local information cannot be relied on, since cloud-controller-manager will not
run on every node. Only one active instance of it will be run in the cluster.
Today, all calls to the cloudprovider are based on the nodename. Nodenames are
unqiue within the kubernetes cluster, but generally not unique within the cloud.
This model of addressing nodes by nodename will not work in the future because
local services cannot be queried to uniquely identify a node in the cloud. Therefore,
I propose that we perform all cloudprovider calls based on ProviderID. This ID is
a unique identifier for identifying a node on an external database (such as
the instanceID in aws cloud).
Each major interface is now in its own file. Any package private
functions that are only referenced by a particular module was also moved
to the corresponding file. All common helper functions were moved to
gce_util.go.
This change is a pure movement of code; no semantic changes were made.
Automatic merge from submit-queue (batch tested with PRs 42452, 43399)
Modify getInstanceByName to avoid calling getInstancesByNames
This PR modify getInstanceByname to loop through all management zones
directly instead of calling getInstancesByNames. Currently
getInstancesByNames use a node name prefix as a filter to list the
instances. If the prefix does not match, it will return all instances
which is very wasteful since getInstanceByName only query one instance
with a specific name.
Partially fix issue #42445
When gce cloud tries to delete a disk, if the disk could not be found
from the zones, the function should return nil error. This modified behavior is also consistent with AWS
This PR modify getInstanceByname to loop through all management zones
directly instead of calling getInstancesByNames. Currently
getInstancesByNames use a node name prefix as a filter to list the
instances. If the prefix does not match, it will return all instances
which is very wasteful since getInstanceByName only query one instance
with a specific name.
Automatic merge from submit-queue (batch tested with PRs 38702, 41810, 41778, 41858, 41872)
gce: Reuse unsuccessfully provisioned volumes.
GCE PD names generated by Kubernetes are guaranteed to be unique - they
contain name of the cluster and UID of the PVC that is behind it.
Presence of a GCE PD that has the same name as we want to provision
indicates that previous provisioning did not go well and most probably
the controller manager process was restarted in the meantime.
Kubernetes should reuse this volume and not provision a new one.
Fixes#38681
GCE PD names generated by Kubernetes are guaranteed to be unique - they
contain name of the cluster and UID of the PVC that is behind it.
Presence of a GCE PD that has the same name as we want to provision
indicates that previous provisioning did not go well and most probably
the controller manager process was restarted in the meantime.
Kubernetes should reuse this volume and not provision a new one.
This method has been unused by k8s for some time, and yet is the last
piece of the cloud provider API that encourages provider names to be
human-friendly strings (this method applies a regex to instance names).
Actually removing this deprecated method is part of a long effort to
migrate from instance names to instance IDs in at least the OpenStack
provider plugin.
At master volume reconciler, the information about which volumes are
attached to nodes is cached in actual state of world. However, this
information might be out of date in case that node is terminated (volume
is detached automatically). In this situation, reconciler assume volume
is still attached and will not issue attach operation when node comes
back. Pods created on those nodes will fail to mount.
This PR adds the logic to periodically sync up the truth for attached volumes kept in the actual state cache. If the volume is no longer attached to the node, the actual state will be updated to reflect the truth. In turn, reconciler will take actions if needed.
To avoid issuing many concurrent operations on cloud provider, this PR
tries to add batch operation to check whether a list of volumes are
attached to the node instead of one request per volume.
More details are explained in PR #33760
Automatic merge from submit-queue
Loadbalanced client src ip preservation enters beta
Sounds like we're going to try out the proposal (https://github.com/kubernetes/kubernetes/issues/30819#issuecomment-249877334) for annotations -> fields on just one feature in 1.5 (scheduler). Or do we want to just convert to fields right now?
'names' is an array of FQDNs. 'instances' is a map indexed by canonicalized
name. Clearly these two won't always match, so when building the final
instance array to return, make sure to look up map entries by their canonicalized
name.
In the below example, "ocp-master-5pob" is clearly found as a GCE instance
but when building the final instance array it cannot be matched as the code
is looking for "ocp-master-5pob.c.ose-refarch.internal" instead. The node
is then deleted from the cluster as it cannot be found by the cloud provider.
gce.go:2519] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): initial node prefix ocp-
gce.go:2530] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): looking for instances map[ocp-master-5pob:<nil>]
gce.go:2533] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): getting zone 'europe-west1-c' (remaining 1)
gce.go:2563] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): instance name <omitted> not requested
gce.go:2563] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): instance name <omitted> not requested
gce.go:2533] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): getting zone 'europe-west1-b' (remaining 1)
gce.go:2563] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): instance name <omitted> not requested
gce.go:2576] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): found instance 'ocp-master-5pob' remaining 0
gce.go:2563] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): instance name <omitted> not requested
gce.go:2533] ### getInstancesByNames([ocp-master-5pob.c.ose-refarch.internal]): getting zone 'europe-west1-d' (remaining 0)
gce.go:2588] Failed to retrieve instance: "ocp-master-5pob.c.ose-refarch.internal"
gce.go:2624] ### getInstanceByName(ocp-master-5pob.c.ose-refarch.internal): got []: instance not found
gce.go:2626] getInstanceByName/multiple-zones: failed to get instance ocp-master-5pob.c.ose-refarch.internal; err: instance not found
nodecontroller.go:587] Deleting node (no longer present in cloud provider): ocp-master-5pob.c.ose-refarch.internal
nodecontroller.go:664] Recording Deleting Node ocp-master-5pob.c.ose-refarch.internal because it's not present according to cloud provider event message for node ocp-master-5pob.c.ose-refarch.internal
- test pd-ssd and pd-standard on GCE,
- test all four volume types on AWS
- test just the default volume type on OpenStack (right now, there is no API
to get list of them)
Contination of #1111
I tried to keep this PR down to just a simple search-n-replace to keep
things simple. I may have gone too far in some spots but its easy to
roll those back if needed.
I avoided renaming `contrib/mesos/pkg/minion` because there's already
a `contrib/mesos/pkg/node` dir and fixing that will require a bit of work
due to a circular import chain that pops up. So I'm saving that for a
follow-on PR.
I rolled back some of this from a previous commit because it just got
to big/messy. Will follow up with additional PRs
Signed-off-by: Doug Davis <dug@us.ibm.com>
We had another bug where we confused the hostname with the NodeName.
To avoid this happening again, and to make the code more
self-documenting, we use types.NodeName (a typedef alias for string)
whenever we are referring to the Node.Name.
A tedious but mechanical commit therefore, to change all uses of the
node name to use types.NodeName
Also clean up some of the (many) places where the NodeName is referred
to as a hostname (not true on AWS), or an instanceID (not true on GCE),
etc.
Bump version of golang.org/x/oauth2
Vendor google.golang.org/cloud/
Vendor google.golang.org/api/
Vendor cloud.google.com/go/compute/
Replace google.golang.org/cloud with cloud.google.com/go/
Fixes#30069
Automatic merge from submit-queue
Do not allow creation of GCE PDs in unmanaged zones.
Such volumes then couldn't be deleted as `getDiskByNameUnknownZone` goes through managed zones only.
Fixes: #31948
@kubernetes/rh-storage
@saad-ali, PTAL.
Automatic merge from submit-queue
Typos and englishify pkg/cloudprovider + pkg/dns + pkg/kubectl
**What this PR does / why we need it**: Just fixed some typos + "englishify" in pkg/cloudprovider + pkg/dns + pkg/kubectl
**Which issue this PR fixes** : None
**Special notes for your reviewer**: It's just fixes typos
**Release note**: `NONE`
Automatic merge from submit-queue
AWS/GCE: Spread PetSet volume creation across zones, create GCE volumes in non-master zones
Long term we plan on integrating this into the scheduler, but in the
short term we use the volume name to place it onto a zone.
We hash the volume name so we don't bias to the first few zones.
If the volume name "looks like" a PetSet volume name (ending with
-<number>) then we use the number as an offset. In that case we hash
the base name.
Tested with 2000 nodes, this actually meets the GCE API specifications
(which is nutty). Previous PR (#25178) was based on a mistaken
understanding of a poorly documented set of limitations, and even
poorer testing, for which I am embarassed.
Lots of comments describing the heuristics, how it fits together and the
limitations.
In particular, we can't guarantee correct volume placement if the set of
zones is changing between allocating volumes.
Filters can't exceed 4k, and GET requests against the GCE API are also
limited, so these break down in different ways at different cluster
counts. Fix it by introducing an advisory node-instance-prefix
configuration in the GCE provider that can hint the
EnsureLoadBalancer/UpdateLoadBalancer code (and the firewall
creation/update code). If it's not there, or wrong (a hostname that's
registered violates it), just ignore it and grab the whole project.
Automatic merge from submit-queue
refuse to create a firewall rule with no target tag
fixes#25145
This modification in gce.firewallObject() will return error when trying
to create or update firewall rule if no node tag can be found. Also add
unit test for this modification.
We had a long-lasting bug which prevented creation of volumes in
non-master zones, because the cloudprovider in the volume label
admission controller is not initialized with the multizone setting
(issue #27656).
This implements a simple workaround: if the volume is created with the
failure-domain zone label, we look for the volume in that zone. This is
more efficient, avoids introducing a new semantic, and allows users (and
the dynamic provisioner) to create volumes in non-master zones.
Fixes#27657
Long term we plan on integrating this into the scheduler, but in the
short term we use the volume name to place it onto a zone.
We hash the volume name so we don't bias to the first few zones.
If the volume name "looks like" a PetSet volume name (ending with
-<number>) then we use the number as an offset. In that case we hash
the base name.
Fixes#27256
Implements #25145
This modification in gce.firewallObject() will return error when trying
to create or update firewall rule if no node tag can be found. Also add
unit test for this modification.
Instead of just rate limits to operation polling, send all API calls
through a rate limited RoundTripper.
This isn't a perfect solution, since the QPS is obviously getting
split between different controllers, etc., but it's also spread across
different APIs, which, in practice, rate limit differently.
Fixes#26119 (hopefully)
