The ExternalWorkload resource we introduced has a minor naming
inconsistency; `Tls` in `meshTls` is not capitalised. Other resources
that we have (e.g. authentication resources) capitalise TLS (and so does
Go, it follows a similar naming convention).
We fix this in the workload resource by changing the field's name and
bumping the version to `v1beta1`.
Upgrading the control plane version will continue to work without
downtime. However, if an existing resource exists, the policy controller
will not completely initialise. It will not enter a crashloop backoff,
but it will also not become ready until the resource is edited or
deleted.
Signed-off-by: Matei David <matei@buoyant.io>
If the readiness of an external workload endpoint changes while traffic
is being sent to it, the update will not be propagated to clients. This
can lead to issues where an endpoint that is marked as `notReady`
continues to figure out as being `ready` by the endpoints watcher.
The issue stems from how endpoint slices are diffed. A utility function
responsible for processing addresses does not consider endpoints whose
targetRef is an external workload. We fix the problem and add two module
tests to validate readiness is propagated to clients correctly.
---------
Signed-off-by: Matei David <matei@buoyant.io>
Adds support for remote discovery to the destination controller.
When the destination controller gets a `Get` request for a Service with the `multicluster.linkerd.io/remote-discovery` label, this is an indication that the destination controller should discover the endpoints for this service from a remote cluster. The destination controller will look for a remote cluster which has been linked to it (using the `linkerd multicluster link` command) with that name. It will look at the `multicluster.linkerd.io/remote-discovery` label for the service name to look up in that cluster. It then streams back the endpoint data for that remote service.
Since we now have multiple client-go informers for the same resource types (one for the local cluster and one for each linked remote cluster) we add a `cluster` label onto the prometheus metrics for the informers and EndpointWatchers to ensure that each of these components' metrics are correctly tracked and don't overwrite each other.
---------
Signed-off-by: Alex Leong <alex@buoyant.io>
Add support for enabling and disabling topology aware routing when hints are added/removed.
The testing setup is very involved because it involves so many moving parts
1) Setup a service which is layered over several availability zones.
1a) The best way to do this is one service object, with 3 replicasets explicitly forced to use a specific AZ each.
2) Add `service.kubernetes.io/topology-aware-hints: Auto` annotation to the Service object
3) Use a load tester like k6 to send meaningful traffic to your service but only in one AZ
3) Scale up your replica sets until k8s adds Hints to your endpointslices
4) Observe that traffic shifts to only hit pods in one AZ
5) Turn down the replicasets count until such time that K8s removes the hints from your endpointslices
6) Observe traffic shifts back to all pods across all AZ.
When processing a `delete` event for an EndpointSlice, regardless of the outcome (whether there are still addresses/endpoints alive or whether we have no endpoints left) we make a call to `noEndpoints` on the subscriber. This will send a `Destination.Get` update to the listeners to advertise a `NoEndpoints` event.
If there are still addresses left, NoEndpoints { exists: true } will be sent (to signal the service still exists). Until an update is registered (i.e a new add) there will be no available endpoints -- this is incorrect, since other EndpointSlices may exist. This change fixes the problem by handling `noEndpoints` in a more specialized way for EndpointSlices.
Signed-off-by: Yannick Utard <yannickutard@gmail.com>
Fixes#9986
After reviewing the k8s API calls in Destination, it was concluded we
could only swap out the calls to the Node and RS resources to use the
metadata API, as all the other resources (Endpoints, EndpointSlices,
Services, Pod, ServiceProfiles, Server) required fields other than those
found in their metadata section.
This also required completing the `NewFakeAPI` implementation by adding
the missing annotations and labels entries.
## Testing Memory Consumption
The gains here aren't as big as in #9650. In order to test this we need
to push hard and create 4000 RS:
``` bash
for i in {0..4000}; do kubectl create deployment test-pod-$i --image=nginx; done
```
In edge-23.2.1 the destination pod's memory consumption goes from 40Mi
to 160Mi after all the RS were created. With this change, it went from
37Mi to 140Mi.
Closes#10130https://kubernetes.io/docs/concepts/services-networking/service-traffic-policy/
1. Update endpoints watcher to include additional field `localTrafficPolicy`.
Set to true when `.spec.internalTrafficPolicy` is set to `Local`
2. Update endpoints translater to filter by node when `localTrafficPolicy` is
set to true. Topology Aware Hints are not used when
`service.pec.internalTrafficPolicy` is set to local
Signed-off-by: Yu Cao <yc185050@ncr.com>
* build(deps): bump sigs.k8s.io/gateway-api from 0.5.1 to 0.6.0
Bumps [sigs.k8s.io/gateway-api](https://github.com/kubernetes-sigs/gateway-api) from 0.5.1 to 0.6.0.
- [Release notes](https://github.com/kubernetes-sigs/gateway-api/releases)
- [Changelog](https://github.com/kubernetes-sigs/gateway-api/blob/main/CHANGELOG.md)
- [Commits](https://github.com/kubernetes-sigs/gateway-api/compare/v0.5.1...v0.6.0)
---
updated-dependencies:
- dependency-name: sigs.k8s.io/gateway-api
dependency-type: direct:production
update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
* Account for possible errors returned from `AddEventHandler`
In v0.26.0 client-go's `AddEventHandler` method for informers started
returning a registration handle (that we ignore) and an error that we
now surface up.
* client-go v0.26.0 removed the openstack plugin
* Temporary changes to trigger tests in k8s 1.21
- Adds an innocuous change to integration.yml so that all tests get
triggered
- Hard-code k8s version in `k3d cluster create` invocation to v1.21
* Revert "Temporary changes to trigger tests in k8s 1.21"
This reverts commit 3e1fdd0e5e.
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Alejandro Pedraza <alejandro@buoyant.io>
Fixes#10003
When endpoints are removed from an EndpointSlice resource, the destination controller builds a list of addresses to remove. However, if any of the removed endpoints have a Pod as their targetRef, we will attempt to fetch that pod to build the address to remove. If that pod has already been removed from the informer cache, this will fail and the endpoint will be skipped in the list of endpoints to be removed. This results in stale endpoints being stuck in the address set and never being removed.
We update the endpoint watcher to construct only a list of endpoint IDs for endpoints to remove, rather than fetching the entire pod object. Since we no longer attempt to fetch the pod, this operation is now infallible and endpoints will no longer be skipped during removal.
We also add a `TestEndpointSliceScaleDown` test to exercise this.
Signed-off-by: Alex Leong <alex@buoyant.io>
Fixes#8592
Increase the minimum supported kubernetes version from 1.20 to 1.21. This allows us to drop support for batch/v1beta1/CronJob and discovery/v1beta1/EndpointSlices, instead using only v1 of those resources. This fixes deprecation warnings about these warnings printed by the CLI.
Signed-off-by: Alex Leong <alex@buoyant.io>
When the `mirror.linkerd.io/remote-gateway-identity` and `mirror.linkerd.io/remote-svc-fq-name` annotations are set on an EndpointSlice object, the destination controller does not return the correct identity hints for that endpoint.
We fix an incorrect assignment to fix this. We also fix some logic that can result in a nil pointer dereference instead of comparing empty strings. We add a test case to exercise these.
Signed-off-by: Alex Leong <alex@buoyant.io>
# Problem
While rolling out often not all pods will be ready in all the same set of
ports, leading the Kubernetes Endpoints API to return multiple subsets,
each covering a different set of ports, with the end result that the
same address gets repeated across subsets.
The old code for endpointsToAddresses would loop through all subsets, and the
later occurrences of an address would overwrite previous ones, with the
last one prevailing.
If the last subset happened to be for an irrelevant port, and the port to
be resolved is named, resolveTargetPort would resolve to port 0, which would
return port 0 to clients, ultimately leading linkerd-proxy to forward
connections to port 0.
This only happens if the pods selected by a service expose > 1 port, the
service maps to > 1 of these ports, and at least one of these ports is named.
# Solution
Never write an address to set of addresses if resolved port is 0, which
indicates named port resolution failed.
# Validation
Added a test case.
Signed-off-by: Riccardo Freixo <riccardofreixo@gmail.com>
This reverts commit f9ab867cbc which renamed the
multicluster label name from `mirror.linkerd.io` to `multicluster.linkerd.io`.
While this change was made to follow similar namings in other extensions, it
complicates the multicluster upgrade process due to the secret creation.
`mirror.linkerd.io` is not that important of a label to change and this will
allow a smoother upgrade process for `stable-2.10.x`
Signed-off-by: Kevin Leimkuhler <kevin@kleimkuhler.com>
This renames the multicluster annotation prefix from `mirror.linkerd.io` to
`multicluster.linkerd.io` in order to reflect other extension naming patterns.
Additionally, it moves labels only used in the Multicluster extension into their
own labels file—again to reflect other extensions.
Signed-off-by: Kevin Leimkuhler <kevin@kleimkuhler.com>
This PR corrects misspellings identified by the [check-spelling action](https://github.com/marketplace/actions/check-spelling).
The misspellings have been reported at aaf440489e (commitcomment-41423663)
The action reports that the changes in this PR would make it happy: 5b82c6c5ca
Note: this PR does not include the action. If you're interested in running a spell check on every PR and push, that can be offered separately.
Signed-off-by: Josh Soref <jsoref@users.noreply.github.com>
EndpointSlices have been made opt-in due to their experimental nature. This PR
introduces a new install flag 'enableEndpointSlices' that will allow adopters to
specify in their cli install or helm install step whether they would like to
use endpointslices as a resource in the destination service, instead of the
endpoints k8s resource.
Signed-off-by: Matei David <matei.david.35@gmail.com>
Introduce support for the EndpointSlice k8s resource (k8s v1.16+) in the destination service.
Through this PR, in the EndpointsWatcher, there will be a dedicated informer for EndpointSlice;
the informer cannot run at the same time as the Endpoints resource informer. The main difference
is that EndpointSlices have a one-to-many relationship with a service, they provide better performance benefits,
dual-stack addresses and more. EndpointSlice support also implies service topology and other k8s related features.
Validated and tested manually, as well as with dedicated unit tests.
Closes#4501
Signed-off-by: Matei David <matei.david.35@gmail.com>
This change removes the target port requirement when resolving ports in the dst service. Based on the comments, it seems that we need to have a target port defined in the port spec in order to resolve to the port in the Endpoints. In reality if target port is note defined when creating the service, k8s will set the port and the target port to the same value. Seems to me that checking for the targetPort to be different than 0, is a no-op.
Signed-off-by: Zahari Dichev zaharidichev@gmail.com
This PR introduces a service mirroring component that is responsible for watching remote clusters and mirroring their services locally.
Signed-off-by: Zahari Dichev <zaharidichev@gmail.com>
Fixes#3444Fixes#3443
## Background and Behavior
This change adds support for the destination service to resolve Get requests which contain a service clusterIP or pod ip as the `Path` parameter. It returns the stream of endpoints, just as if `Get` had been called with the service's authority. This lays the groundwork for allowing the proxy to TLS TCP connections by allowing the proxy to do destination lookups for the SO_ORIG_DST of tcp connections. When that ip address corresponds to a service cluster ip or pod ip, the destination service will return the endpoints stream, including the pod metadata required to establish identity.
Prior to this change, attempting to look up an ip address in the destination service would result in a `InvalidArgument` error.
Updating the `GetProfile` method to support ip address lookups is out of scope and attempts to look up an ip address with the `GetProfile` method will result in `InvalidArgument`.
## Implementation
We do this by creating a `IPWatcher` which wraps the `EndpointsWatcher` and supports lookups by ip. `IPWatcher` maintains a mapping up clusterIPs to service ids and translates subscriptions to an IP address into a subscription to the service id using the underlying `EndpointsWatcher`.
Since the service name is no longer always infer-able directly from the input parameters, we restructure `EndpointTranslator` and `PodSet` so that we propagate the service name from the endpoints API response.
## Testing
This can be tested by running the destination service locally, using the current kube context to connect to a Kubernetes cluster:
```
go run controller/cmd/main.go destination -kubeconfig ~/.kube/config
```
Then lookups can be issued using the destination client:
```
go run controller/script/destination-client/main.go -path 192.168.54.78:80 -method get -addr localhost:8086
```
Service cluster ips and pod ips can be used as the `path` argument.
Signed-off-by: Alex Leong <alex@buoyant.io>
* The linkerd proxy does not work with headless services (i.e. endpoints not referencing a pod).
Changed endpoints_watcher to also return endpoints with no targetref.
Fixes#3308
Signed-off-by: Johannes Hansen <johannesh1980@gmail.com>
* Fix panic in endpoint_translator
Signed-off-by: Johannes Hansen <johannesh1980@gmail.com>
Fixes https://github.com/linkerd/linkerd2/issues/2800#issuecomment-513740498
When the Linkerd proxy sends a query for a Kubernetes external name service to the destination service, the destination service returns `NoEndpoints: exists=false` because an external name service has no endpoints resource. Due to a change in the proxy's fallback logic, this no longer causes the proxy to fallback to either DNS or SO_ORIG_DST and instead fails the request. The net effect is that Linkerd fails all requests to external name services.
We change the destination service to instead return `InvalidArgument` for external name services. This causes the proxy to fallback to SO_ORIG_DST instead of failing the request.
Signed-off-by: Alex Leong <alex@buoyant.io>
The destination service's endpoints watcher assumed every `Endpoints`
object contained a `TargetRef`. This field is optional, and in cases
such as the default `ep/kubernetes` object, `TargetRef` is nil, causing
a nil pointer dereference.
Fix endpoints watcher to check for `TargetRef` prior to dereferencing.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
We add support for looking up individual pods in a stateful set with the destination service. This allows Linkerd to correctly proxy requests which address individual pods. The authority structure for such a request is `<pod-name>.<service>.<namespace>.svc.cluster.local:<port>`.
Fixes#2266
Signed-off-by: Alex Leong <alex@buoyant.io>
This change implements the DstOverrides feature of the destination profile API (aka traffic splitting).
We add a TrafficSplitWatcher to the destination service which watches for TrafficSplit resources and notifies subscribers about TrafficSplits for services that they are subscribed to. A new TrafficSplitAdaptor then merges the TrafficSplit logic into the DstOverrides field of the destination profile.
Signed-off-by: Alex Leong <alex@buoyant.io>
This is a major refactor of the destination service. The goals of this refactor are to simplify the code for improved maintainability. In particular:
* Remove the "resolver" interfaces. These were a holdover from when our decision tree was more complex about how to handle different kinds of authorities. The current implementation only accepts fully qualified kubernetes service names and thus this was an unnecessary level of indirection.
* Moved the endpoints and profile watchers into their own package for a more clear separation of concerns. These watchers deal only in Kubernetes primitives and are agnostic to how they are used. This allows a cleaner layering when we use them from our gRPC service.
* Renamed the "listener" types to "translator" to make it more clear that the function of these structs is to translate kubernetes updates from the watcher to gRPC messages.
Signed-off-by: Alex Leong <alex@buoyant.io>
Matei David
Zahari Dichev
Alex Leong
Mark Robinson
Yannick Utard
Alejandro Pedraza
Yu Cao
dependabot[bot]
Riccardo Freixo
Kevin Leimkuhler
Josh Soref
Matei David
Johannes Hansen
Andrew Seigner