Add support for service profiles created on external (non-service) authorities. For example, this allows you to create a service profile named `linkerd.io` which will apply to calls made to `linkerd.io`.
This is done by changing the `LINKERD2_PROXY_DESTINATION_PROFILE_SUFFIXES` to `.` so that the proxy will attempt to lookup a service profile for any authority. We provide the `--disable-external-profiles` proxy flag to revert this behavior in case it is a problem.
We also refactor the proxy-api implementation of GetProfiles so that it does the profile lookup, regardless of if the authority looks like a Kubernetes service name or not. To simplify this, support for multiple resolves (which was unused) was removed.
Signed-off-by: Alex Leong <alex@buoyant.io>
The proxy-api service _always_ suggests that two meshed pods communicate
via HTTP/2 (i.e. via transparent protocol upgrading, if necessary).
This can complicate debugging and diagnostics at times, so it's
important that we have a way to deploy linkerd without this auto-upgrade
behavior.
This change adds a `-disable-h2-upgrade` flag to the `linkerd install`
command that disables transparent upgrading for the whole cluster.
Filtering by Kubernetes job was not supported. Also filtering by any unknown
type caused a panic.
Add filtering support by Kubernetes job, with special case mapping `job` to
`k8s_job`, to not conflict with Prometheus' job label.
Fix panic when unknown type specified as a `--from` or `--to` flag.
Fix `job` label from `linkerd-proxy` overwriting Prometheus `job` label at
collection time. This caused all metrics collected by proxy sidecars in
Kubernetes jobs to be collected into an incorrect Prometheus job, rather than
the expected `linkerd-proxy` Prometheus job.
Fix `unsupported resource type` tap error message incorrectly printing the
target resource rather than the destination.
Set `--controller-log-level debug` in `install_test.go` for easier debugging.
Expose `slow-cooker`'s metrics via a k8s service in the tap integration test, to
validate proxy requests with a job as destination.
Fixes#1872
Part of #627
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
* Adjust proxy, Prometheus, and Grafana probes
High `readinessProbe.initialDelaySeconds` values delayed the controller's
readiness by up to 30s, preventing cli commands from succeeding shortly after
control plane deployment.
Decrease `readinessProbe.initialDelaySeconds` in the proxy, Prometheus, and
Grafana to the default 0s. Also change `linkerd check` controller pod ordering
to: controller, prometheus, web, grafana.
Detailed probe changes:
- proxy
- decrease `readinessProbe.initialDelaySeconds` from 10s to 0s
- prometheus
- decrease `readinessProbe.initialDelaySeconds` from 30s to 0s
- decrease `readinessProbe.timeoutSeconds` from 30s to 1s
- decrease `livenessProbe.timeoutSeconds` from 30s to 1s
- grafana
- decrease `readinessProbe.initialDelaySeconds` from 30s to 0s
- decrease `readinessProbe.timeoutSeconds` from 30s to 1s
- decrease `readinessProbe.failureThreshold` from 10 to 3
- increase `livenessProbe.initialDelaySeconds` from 0s to 30s
Fixes#1804
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
This change allows some advised production config to be applied to the install of the control plane.
Currently this runs 3x replicas of the controller and adds some pretty sane requests to each of the components + containers of the control plane.
Fixes#1101
Signed-off-by: Ben Lambert <ben@blam.sh>
A container called `proxy-api` runs in the Linkerd2 controller pod. This container listens on port 8086 and serves the proxy-api but does nothing other than forward gRPC requests to the destination container which listens on port 8089.
We remove the proxy-api container altogether and change the destination container to listen on port 8086 instead of 8089. The result is that clients still use the proxy-api by connecting to `proxy-api.<ns>.svc.cluster.local:8086` but the controller has one fewer containers. This results in a simpler system that is easier to reason about.
Signed-off-by: Alex Leong <alex@buoyant.io>
We implement the getProfiles method in the destination service. This method returns a stream of destination profiles for a given authority. It does this by looking up the ServiceProfile resource in the controller namespace named `<svc>.<ns>` where `<svc>` is the name of the service and `<ns>` is the namespace of the service.
This PR includes:
* Adding a ServiceProfile Custom Resource Definition to linkerd install
* A watch based implementation of the getProfiles method in the destination service, similar to the implementation of get.
* An update to the destination client script that allows querying the getProfiles method.
Signed-off-by: Alex Leong <alex@buoyant.io>
* Add --single-namespace install flag for restricted permissions
* Better formatting in install template
* Mark --single-namespace and --proxy-auto-inject as experimental
* Fix wording of --single-namespace check flag
* Small healthcheck refactor
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Support auto sidecar-injection
1. Add proxy-injector deployment spec to cli/install/template.go
2. Inject the Linkerd CA bundle into the MutatingWebhookConfiguration
during the webhook's start-up process.
3. Add a new handler to the CA controller to create a new secret for the
webhook when a new MutatingWebhookConfiguration is created.
4. Declare a config map to store the proxy and proxy-init container
specs used during the auto-inject process.
5. Ignore namespace and pods that are labeled with
linkerd.io/auto-inject: disabled or linkerd.io/auto-inject: completed
6. Add new flag to `linkerd install` to enable/disable proxy
auto-injection
Proposed implementation for #561.
* Resolve missing packages errors
* Move the auto-inject label to the pod level
* PR review items
* Move proxy-injector to its own deployment
* Ignore pods that already have proxy injected
This ensures the webhook doesn't error out due to proxy that are injected using the command
* PR review items on creating/updating the MWC on-start
* Replace API calls to ConfigMap with file reads
* Fixed post-rebase broken tests
* Don't mutate the auto-inject label
Since we started using healhcheck.HasExistingSidecars() to ensure pods with
existing proxies aren't mutated, we don't need to use the auto-inject label as
an indicator.
This resolves a bug which happens with the kubectl run command where the deployment
is also assigned the auto-inject label. The mutation causes the pod auto-inject
label to not match the deployment label, causing kubectl run to fail.
* Tidy up unit tests
* Include proxy resource requests in sidecar config map
* Fixes to broken YAML in CLI install config
The ignore inbound and outbound ports are changed to string type to
avoid broken YAML caused by the string conversion in the uint slice.
Also, parameterized the proxy bind timeout option in template.go.
Renamed the sidecar config map to
'linkerd-proxy-injector-webhook-config'.
Signed-off-by: ihcsim <ihcsim@gmail.com>
Previously, we would tap any resource's pods, regardless of whether the pods
were meshed or not. We can't actually tap non-meshed pods, so I'm adding a check
that will filter out non-meshed pods from the pods that tap watches.
Previous behaviour:
When attempting to hang a non meshed pod, it would establish
a watch on the pods, but then never return any results. In the CLI you could
just cancel it with Ctrl-C. In the web, clicking Stop would send a
WebSocket.close(1000) but wouldn't actually close the connection...
Behaviour after change :
If no pods under the specified resource are meshed, it'll
return an error of no pods being found to tap
`ca-bundle-distributor` described the original role of the program but
`ca` ("Certificate Authority") better describes its current role.
Signed-off-by: Brian Smith <brian@briansmith.org>
* update grafana dashboards to remove conduit reference and replace with linkerd instances
* update test install fixtures to reflect changes
Fixes: #1315
Signed-off-by: Franziska von der Goltz <franziska@vdgoltz.eu>
The control-plane's `ClusterRole` and `ClusterRoleBinding` objects are
global. Because their names did not vary across multiple control-plane
deployments, it prevented multiple control-planes from coexisting (when
RBAC is enabled).
Modify the `ClusterRole` and `ClusterRoleBinding` objects to include the
control-plane's namespace in their names. Also modify the integration
test to first install two control-planes, and then perform its full
suite of tests, to prevent regression.
Fixes#1292.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Create a ephemeral, in-memory TLS certificate authority and integrate it into the certificate distributor.
Remove the re-creation of deleted ConfigMaps; this will be added back later in #1248.
Signed-off-by: Brian Smith brian@briansmith.org
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Add TLS support to `conduit inject`.
Add the settings needed to enable TLs when `--tls=optional` is passed on the
commend line. Later the requirement to add `--tls` will be removed.
Signed-off-by: Brian Smith <brian@briansmith.org>
* Add CA certificate bundle distributor to conduit install
* Update ca-distributor to use shared informers
* Only install CA distributor when --enable-tls flag is set
* Only copy CA bundle into namespaces where inject pods have the same controller
* Update API config to only watch pods and configmaps
* Address review feedback
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Add controller admin servers and readiness probes
* Tweak readiness probes to be more sane
* Refactor based on review feedback
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
The `kubernetes-nodes-cadvisor` Prometheus queries node-level data via
the Kubernetes API server. In some configurations of Kubernetes, namely
minikube and at least one baremetal kubespray cluster, this API call
requires the `get` verb on the `nodes/proxy` resource.
Enable `get` for `nodes/proxy` for the `conduit-prometheus` service
account.
Fixes#912
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
* Add readiness/liveness checks for third party components
Any possible issues with the third party control plane components can wedge the services.
Take the best practices for prometheus/grafana and add them to our template. See #1116
* Update test fixtures for new output
Running `conduit install --api-port xxx` where xxx != 8086 would yield a
broken install.
Fix the install command to correctly propagate the `api-port` flag,
setting it as the serve address in the proxy-api container.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Grafana provides default dashboards for Prometheus and Grafana health.
The community also provides Kubernetes-specific dashboards. Conduit was
not taking advantage of these.
Introduce new Grafana dashboards focused on Grafana, Kubernetes, and
Prometheus health. Tag all Conduit dashboards for easier UI navigation.
Also fix layout in Conduit Health dashboard.
Part of #420
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
* Add namespace as a resource type in public-api
The cli and public-api only supported deployments as a resource type.
This change adds support for namespace as a resource type in the cli and
public-api. This also change includes:
- cli statsummary now prints `-`'s when objects are not in the mesh
- cli statsummary prints `No resources found.` when applicable
- removed `out-` from cli statsummary flags, and analagous proto changes
- switched public-api to use native prometheus label types
- misc error handling and logging fixes
Part of #627
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
* Refactor filter and groupby label formulation
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Rename stat_summary.go to stat.go in cli
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Update rbac privileges for namespace stats
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Remove the telemetry service
The telemetry service is no longer needed, now that prometheus scrapes
metrics directly from proxies, and the public-api talks directly to
prometheus. In this branch I'm removing the service itself as well as
all of the telemetry protobuf, and updating the conduit install command
to no longer install the service. I'm also removing the old version of
the stat command, which required the telemetry service, and renaming the
statsummary command to stat.
* Fix time window tests
* Remove deprecated controller scrape config
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
Start implementing new conduit stat summary endpoint.
Changes the public-api to call prometheus directly instead of the
telemetry service. Wired through to `api/stat` on the web server,
as well as `conduit statsummary` on the CLI. Works for deployments only.
Current implementation just retrieves requests and mesh/total pod count
(so latency stats are always 0).
Uses API defined in #663
Example queries the stat endpoint will eventually satisfy in #627
This branch includes commits from @klingerf
* run ./bin/dep ensure
* run ./bin/update-go-deps-shas
The Destination service used slightly different labels than the
telemetry pipeline expected, specifically, prefixed with `k8s_*`.
Make all Prometheus labels consistent by dropping `k8s_*`. Also rename
`pod_name` to `pod` for consistency with `deployement`, etc. Also update
and reorganize `proxy-metrics.md` to reflect new labelling.
Fixes#655
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Using a vanilla Grafana Docker image as part of `conduit install`
avoided maintaining a conduit-specific Grafana Docker image, but made
packaging dashboard json files cumbersome.
Roll our own Grafana Docker image, that includes conduit-specific
dashboard json files. This significantly decreases the `conduit install`
output size, and enables dashboard integration in the docker-compose
environment.
Fixes#567
Part of #420
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
The Grafana dashboards were displaying all proxy-enabled pods, including
conduit controller pods. In the old telemetry pipeline filtering these
out required knowledge of the controller's namespace, which the
dashboards are agnostic to.
This change leverages the new `conduit_io_control_plane_component`
prometheus label to filter out proxy-enabled controller components.
Part of #420
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
The Top-line and Deployment Grafana dashboards relied on the
soon-to-be-removed telemetry pipeline metrics.
Update the Grafana dashboards to query for the new, proxy-based metrics.
Grafana dashboard layouts have not changed.
Depends on #635 to render metrics.
Part of #420.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Previously we were using the instance label to uniquely identify a pod.
This meant that getting stats by pod name would require extra queries to
Kubernetes to map pod name to instance.
This change adds a pod_name label to metrics at collection time. This
should not affect cardinality as pod_name is invariant with respect to
instance.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
The Prometheus scrape config collects from Conduit proxies, and maps
Kubernetes labels to Prometheus labels, appending "k8s_".
This change keeps the resultant Prometheus labels consistent with their
source Kubernetes labels. For example: "deployment" and
"pod_template_hash".
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
The inject code detects the object it is being injected into, and writes
self-identifying information into the CONDUIT_PROMETHEUS_LABELS
environment variable, so that conduit-proxy may read this information
and report it to Prometheus at collection time.
This change puts the self-identifying information directly into
Kubernetes labels, which Prometheus already collects, removing the need
for conduit-proxy to be aware of this information. The resulting label
in Prometheus is recorded in the form `k8s_deployment`.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
In addition to dashboards display service health, we need a dashboard to
display health of the Conduit service mesh itself.
This change introduces a conduit-health dashboard. It currently only
displays health metrics for the control plane components. Proxy health
will come later.
Fixes#502
Part of #420
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
The existing telemetry pipeline relies on Prometheus scraping the
Telemetry service, which will soon be removed.
This change configures Prometheus to scrape the conduit proxies directly
for telemetry data, and the control plane components for control-plane
health information. This affects the output of both conduit install
and conduit inject.
Fixes#428, #501
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
The existing `conduit dashboard` command supported opening the conduit
dashboard, or displaying the conduit dashboard URL, via a `url` boolean
flag.
Replace the `url` boolean flag with a `show` string flag, with three
modes:
`conduit dashboard --show conduit`: default, open conduit dashboard
`conduit dashboard --show grafana`: open grafana dashboard
`conduit dashboard --show url`: display dashboard URLs
Part of #420
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Existing Grafana configuration contained no dashboards, just a skeleton
for testing.
Introduce two Grafana dashboards:
1) Top Line: Overall health of all Conduit-enabled services
2) Deployment: Health of a specific conduit-enabled deployment
Fixes#500
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
* Most controller listeners should only bind on localhost
* Use default listening addresses in controller components
* Review feedback
* Revert test_helper change
* Revert use of absolute domains
Signed-off-by: Alex Leong <alex@buoyant.io>
Alex Leong
Oliver Gould
Kevin Lingerfelt
Andrew Seigner
Ben Lambert
Ivan Sim
Risha Mars
Brian Smith
Franziska von der Goltz
Thomas Rampelberg