Jaeger preserves the trace data of router healthz endpoints
which is not so helpful for monitoring function metrics and
consumes storage to store such trace data.
This PR uses `GetStartOptions` to examine the request's path and
check if it's a request to healthz endpoint. If yes, skip and not
to trace it.
To follow 12-factor app rules, make jaeger collector endpoint
as an environment variable instead of CLI args. It's easier to
replace the value in different deployments. Also, we can utilize
valueFrom to get value from the configmap.
Helm treats the single line "extraCoreComponentPodConfig" in
values.yaml as a key value pair config.
```
extraCoreComponentPodConfig:
```
And when a user tries to install helm with providing config
like following, it failed.
```
extraCoreComponentPodConfig:
nodeSelector:
serverless: fission
```
The root cause is that helm treats config in default values.yaml
and providing config as two different types of config, which is
key-value vs. value table. And so it's unable for helm to merge these
two configs even with the same key name.
This PR comments out the single line config in default values.yaml
to prevent the issue.
Encoded path means the URL path contains encoded string like "/foo%2Fbar".
Gorilla/Mux by default doesn't enable encoded path support, you need to enable
it when initializing the router.
This PR adds a new environment variable USE_ENCODED_PATH to the router
deployment to enable encoded path support.
If a user deploys fission in the namespace which is different from the one in the single YAML file generated by helm, fission components won't be able to talk to each other due to the wrong namespace appends after the service address.
This PR adds --namespace when generating the YAML file to prevent the mismatch problem.
ReplicaSet generated by Deployment distributes pods to nodes based
on nodes resource usage, which means in some cases the newly created
pods may be assigned to only a few nodes. The requests will go to the
same node if there multiple pods on it and may increase the node
resource usage and overall request latency.
To solve the problem mentioned above, this PR allows users to deploy
the router as DaemonSet so that we can distribute requests across all
nodes for better workload distribution and lower latency.
In 1.9, kubernetes prompted Deployment to apps/v1 and finally deprecated
extensions/v1beta1 support in 1.16. Users will see the following error message
in log when fission components try to submit Deployment to the k8s API server
and specify extensions/v1beta1 as API group.
error: unable to recognize "deployment": no matches for kind "Deployment" in version "extensions/v1beta1"
This PR updates call to k8s API server and apiKind in chart files to avoid fission failure in 1.16.
This fix allows the empty repository tag in values.yaml file. The changes are done in the templates of deployment and fluentbit where repository tag will be added in the docker image name only when specified with some value in case of empty repository tag image name contain the name of image and tag only.
* Add support for setting nodeSelector/affinity/tolerations on all deployments of fission components.
* Add extraCoreComponmentPodConfig helm value to allow setting of affinty/tolerations/nodeSelectors/etc on core pods
Consolidates the addition of the fetcher container to the pod into a new type FetcherConfig and takes care of serviceAccountName if not set. Also added PreStop lifecycle handler if pod set TerminationGracePeriodSeconds
As of K8s 1.8, RBAC mode is stable and backed by the rbac.authorization.k8s.io/v1
API. Using v1beta1 API version for RBAC based objects fails helm deployment
of fission on minikube
* Optimize router response time by adding update lock
In 0.9.2, the router sends multiple GetServiceForFunction requests to executor
to get the service URL. However, the response time of executor will increase
under high-concurrency situation due to too many requests are waiting for
processing.
To solve the problem, an update lock was added to the router. All of goroutines belongs
to the same function need to grab the update lock before sending the request.
Only the first goroutine which gets the update lock is allowed to send request.
In this way, we reduce the burden of executor and lower the failure rate.
