Previously, once the executor is deleted for reasons (like upgrade or cluster scale-in),
the new executor deletes all existing resources created by the old executor and creates
new one. This mechanism becomes a problem when there are requests connecting to the
existing pods. Also in the worst case, the cluster may not have enough resources to create
new pods and cause service downtime.
This PR let each executor type adopts existing resources before starting the executor
API services, and so the alive connections won't experience failure. However, the requests
send to the function that doesn't have alive function pods will still fail due to the
executor is in bootstrapping.
The pool manager keeps terminating function pod periodically even there are
traffic to the function. The root cause is that executor, poolmgr, newdeploy
manage their own functionServiceCache separately. And when router taps a
function, executor updates the access time of the function service entry in its
own cache without notifying executor types to do the update as well. Hence,
the access time of function service entry in poolmanager cache never gets updated.
Due to the access time never gets updated, the idle pod reaper in poolmanager
then thinks the function pod is in idle state and recycle it.
This PR removes the cache in executor itself, and when router tries to tap a function,
executor will call executor type to tap the function and update access time.
To keep archive creation implementation simple and prevent any
confusion, we decided to remove `--keepurl` flag and embed URL
directly without downloading the file from it . In this way, we can
ensure consistent behavior in either package creation or spec file
creation. Also, it increases the portability of spec file.
1. The records are stored in redis which is not migratable to another cluster for the testing purposes.
2. Some of the requests fields are not recorded.
3. People should consider using https://github.com/buger/goreplay which is an existing mature and well-tested solution for testing purposes.
When `fn test` failed to retrieve logs from the function
pod, it turns to query logs from the log database. However,
the logdb type flag is not set to `fn test`, hence the logdb
type is empty (even without default value) and cause `fn logs`
returns an error due to unable to find the corresponding log
database type.
This PR adds logdb type flag to `fn test` to resolve the problem.
In the case of large files, it takes a long time for the user to download
the source from the URL and upload it to StorgeSvc through CLI.
This PR allows a user to use URL as the function source when creating a function
and provides a new flag "--keeparchiveurl" to let the user to decided
whether the CLI should download the file first or store the file URL in the
archive directly. If "--keeparchiveurl" is true, then no checksum will be
generated, it's the user's responsibility to ensure the file won't be changed.
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.
The router prints error no matter what error type it is.
It's useful for troubleshooting, however, it also prints
the context canceled error, which means that users abort
request before reply and its really normal nowadays. Also,
the router returns 502 if error is not nil and may confused
client if it's a timeout error.
To solve these problems, this PR adds an error handler to
reverse proxy to examine the return error and change the
status code when needed.
For each ingress controller, the format of ingress host,
path and annotations are different. To support different
kinds of controller, this PR adds new ingress config field
to http trigger spec. A user can set annotations, host and
path based on the type of underlying ingress controller
with CLI.
Command example:
fission route create --name foo \
--url /foo/{bar} --function foofn --createingress \
--ingressannotation "nginx.ingress.kubernetes.io/ssl-redirect=false" \
--ingressannotation "nginx.ingress.kubernetes.io/use-regex=true" \
--ingressrule "*=/foo/*"
Currently, only one CI build can be started at the same time due to we only have only one testing cluster. This PR aims to enable multiple concurrent builds can be triggered to reduce the waiting time for CI builds. The major changes listed below:
1. Each build generate unique images for testing and debugging
2. Check whether the testing cluster is being used by another build.
3. Cache docker image & go mod after build finished
4. Improve dockerfiles for reusing docker build cache
* Use terraform to manage ci cluster. Upgrade ci cluster to 1.13
* Upgrade helm version
* Do not break the test if no previous resources
* Do not skip install helm if helm is old version
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
This removes fluentd in favor of using fluentbit, which is lighter (in
memory usage) and seems to be more actively maintained.
Fluentbit's config file format is different from fluentd's. It also
doesn't support the same record modification stuff that fluentd
supports, so we have to change the influxdb query slightly. This
means that after an upgrade, the new CLI may won't work for querying older
logs. Hopefully, this slight breakage is acceptable; if users
really need older logs they can use the older CLI.
The root cause of the issue was introduced by PR https://github.com/fission/fission/pull/1009/files .
To be short, even the CRD of environment was delete, it still takes time for executor (poolmgr) to destroy env pool. In our cases, the previous test creates an env and delete it when test finished, then the next one creates the same name env, but failed to create pool due to the deploy name conflict. So the executor selects the pod from the first created env pool. Then, executor starts to delete the env pool, and makes the pod state became Termination state. To fix this problem, a unique name of deployment will be returned after this PR to prevent the name conflict.
Added support for Kafka headers to be passed and retrieved from functions. The headers and supported and work only for Kafka version 0.11.0.0 and higher.
* Refactor specialization process
* Separate readiness and liveness probe to different routes
For newdeploy, readiness probe should check whether a fetcher specializes env container successfully or not. In this commit, fetcher returns the actual state of current specialization status instead of returning 200ok directly.
