- Added metav1.Status() that enforces '406 Not Acceptable' response if
protobuf serialization is not fully supported for the API resource type.
- JSON and YAML serialization are supposed to be more completely baked
in, so serialization involving those, and general errors with seralizing
protobuf, will return '500 Internal Server Error'.
- If serialization failure occurs and original HTTP status code is
error, use the original status code, else use the serialization failure
status code.
- Write encoded API responses to intermediate buffer
- Use apimachinery/runtime::Encode() instead of
apimachinery/runtime/protocol::Encode() in
apiserver/endpoints/handlers/responsewriters/writers::SerializeObject()
- This allows for intended encoder error handling to fully work, facilitated by
apiserver/endpoints/handlers/responsewriters/status::ErrorToAPIResponse() before officially
writing to the http.ResponseWriter
- The specific part that wasn't working by ErrorToAPIResponse() was the
HTTP status code set. A direct call to
http.ResponseWriter::WriteHeader(statusCode) was made in
SerializeObject() with the original response status code, before
performing the encode. Once this
method is called, it can not again update the status code at a later
time, with say, an erro status code due to encode failure.
- Updated relevant apiserver unit test to reflect the new behavior
(TestWriteJSONDecodeError())
- Add build deps from make update for protobuf serializer
50342: Code review suggestion impl
- Ensure that http.ResponseWriter::Header().Set() is called before http.ResponseWriter::WriteHeader()
- This will avert a potential issue where changing the response media type to text/plain wouldn't work.
- We want to respond with plain text if serialization fails of the original response, and serialization also fails for the resultant error response.
50342: wrapper for http.ResponseWriter
- Prevent potential performance regression caused by modifying encode to use a buffer instead of streaming
- This is achieved by creating a wrapper type for http.ResponseWriter that will use WriteHeader(statusCode) on the first
call to Write(). Thus, on encode success, Write() will write the original statusCode. On encode failure, we pass control
onto responsewriters::errSerializationFatal(), which will process the error to obtain potentially a new status code, depending
on whether or not the original status code was itself an error.
50342: code review suggestions
- Remove historical note from unit test comment
- Don't export httpResponseWriterWithInit type (for now)
Kubernetes-commit: bcdf3bb64333ce12f15b1beebef48f554d69027f
This ensures that request cancellation will be propagated properly to
the client used to create the stream. Without this fix, the apiserver
and the kubelet may leak resources (e.g., goroutine, inotify watches).
One such example is that if user run `kubectl logs -f <container that
don't produce new logs)` and then enter ctrl-c, both kubelet and
apiserver will hold on to the connection and resources indefinitely.
Kubernetes-commit: 31d1607a514b62ef46452e402f5438d827314b98
Make apiserver pass connectRequest.Options directly to the admission layer. All
the information in rest.ConnectRequest is present in admission attributes.
Kubernetes-commit: 355691d310803ea3a0cd8ff284a39ead38857602
This makes the error consistent with the timeout filter and also helps
the user understand that they requested a specific timeout.
Kubernetes-commit: 8a2d037bc51c97758c0a68f2726f104953846cd5
remove create-on-update logic for quota controller
review: add more error check
remove unused args
revert changes in patch.go
use hasUID to judge if it's a create-on-update
Kubernetes-commit: ccb1ec7a3695082326fe60ec06890f91004dc043
Got
```
E0628 00:23:07.106285 1 watch.go:274] unable to encode watch object: expected pointer, but got invalid kind
```
on a production system and had no way to debug what type was being sent.
Kubernetes-commit: 307849baef076d8ee61a3b9649f9260a765f7ac0
builds on #62868
1. When the incoming patch specified a resourceVersion that failed as a precondition,
the patch handler would retry uselessly 5 times. This PR collapses onto GuaranteedUpdate,
which immediately stops retrying in that case.
2. When the incoming patch did not specify a resourceVersion, and persisting to etcd
contended with other etcd updates, the retry would try to detect patch conflicts with
deltas from the first 'current object' retrieved from etcd and fail with a conflict error
in that case. Given that the user did not provide any information about the starting version
they expected their patch to apply to, this does not make sense, and results in arbitrary
conflict errors, depending on when the patch was submitted relative to other changes made
to the resource. This PR changes the patch application to be performed on the object retrieved
from etcd identically on every attempt.
fixes#58017
SMP is no longer computed for CRD objects
fixes#42644
No special state is retained on the first attempt, so the patch handler correctly handles
the cached storage optimistically trying with a cached object first
Kubernetes-commit: fbd6f3808480d27a83643e82a11c217601b76cbc
This is the combination of a series of changes which individually don't
make any behavioral changes. The original commits are preserved in my
own fork in the refactor-patch-complete branch, as when squashed this is
impossible to review.
This turned a big function with lots of parameters and closures into an
object with multiple functions, fewer closures and more well documented
state transitions.
Kubernetes-commit: 349a99b80e7e6c0c92218c814ae0858fd71609fc
Since we have a custom handler for apiextensions-apiserver,
we need to record the metrics here.
Kubernetes-commit: 74cd45fb21b349dd037e3bfd844459ca5834cca1
Jordan Liggitt
Cao Shufeng
jennybuckley
Antoine Pelisse
Tim Allclair
Tristan Burgess
Yu-Ju Hong
Jordan Liggitt
Chao Xu
Clayton Coleman
yue9944882
xuzhonghu
Mikhail Mazurskiy
Guoliang Wang
wojtekt
Dr. Stefan Schimanski
hangaoshuai
Daniel Smith
Mike Danese
Mengqi Yu
Kubernetes Publisher
Jeff Grafton
Nikhita Raghunath
halfcrazy