This raises awareness so that developers hopefully get it right without having to
learn about it from compile errors. It also explains that creating a PR to
fix the naming is not desired.
Kubernetes-commit: f40892f4813f29fc305d59073d2464a1fd7d6fe5
- Add a new VeryShortWatchError struct for error matching, returned
by `handleAnyWatch`, up through `Reflector.ListAndWatch`.
- Update test expectations to match exact errors.
Kubernetes-commit: 6eff9db0f10db72f2c64390e106a80621d136439
Change:
- refactor Reflector.ListAndWatch and Reflector.watch to always call
watcher.Stop.
- refactor Reflector.handleAnyWatch to always call watcher.Stop,
EXCEPT when exitOnWatchListBookmarkReceived &&
watchListBookmarkReceived.
- Update unit tests with these new expectations.
Effect:
- ensures watcher.Stop is always called at least once
- avoids deadlocks in watcher implementations when watcher.Watch is
called, but watcher.Stop is never called.
Note: It's impossible to guarantee that Stop will only be called once.
So watch.Interface implementations must tollerate Stop being called
multiple times.
Kubernetes-commit: 3e609ecf6e945bf4562bddfc563fde9a4c3d0d90
The Lister and Watcher interfaces only supported methods without context, but
were typically implemented with client-go API calls which need a context. New
interfaces get added using the same approach as in
https://github.com/kubernetes/kubernetes/pull/129109.
Kubernetes-commit: 6688adae142e37114d9dfa8d94cd1d8a91fbcc13
cache.ErrRequeue advertised itself as a way to requeue failures on a
FIFO, but it suffers the same problems as AddIfNotPresent. If we do
requeue an item at the end, we'll move the informer back in time. If we
requeue at the beginning we'll simply wedge FIFO.
We didn't find examples in the wild, but by removing the error type
those impacted will get a compile error and get to decide what action is
most appropriate for their failure. Most of the time, proceeding to the
next item is best.
Kubernetes-commit: 238c32a1d9b2c72d648193fa8642a53a2884975f
Logically a cache.Queue.AddIfNotPresent means that the informer can move
back in time since an older item is placed after newer items. The
alternative of placing errors at the head of the queue leads to
indefinite memory growth and repeated failures on retry.
Luckily this behavior was behind RetryOnError, which was always set to
false and impossible for normal users to set to true. By removing the
function and setting, impacted users (none found in a github search)
will get a compile failure.
Kubernetes-commit: 8e77ac000131019d5aa49c19aa1f477f6dac4d59
The test relied on a 100ms sleep to ensure that controller was done. If that
race was lost, one goroutine was intentionally prevented from completing by
locking a mutex permanently. A TODO was left about detecting that.
Adding goroutine leak checking in
https://github.com/kubernetes/kubernetes/pull/126387 revealed that this race
indeed sometimes is lost because the goroutine
leaked (https://github.com/kubernetes/kubernetes/issues/129400).
Waiting for controller shutdown instead of relying on timing should fix this.
Kubernetes-commit: 8e1403563a60f3b7a258e3bbb64b5c3a7f6548fb
The context is used for cancellation and to support contextual logging.
In most cases, alternative *WithContext APIs get added, except for
NewIntegerResourceVersionMutationCache where code searches indicate that the
API is not used downstream.
An API break around SharedInformer couldn't be avoided because the
alternative (keeping the interface unchanged and adding a second one with
the new method) would have been worse. controller-runtime needs to be updated
because it implements that interface in a test package. Downstream consumers of
controller-runtime will work unless they use those test package.
Converting Kubernetes to use the other new alternatives will follow. In the
meantime, usage of the new alternatives cannot be enforced via logcheck
yet (see https://github.com/kubernetes/kubernetes/issues/126379 for the
process).
Passing context through and checking it for cancellation is tricky for event
handlers. A better approach is to map the context cancellation to the normal
removal of an event handler via a helper goroutine. Thanks to the new
HandleErrorWithLogr and HandleCrashWithLogr, remembering the logger is
sufficient for handling problems at runtime.
Kubernetes-commit: 4638ba971661497b147906b8977ae206c9dd6e44
Several tests leaked goroutines. All of those get fixed where possible
without API changes. Goleak is used to prevent regressions.
One new test specifically covers shutdown of an informer and its event
handlers.
Kubernetes-commit: 0ba43734b4c8998b4aaeb1fa2bec8dee609fa50a
The "// import <path>" comment has been superseded by Go modules.
We don't have to remove them, but doing so has some advantages:
- They are used inconsistently, which is confusing.
- We can then also remove the (currently broken) hack/update-vanity-imports.sh.
- Last but not least, it would be a first step towards avoiding the k8s.io domain.
This commit was generated with
sed -i -e 's;^package \(.*\) // import.*;package \1;' $(git grep -l '^package.*// import' | grep -v 'vendor/')
Everything was included, except for
package labels // import k8s.io/kubernetes/pkg/util/labels
because that package is marked as "read-only".
Kubernetes-commit: 8a908e0c0bd96a3455edf7e3b5f5af90564e65b0
I am moving myself to emeritus as I am now firmly on the end-user side
of things.
Signed-off-by: Andy Goldstein <andy.goldstein@gmail.com>
Kubernetes-commit: 3ab816dcabf37acda33c665ab5aa85f1f6163bc1
The watch.Interface design is hard to change, because it would break
most client-go users that perform watches. So instead of changing the
interface to be more user friendly, this change updates the method
comments to explain the different responsibilities of the consumer
(client user) and the producer (interface implementer).
Kubernetes-commit: 1f35231a1d4f7b8586a7ec589c799729eeb4f7c4
- Extract watchWithResync to simplify ListAndWatch
- Wrap watchHandler with two variants, one for WatchList and one for
just Watch.
- Replace a bool pointer arg with a bool arg and bool return, to
improve readability.
- Use errors.Is to satisfy the linter
- Use %w to wrap the store.Replace error, to allow unwrapping.
Kubernetes-commit: 65fc1bb463c85a4c85e619bf7acac9503e23a253
- Add tests to confirm that Stop is always called.
- Add TODOs to show were Stop is not currently being called
(to fix in a future PR)
Kubernetes-commit: ab5aa4762fd5206d0dbd8412d7c6f3b76533a122
the signature of the method was tightly connected to the reflector,
making it difficult to use for anything other than a reflector.
this simple refactor makes the method more generic.
Kubernetes-commit: 83c7542abc8c542c01ecb67376f134b2071c5304
In particular, document that ListAllByNamespace delegates to ListAll
if no namespace is specified.
Signed-off-by: Stephen Kitt <skitt@redhat.com>
Kubernetes-commit: 54e899317ef46e3b70827cacee244717022db0ad
until #115478(use streaming against the etcd storage)
is resolved the cacher need a way to disable the streaming.
Kubernetes-commit: 41e706600aea7468f486150d951d3b8948ce89d5
checkWatchListConsistencyIfRequested performs a data consistency check only when
the KUBE_WATCHLIST_INCONSISTENCY_DETECTOR environment variable was set during a binary startup.
The consistency check is meant to be enforced only in the CI, not in production.
The check ensures that data retrieved by the watch-list api call
is exactly the same as data received by the standard list api call.
Note that this function will panic when data inconsistency is detected.
This is intentional because we want to catch it in the CI.
Kubernetes-commit: b31e7793d0d873a71c90caf8455556aa905cf88d
