how to use informer to list k8s object,like to list pods?

[vCassius]

apiserver would be pressure if all requests went to it,so i want to use informer to cache it.

[clux]

the informer equivalent is to pass a watcher stream through a reflector and giving it the write access to a store.
see e.g. node_reflector example:

kube/examples/node_reflector.rs

Lines 20 to 41 in 5e98a92

	let (reader, writer) = reflector::store();
	let stream = watcher(nodes, wc)
	.default_backoff()
	.reflect(writer)
	.applied_objects()
	.predicate_filter(predicates::labels.combine(predicates::annotations)); // NB: requires an unstable feature
	// Periodically read our state in the background
	tokio::spawn(async move {
	reader.wait_until_ready().await.unwrap();
	loop {
	let nodes = reader.state().iter().map(|r| r.name_any()).collect::<Vec<_>>();
	info!("Current {} nodes: {:?}", nodes.len(), nodes);
	tokio::time::sleep(std::time::Duration::from_secs(10)).await;
	}
	});
	// Log applied events with changes from the reflector
	pin_mut!(stream);
	while let Some(node) = stream.try_next().await? {
	info!("saw node {} with new labels/annots", node.name_any());
	}

[clux]

the writer is deliberately not clone-able, see doc str for it.

if you are customizing how storage and cleanup is done, you probably want to implement you own "writer-like" thing, with your own "apply-fn": https://docs.rs/kube-runtime/0.85.0/src/kube_runtime/reflector/store.rs.html#61-84 and call that on every event like reflector does

[vCassius]

thx, i will try it with apply_watcher_event

[vCassius]

seems event informer working,but i got this err:
2023-09-11 11:57:33.671 ERROR v_kubernetes_action::pkg::global::objects: event try next exception: error returned by apiserver during watch: The resourceVersion for the provided watch is too old.: Expired
at src/pkg/global/objects.rs:95

[clux]

i'm guessing if you don't have enough events flowing through the system, the normal watcher attempt to resume from a (now wiped) resume point will not work. because events wipe so frequently, the watcher will probably have to restart from "0" again a lot more frequently than a watcher for any other resource.

[clux]

so if you see every event being passed through then it shouldn't be anything to worry about in your case - you could filter out the error by matching on it if it's annoying.

[vCassius]

it's a way to forwarding all watcher objects to a vec then clean, when i used client-go to create informer, It automatically manages objects when the kubernetes create or destory object.