plugin,etc: Rewrite to get state from Pod annotations #1163

sharnoff · 2024-12-03T06:39:10Z

Note

Hey! This is a big PR — sorry! There's a lot of smaller commits (see below), plus one GIANT commit that does the bulk of the rewrite, building on those smaller commits.

There's an overview of the changes in the commit message for the full rewrite. That might be a useful place to start to get your bearings on the changes.

Commits broken down by theme:

~~Background work: neonvm~~ MERGED in #1194

neonvm-controller: Use non-controller owner ref for migration source (d6fa7d0)
In short, there's currently no reliable way to tell whether a Pod is the source or target in a live migration without checking against the migration object itself.

This is because the migration object initially has a unique "controller" reference on the target pod (with the source pod having a non-controlling owner reference), and once the migration is complete, the source pod is changed to have the unique controller reference setup.

So this change is just to keep using a non-controller reference on the source pod after the migration completes.

That means that we will have the following guarantees:
- If a Pod has a "controller" owner reference for a live migration, it's the target pod in an ongoing migration.
- If a Pod has a non-controller owner reference for a live migration, it's the source pod in a maybe-ongoing migration.
- If the Pod has a "controller" owner reference for a virtual machine, it's the current source of the VM. So if a migration source pod doesn't have an owner reference for the VM, then the migration has completed and there is no longer a use for the old source pod.
neonvm: Add helpers to get pod ownership (e77ac86)

We had some existing, similar helpers in pkg/util, but I figured it makes more sense for those to be defined in neonvm/apis/....

This change originated while working on the stateless scheduler prototype, where I found myself wanting a reliable way to determine what role a pod has in a live migration from only the pod's metadata.
neonvm-controller: Update runner pod metadata while pending (a1b55f7)

We weren't doing this previously, and it means that propagation of labels/annotations can be arbitrarily delayed while the VM starts.

Related changes in util/watch

util/watch: Store HandlerFuncs[*T] in Store[T] (ad2340f)

Also, change handleEvent[T any, P ~*T]() from a function to a method on *Store[T], now that we have the handlers translated from P → *T.

This opens up a lot of ways to make the code cleaner, and the handlers part in particular is required to implement (*Store[T]).NopUpdate() in a later commit.
util/watch: Add (*Store[T]).NopUpdate() method (1f43c44)

There's some info about this in the added comment.

tl;dr: We need it for the stateless scheduler work to be able to re-inject items into the reconcile queue while maintaining the watch as the source of truth for what each object is.
util/watch: Override GVK on incoming objects (2580422)

The K8s API server and client-go together have the behavior that objects returned from List() calls do not have TypeMeta set.

For one-off List() requests this is fine becuase you already know the type! But this interacts poorly with the generated implementations of objects' .GetObjectKind().GroupVersionKind(), as those just directly read from the TypeMeta fields (which again: are not set).

So this commit works around this behavior by getting the GVK at the start of the Watch() call and explicitly setting it on all incoming objects.
util/watch: Add (*Store[T]).Listen() method (341c0b5)

The Listen() method returns a util.BroadcastReceiver that will be updated whenever the object is modified or deleted.

This is required for now for the stateless scheduler work, so that we can be separately notified when there's changes to an object without hooking deeper into the internal state.

We can probably remove this once the scheduler plugin's agent request handler server is removed, at the end of the stateless scheduler work.

The one big commit doing the rewrite

plugin: Rewrite to get state from Pod annotations (d04ce1b)
a.k.a. "Stateless Scheduler".

This is effectively a full rewrite of the scheduler plugin. At a high level, the existing external interfaces are preserved:
- The scheduler plugin still exposes an HTTP server for the autoscaler-agent (for now); and
- The scheduler plugin is still a plugin.
However, instead of storing the state for approved resources in-memory, in the scheduler plugin, we now treat annotations on the Pod as the source of truth for requested/approved resources.

A brief overview of the internal changes to make this happen:
1. The state of resource reservations can be constructed entirely from Node and Pod objects. We do store that, and update as objects change, but it's only for convenience and not a strict requirement.
  
  One tricky piece is with scheduling. For that, we store a set of pods that have gone through the plugin methods but haven't actually had the spec.nodeName field set.
  
  For more info, the pkg/plugin/state package contains all the pure logic for manipulating resources.
2. Each watch event on Node/Pod objects is now placed into a "reconcile" queue similar to the controller framework. Reconcile operations are a tuple of (object, event type, desired reconcile time) and are retried with backoff on error/panic.
  
  For a detailed look, the 'pkg/plugin/reconcile' package defines the reconcile queue and all its related machinery.
3. The handler for autoscaler-agent requests no longer accesses the internal state and instead directly patches the VirtualMachine object to set the annotation for requested resources, and then waits for that object to be updated.
  
  Once the autoscaler-agent is converted to read and write those annotations directly, we will remove the HTTP server.
4. pkg/util/watch was changed to allow asking to be notified when
  there's changes to an object, via the new (*Store[T]).Listen() API.
  
  This was required to implement (3), and can be removed once (3) is no longer needed, if it doesn't become used in the autoscaler-agent.
5. pkg/util/watch was changed to allow triggering no-op update events, which - for our usage - will trigger requeuing the object. This solves two problems:
  1. During initial startup, we need to defer resource approval until all Pods on the Node have been processed -- otherwise, we may end up unintentionally overcommitting resources based on partial information.
    
    So during startup, we track the set of Pods with deferred approvals, and then requeue them all once startup is over by triggering no-op update events in the watch store.
  2. Whenever we handle changes for some Pod, it's worthwhile to handle certain operations on the Node -- e.g., triggering live migration if the reserved resources are too high.
    
    While we could do this as part of the Pod reconciling, we get more fair behavior (and, better balancing under load) by instead triggering re-reconciling the Pod's Node.
  Why can't this be done elsewhere? In short, consistency. Fundamentally we need to use a consistent view of the object that we're reconciling (else, it might not be no-op), and the source of truth for the current value of an object within the scheduler plugin is the watch store.

Remaining TODOs:

Flesh out the reconcile metrics (3693980)
Add metrics for k8s CRUD operations (mainly: rate of VirtualMachine update operations) (3693980)
Test node metrics (40b43c1)
Test that deferred reconciles on startup work correctly
Test that setting VM CPU/memory use fields and enabling autoscaling at the same time works correctly.
Load testing on staging (link)

Open questions:

Which commits should be squashed, or extracted into separate PRs?
We have a single global state mutex (same as before, but different usage pattern). How big a cluster can we support before that falls over? Do we need to have individual state locks per node?

github-actions · 2024-12-03T06:42:35Z

No changes to the coverage.

HTML Report

Click to open

Probably an inadvertent merge conflict between #1090 and #989 meaning we accidentally weren't using go-chef for neonvm-daemon. Noticed this while working on #1163 locally and saw that it was re-downloading all of the dependencies for neonvm-daemon every time, even though I was making changes in the scheduler and the dependencies hadn't changed.

It wasn't correct; the separator is '/', not ':'. (I think once upon a time, we used to format it with ':', but that's no longer the case). Noticed this as part of #1163.

sharnoff · 2024-12-13T17:17:22Z

Squashed 06240dd..0b64f08 (diff)
Rebased onto main: 0b64f08..6c76e9d (diff)

Omrigan

Partial review, got to the middle of handle_pod.

neonvm/apis/neonvm/v1/pod_helpers.go

pkg/plugin/reconcile/event.go

pkg/neonvm/controllers/vmmigration_controller.go

pkg/plugin/globalstate.go

pkg/plugin/handle_pod.go

petuhovskiy · 2024-12-20T11:54:21Z

Which commits should be squashed, or extracted into separate PRs?

The first 7 commits look quite isolated from the big "stateless scheduler" commit. They also LGTM, so you can create a PR out of these commits and merge them to main, if you want.

pkg/plugin/reconcile/middleware.go

pkg/plugin/reconcile/queue.go

pkg/neonvm/controllers/vmmigration_controller.go

pkg/plugin/reconcile/options.go

petuhovskiy

Finished my first reading, overall the code looks very good. Will come back for a second iteration next year :)

Made some small refactorings along the way, mostly it is about moving metrics-related code to a new package: https://github.com/neondatabase/autoscaling/pull/1182/commits

The changes are described in commit messages, feel free to accept/reject.

pkg/plugin/reconcile/middleware.go

pkg/plugin/state/xactmap.go

pkg/plugin/state/node.go

pkg/plugin/reconcile/event.go

pkg/plugin/handle_node.go

Extracted from #1163, where `gci` incorrectly groups the "iter" import away from other stdlib pacakges, because it was only added in Go 1.23.

sharnoff · 2025-01-02T21:33:42Z

Squashed 71fc62f..11b50ac (diff)
Rebased 11b50ac..d04ce1b (diff)
Extracted the neonvm commits into neonvm: Small changes for stateless scheduler #1194
Responded to most of the review comments; still a few left.
- Main thing I still want to sort out is the reconcile queue's event types — probably worth a brief discussion (maybe async) about our options there

petuhovskiy

Found a couple of small issues. I can approve everything except framework_methods.go, will take a look at it again tomorrow.

UPD: accidentally pressed approve instead of comment, but no big difference :)

autoscale-scheduler/cmd/main.go

autoscale-scheduler/config_map.yaml

pkg/plugin/setup.go

pkg/plugin/state/node.go

pkg/plugin/framework_methods.go

pkg/plugin/handle_node.go

sharnoff · 2025-01-21T10:53:56Z

Updates:

Rebased onto main (d04ce1b..a11094d) (diff)
Small review fixes (a11094d..b47ba95) (diff)
Squashed (b47ba95..86376ea) (diff)

Omrigan

Almost done!

pkg/plugin/reconcile/options.go

neonvm/apis/neonvm/v1/pod_helpers.go

pkg/plugin/reconcile/event.go

pkg/plugin/reconcile/middleware.go

pkg/plugin/reconcile/queue.go

pkg/plugin/globalstate.go

pkg/plugin/reconcile/queue.go

sharnoff · 2025-01-30T08:41:53Z

Updates:

Small changes from review: 86376ea..987d73b (diff)
Squashed 987d73b..289735a (diff)
Rebased onto main 289735a..dcfb842 (diff)

sharnoff · 2025-02-04T09:30:24Z

Updates:

Rebased onto main dcfb842..24b5ff5 (diff)

Also, change handleEvent[T any, P ~*T]() from a function to a method on *Store[T], now that we have the handlers translated from P -> *T. This opens up a lot of ways to make the code cleaner, and the handlers part in particular is required to implement (*Store[T]).NopUpdate() in a later commit.

There's some info about this in the added comment. tl;dr: We need it for the stateless scheduler work to be able to re-inject items into the reconcile queue while maintaining the watch as the source of truth for what each object is.

The K8s API server and client-go together have the behavior that objects returned from List() calls do not have TypeMeta set. For one-off List() requests this is fine becuase you already know the type! But this interacts poorly with the generated implementations of objects' .GetObjectKind().GroupVersionKind(), as those just directly read from the TypeMeta fields (which again: are not set). So this commit works around this behavior by getting the GVK at the start of the Watch() call and explicitly setting it on all incoming objects.

The Listen() method returns a util.BroadcastReceiver that will be updated whenever the object is modified or deleted. This is required *for now* for the stateless scheduler work, so that we can be separately notified when there's changes to an object without hooking deeper into the internal state. We can probably remove this once the scheduler plugin's agnent request handler server is removed, at the end of the stateless scheduler work.

a.k.a. "Stateless Scheduler". This is effectively a full rewrite of the scheduler plugin. At a high level, the existing external interfaces are preserved: - The scheduler plugin still exposes an HTTP server for the autoscaler-agent (for now); and - The scheduler plugin is still a plugin. However, instead of storing the state for approved resources in-memory, in the scheduler plugin, we now treat annotations on the Pod as the source of truth for requested/approved resources. A brief overview of the internal changes to make this happen: 1. The state of resource reservations can be constructed entirely from Node and Pod objects. We *do* store that, and update as objects change, but it's only for convenience and not a strict requirement. One tricky piece is with scheduling. For that, we store a set of pods that have gone through the plugin methods but haven't actually had the spec.nodeName field set. For more info, the 'pkg/plugin/state' package contains all the pure logic for manipulating resources. 2. Each watch event on Node/Pod objects is now placed into a "reconcile" queue similar to the controller framework. Reconcile operations are a tuple of (object, event type, desired reconcile time) and are retried with backoff on error/panic. For a detailed look, the 'pkg/plugin/reconcile' package defines the reconcile queue and all its related machinery. 3. The handler for autoscaler-agent requests no longer accesses the internal state and instead directly patches the VirtualMachine object to set the annotation for requested resources, and then waits for that object to be updated. Once the autoscaler-agent is converted to read and write those annotations directly, we will remove the HTTP server. 4. 'pkg/util/watch' was changed to allow asking to be notified when there's changes to an object, via the new (*Store[T]).Listen() API. This was required to implement (3), and can be removed once (3) is no longer needed, if it doesn't become used in the autoscaler-agent. 5. 'pkg/util/watch' was changed to allow triggering no-op update events, which - for our usage - will trigger requeuing the object. This solves two problems: a. During initial startup, we need to defer resource approval until all Pods on the Node have been processed -- otherwise, we may end up unintentionally overcommitting resources based on partial information. So during startup, we track the set of Pods with deferred approvals, and then requeue them all once startup is over by triggering no-op update events in the watch store. b. Whenever we handle changes for some Pod, it's worthwhile to handle certain operations on the Node -- e.g., triggering live migration if the reserved resources are too high. While we *could* do this as part of the Pod reconciling, we get more fair behavior (and, better balancing under load) by instead triggering re-reconciling the Pod's Node. Why can't this be done elsewhere? In short, consistency. Fundamentally we need to use a consistent view of the object that we're reconciling (else, it might not be no-op), and the source of truth for the current value of an object *within the scheduler plugin* is the watch store. Co-authored-by: Arthur Petukhovsky <[email protected]>

sharnoff · 2025-02-10T20:06:17Z

Updates:

Small review fix 24b5ff5..f1f164d (diff)
Squash f1f164d..d1a208b (diff)
Rebase onto main d1a208b..76e0e92 (diff)

Also, change handleEvent[T any, P ~*T]() from a function to a method on *Store[T], now that we have the handlers translated from P -> *T. This opens up a lot of ways to make the code cleaner, and the handlers part in particular is required to implement (*Store[T]).NopUpdate() in a later commit.

There's some info about this in the added comment. tl;dr: We need it for the stateless scheduler work to be able to re-inject items into the reconcile queue while maintaining the watch as the source of truth for what each object is.

The K8s API server and client-go together have the behavior that objects returned from List() calls do not have TypeMeta set. For one-off List() requests this is fine becuase you already know the type! But this interacts poorly with the generated implementations of objects' .GetObjectKind().GroupVersionKind(), as those just directly read from the TypeMeta fields (which again: are not set). So this commit works around this behavior by getting the GVK at the start of the Watch() call and explicitly setting it on all incoming objects.

The Listen() method returns a util.BroadcastReceiver that will be updated whenever the object is modified or deleted. This is required *for now* for the stateless scheduler work, so that we can be separately notified when there's changes to an object without hooking deeper into the internal state. We can probably remove this once the scheduler plugin's agnent request handler server is removed, at the end of the stateless scheduler work.

This was referenced Dec 3, 2024

util: Fix comment in NamespacedName formatter #1164

Merged

neonvm-daemon: Use go-chef in Dockerfile #1165

Merged

sharnoff force-pushed the sharnoff/stateless-scheduler branch from cf468d8 to 34f1106 Compare December 10, 2024 20:43

Omrigan self-requested a review December 12, 2024 13:40

sharnoff force-pushed the sharnoff/stateless-scheduler branch 2 times, most recently from 0b64f08 to 6c76e9d Compare December 13, 2024 17:15

Omrigan reviewed Dec 16, 2024

View reviewed changes

sharnoff mentioned this pull request Dec 16, 2024

agent,plugin: Remove metrics from AgentRequests #1145

Closed

Omrigan reviewed Dec 20, 2024

View reviewed changes

petuhovskiy reviewed Dec 25, 2024

View reviewed changes

sharnoff mentioned this pull request Dec 30, 2024

Internal feature: Stateless scheduler #995

Open

sharnoff added a commit that referenced this pull request Dec 30, 2024

golangci: Bump v1.60.3 -> v1.62.2

8bd5732

Extracted from #1163, where `gci` incorrectly groups the "iter" import away from other stdlib pacakges, because it was only added in Go 1.23.

sharnoff added a commit that referenced this pull request Dec 30, 2024

golangci-lint: Bump v1.60.3 -> v1.62.2

6938018

Extracted from #1163, where `gci` incorrectly groups the "iter" import away from other stdlib pacakges, because it was only added in Go 1.23.

sharnoff mentioned this pull request Dec 30, 2024

golangci-lint: Bump v1.60.3 -> v1.62.2 #1191

Merged

sharnoff added a commit that referenced this pull request Jan 2, 2025

golangci-lint: Bump v1.60.3 -> v1.62.2 (#1191)

10f0b1d

Extracted from #1163, where `gci` incorrectly groups the "iter" import away from other stdlib pacakges, because it was only added in Go 1.23.

sharnoff force-pushed the sharnoff/stateless-scheduler branch 2 times, most recently from 11b50ac to d04ce1b Compare January 2, 2025 21:33

sharnoff mentioned this pull request Jan 2, 2025

neonvm: Small changes for stateless scheduler #1194

Merged

sharnoff assigned Omrigan and petuhovskiy Jan 2, 2025

petuhovskiy self-requested a review January 2, 2025 22:31

This was referenced Jan 3, 2025

WIP: Arthur's refactoring on top of #1163 #1182

Closed

Get rid of VM watcher in plugin #873

Open

petuhovskiy approved these changes Jan 6, 2025

View reviewed changes

sharnoff force-pushed the sharnoff/stateless-scheduler branch from b47ba95 to 86376ea Compare January 21, 2025 12:35

sharnoff assigned petuhovskiy and unassigned sharnoff Jan 21, 2025

petuhovskiy approved these changes Jan 22, 2025

View reviewed changes

petuhovskiy removed their assignment Jan 22, 2025

Omrigan approved these changes Jan 27, 2025

View reviewed changes

sharnoff assigned sharnoff and unassigned Omrigan Jan 28, 2025

sharnoff force-pushed the sharnoff/stateless-scheduler branch 2 times, most recently from 289735a to dcfb842 Compare January 30, 2025 08:36

sharnoff force-pushed the sharnoff/stateless-scheduler branch from dcfb842 to 24b5ff5 Compare February 4, 2025 09:30

sharnoff mentioned this pull request Feb 7, 2025

Add cluster-trace-collector tool #1247

Draft

sharnoff force-pushed the sharnoff/stateless-scheduler branch from f1f164d to d1a208b Compare February 10, 2025 20:02

sharnoff and others added 5 commits February 10, 2025 20:04

sharnoff force-pushed the sharnoff/stateless-scheduler branch from d1a208b to 76e0e92 Compare February 10, 2025 20:05

sharnoff enabled auto-merge (rebase) February 10, 2025 20:06

sharnoff merged commit 9164afc into main Feb 10, 2025
23 checks passed

sharnoff deleted the sharnoff/stateless-scheduler branch February 10, 2025 20:22

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

plugin,etc: Rewrite to get state from Pod annotations #1163

plugin,etc: Rewrite to get state from Pod annotations #1163

sharnoff commented Dec 3, 2024 •

edited

Loading

github-actions bot commented Dec 3, 2024 •

edited

Loading

sharnoff commented Dec 13, 2024

Omrigan left a comment

petuhovskiy commented Dec 20, 2024

petuhovskiy left a comment

sharnoff commented Jan 2, 2025 •

edited

Loading

petuhovskiy left a comment •

edited

Loading

sharnoff commented Jan 21, 2025 •

edited

Loading

Omrigan left a comment

sharnoff commented Jan 30, 2025

sharnoff commented Feb 4, 2025

sharnoff commented Feb 10, 2025

plugin,etc: Rewrite to get state from Pod annotations #1163

plugin,etc: Rewrite to get state from Pod annotations #1163

Conversation

sharnoff commented Dec 3, 2024 • edited Loading

github-actions bot commented Dec 3, 2024 • edited Loading

HTML Report

sharnoff commented Dec 13, 2024

Omrigan left a comment

Choose a reason for hiding this comment

petuhovskiy commented Dec 20, 2024

petuhovskiy left a comment

Choose a reason for hiding this comment

sharnoff commented Jan 2, 2025 • edited Loading

petuhovskiy left a comment • edited Loading

Choose a reason for hiding this comment

sharnoff commented Jan 21, 2025 • edited Loading

Omrigan left a comment

Choose a reason for hiding this comment

sharnoff commented Jan 30, 2025

sharnoff commented Feb 4, 2025

sharnoff commented Feb 10, 2025

sharnoff commented Dec 3, 2024 •

edited

Loading

github-actions bot commented Dec 3, 2024 •

edited

Loading

sharnoff commented Jan 2, 2025 •

edited

Loading

petuhovskiy left a comment •

edited

Loading

sharnoff commented Jan 21, 2025 •

edited

Loading