*: Implementation of weighted random shuffling (A113) by easwars · Pull Request #8864 · grpc/grpc-go

easwars · 2026-01-29T07:07:50Z

This PR implements the currently in-review gRFC A113: grpc/proposal#535. I've split the PR into logically separate commits to help with the review process.

Summary of changes:

Commit 1: simplify the implementation of groupLocalitiesByPriority
- Change the implementation to use newly added methods in the stdlib maps and slices package to significantly simplify the implementation (and get rid of an unnecessary test)
Commit 2: Remove code that handles localities and endpoints of weight 0
- Remove unnecessary checks for locality and endpoint weights of 0 in cluster_resolver. The xDS client already guarantees that these weights will never be set to 0.
Commit 3: add the env var GRPC_EXPERIMENTAL_PF_WEIGHTED_SHUFFLING
Commit 4: Weight computation changes in cluster_resolver LB policy
- This performs the weight normalization and fixed-point arithmetic specified in A113
- The change here is guarded by the above env var
- Ended up duplicating the tests that verify the weight computation behavior. This will make it easier to delete the old tests when the env var is removed.
Commit 5: Fix a broken test in ring_hash due to the new weight computation
Commit 6: Weighted shuffling in pick_first
- Contains the changes specified in A113 for the pick_first LB policy
- Changes are guarded by the env var

RELEASE NOTES:

pickfirst: Add support for weighted random shuffling of endpoints, as described in gRFC A113

This is already taken care of by the xDS client

codecov · 2026-01-29T07:14:06Z

Codecov Report

❌ Patch coverage is 92.00000% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 83.17%. Comparing base (69b542a) to head (3f4487d).
⚠️ Report is 4 commits behind head on master.

Files with missing lines	Patch %	Lines
balancer/pickfirst/pickfirst.go	92.59%	1 Missing and 1 partial ⚠️
...rnal/xds/balancer/clusterresolver/configbuilder.go	91.30%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #8864      +/-   ##
==========================================
+ Coverage   82.42%   83.17%   +0.75%     
==========================================
  Files         414      414              
  Lines       32750    32755       +5     
==========================================
+ Hits        26994    27245     +251     
+ Misses       4096     4090       -6     
+ Partials     1660     1420     -240

Files with missing lines	Coverage Δ
balancer/pickfirst/internal/internal.go	`100.00% <ø> (ø)`
internal/envconfig/envconfig.go	`100.00% <ø> (+26.66%)`	⬆️
balancer/pickfirst/pickfirst.go	`89.28% <92.59%> (-0.34%)`	⬇️
...rnal/xds/balancer/clusterresolver/configbuilder.go	`92.42% <91.30%> (-1.61%)`	⬇️

... and 26 files with indirect coverage changes

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

balancer/pickfirst/pickfirst.go

balancer/pickfirst/pickfirst_ext_test.go

balancer/pickfirst/pickfirst.go

arjan-bal · 2026-01-29T19:59:56Z

balancer/pickfirst/pickfirst.go

 		// will change the order of endpoints but not touch the order of the
 		// addresses within each endpoint. - A61
 		if cfg.ShuffleAddressList {
-			endpoints = append([]resolver.Endpoint{}, endpoints...)


We should continue cloning the endpoints list here. Directly mutating the ResolverState can lead to data races if the caller of UpdateClientConnState reads the state concurrently. It would also be helpful to add a comment explaining this to prevent the accidental removal of the copy in the future.

Argh. I feel like we need to either make resolver.State.Endpoints an immutable type (EndpointList) or guarantee it's deeply copied when resolver.State is copied.

balancer/ringhash/ringhash_e2e_test.go

internal/xds/balancer/clusterresolver/configbuilder.go

internal/xds/balancer/clusterresolver/configbuilder_test.go

arjan-bal

LGTM, with some minor comments.

arjan-bal · 2026-01-30T12:47:13Z

internal/xds/balancer/clusterresolver/configbuilder_test.go

+	// Endpoints weights are the product of normalized locality weight and
+	// endpoint weight, represented as a fixed-point number in uQ1.31 format.
+	// Locality weights are normalized as:
+	//   P1: locality 0: 80 / (100) = 0.8


I believe locality 0 has weight 20 and locality 3 has weight 80. They seem to be swapped in this comment.

OMG. This test is so messed up and I've made a copy of it, making it two problems instead of one. I've anyways filed #8868 to cleanup these tests.

arjan-bal · 2026-01-30T12:51:57Z

internal/xds/balancer/clusterresolver/configbuilder_test.go

 			}
-			if diff := cmp.Diff(got, tt.wantConfig); diff != "" {
+			if diff := cmp.Diff(gotConfig, tt.wantConfig); diff != "" {
 				t.Errorf("localitiesToWeightedTarget() diff (-got +want) %v", diff)


The error message logs the wrong function name. Same issue a couple of lines below and in the corresponding test with weighted sorting enabled.

arjan-bal · 2026-01-30T13:01:53Z

internal/xds/balancer/clusterresolver/configbuilder.go

+				}
+				resolverEndpoint = weight.Set(resolverEndpoint, weight.EndpointInfo{Weight: endpointWeight})
+			} else {
+				resolverEndpoint = weight.Set(resolverEndpoint, weight.EndpointInfo{Weight: locality.Weight * endpoint.Weight})


Not related to the changes introduced in this PR, but can't this multiplication result in an overflow since EndpointInfo.Weight is also a uint32?

Yes, it definitely can overflow. But since no one has complained so far and we will get rid of this code path soonish (once we delete the env var), I'm Ok letting it be as-is.

dfawley · 2026-01-30T17:01:22Z

balancer/pickfirst/pickfirst.go

 		// will change the order of endpoints but not touch the order of the
 		// addresses within each endpoint. - A61
 		if cfg.ShuffleAddressList {
-			endpoints = append([]resolver.Endpoint{}, endpoints...)


Argh. I feel like we need to either make resolver.State.Endpoints an immutable type (EndpointList) or guarantee it's deeply copied when resolver.State is copied.

dfawley · 2026-01-30T17:04:18Z

balancer/pickfirst/pickfirst.go

+				// Here, and in the "else" block below, we clone the endpoints
+				// slice to avoid mutating the resolver state. Doing the latter
+				// would lead to data races if the caller is accessing the same
+				// slice concurrently.


I think this is important even if it's not a concurrent update. We just shouldn't be mutating the incoming update's endpoints because that likely changes state in the parent.

dfawley · 2026-01-30T17:11:27Z

balancer/ringhash/ringhash_e2e_test.go

+	// in each locality, the probability of selecting the endpoint within the
+	// locality is 1. Therefore, the target fractions end up as 1/3 and 2/3
+	// respectively.
+	const wantRPCs0 = float64(1) / float64(3)


FWIW you only need to float64 either the numerator or the denominator.

This PR implements the currently in-review gRFC A113: grpc/proposal#535. I've split the PR into logically separate commits to help with the review process. Summary of changes: - Commit 1: simplify the implementation of `groupLocalitiesByPriority` - Change the implementation to use newly added methods in the stdlib `maps` and `slices` package to significantly simplify the implementation (and get rid of an unnecessary test) - Commit 2: Remove code that handles localities and endpoints of weight 0 - Remove unnecessary checks for locality and endpoint weights of `0` in `cluster_resolver`. The xDS client already guarantees that these weights will never be set to `0`. - Commit 3: add the env var GRPC_EXPERIMENTAL_PF_WEIGHTED_SHUFFLING - Commit 4: Weight computation changes in cluster_resolver LB policy - This performs the weight normalization and fixed-point arithmetic specified in A113 - The change here is guarded by the above env var - Ended up duplicating the tests that verify the weight computation behavior. This will make it easier to delete the old tests when the env var is removed. - Commit 5: Fix a broken test in ring_hash due to the new weight computation - Commit 6: Weighted shuffling in pick_first - Contains the changes specified in A113 for the pick_first LB policy - Changes are guarded by the env var RELEASE NOTES: - pickfirst: Add support for weighted random shuffling of endpoints, as described in gRFC A113

Original PR : #8864 RELEASE NOTES: - pickfirst: Add support for weighted random shuffling of endpoints, as described in gRFC A113 Co-authored-by: Easwar Swaminathan <easwars@google.com>

This PR implements the currently in-review gRFC A113: grpc/proposal#535. I've split the PR into logically separate commits to help with the review process. Summary of changes: - Commit 1: simplify the implementation of `groupLocalitiesByPriority` - Change the implementation to use newly added methods in the stdlib `maps` and `slices` package to significantly simplify the implementation (and get rid of an unnecessary test) - Commit 2: Remove code that handles localities and endpoints of weight 0 - Remove unnecessary checks for locality and endpoint weights of `0` in `cluster_resolver`. The xDS client already guarantees that these weights will never be set to `0`. - Commit 3: add the env var GRPC_EXPERIMENTAL_PF_WEIGHTED_SHUFFLING - Commit 4: Weight computation changes in cluster_resolver LB policy - This performs the weight normalization and fixed-point arithmetic specified in A113 - The change here is guarded by the above env var - Ended up duplicating the tests that verify the weight computation behavior. This will make it easier to delete the old tests when the env var is removed. - Commit 5: Fix a broken test in ring_hash due to the new weight computation - Commit 6: Weighted shuffling in pick_first - Contains the changes specified in A113 for the pick_first LB policy - Changes are guarded by the env var RELEASE NOTES: - pickfirst: Add support for weighted random shuffling of endpoints, as described in gRFC A113

easwars added 6 commits January 29, 2026 06:55

simplify the implementation of groupLocalitiesByPriority

6f9d9cf

Remove code that handles localities and endpoints of weight 0.

633fab8

This is already taken care of by the xDS client

add the env var GRPC_EXPERIMENTAL_PF_WEIGHTED_SHUFFLING

3bd5a7d

Weight computation changes in cluster_resolver LB policy

3a582da

Fix a broken test in ring_hash due to the new weight computation

aaa9d86

Weighted shuffling in pick_first

d4d3d88

easwars requested review from arjan-bal and dfawley January 29, 2026 07:07

easwars assigned dfawley and arjan-bal Jan 29, 2026

easwars added Area: Resolvers/Balancers Includes LB policy & NR APIs, resolver/balancer/picker wrappers, LB policy impls and utilities. Type: Feature New features or improvements in behavior labels Jan 29, 2026

easwars added this to the 1.79 Release milestone Jan 29, 2026

arjan-bal reviewed Jan 29, 2026

View reviewed changes

arjan-bal assigned easwars and unassigned arjan-bal Jan 29, 2026

easwars mentioned this pull request Jan 30, 2026

Improve the configbuilder tests in clusterresolver #8868

Closed

first round of review comments from Arjan

fa17114

easwars assigned arjan-bal and unassigned easwars Jan 30, 2026

arjan-bal approved these changes Jan 30, 2026

View reviewed changes

arjan-bal assigned easwars and unassigned arjan-bal Jan 30, 2026

second round of review comments from Arjan

3f4487d

dfawley reviewed Jan 30, 2026

View reviewed changes

dfawley approved these changes Jan 30, 2026

View reviewed changes

easwars merged commit 61d569d into grpc:master Jan 30, 2026
21 of 22 checks passed

easwars deleted the directpath_soft_affinity branch January 30, 2026 19:40

eshitachandwani mentioned this pull request Feb 11, 2026

Cherry pick #8864 to v1.79.x #8896

Merged

Conversation

easwars commented Jan 29, 2026

Uh oh!

codecov bot commented Jan 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

arjan-bal left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

codecov bot commented Jan 29, 2026 •

edited

Loading