feat: add go local provider #100

vahidlazio · 2025-10-31T00:34:37Z

No description provided.

openfeature-provider/go/local_resolver_provider.go

openfeature-provider/go/local_resolver_factory.go

openfeature-provider/go/provider_builder.go

andreas-karlsson · 2025-11-03T10:20:40Z

openfeature-provider/go/swap_wasm_resolver_api.go

+	response, err := instance.Resolve(request)
+
+	// If instance is closed, retry with the current instance (which may have been swapped)
+	if err != nil && errors.Is(err, ErrInstanceClosed) {


Should rather be a while loop?

for how long?

it only happens once since the instance had to be swapped for this to happen.

Yes. It's very rare, but technically it could happen more than once if we for some reason flush more frequently.

We won't swap the instance very often, so it's unlikely we end up in a loop of tries on unloaded instances. I think we can keep this code as is

andreas-karlsson · 2025-11-03T10:24:44Z

openfeature-provider/go/swap_wasm_resolver_api.go

+	oldInstance := s.currentInstance.Swap(newInstance).(*ResolverApi)
+
+	// Close the old instance
+	oldInstance.Close(ctx)


There might be concurrent resolves while we close, so close can't call into the wasm module.

we have a lock in the resolver api, might be multiple calls to the resolve but if it's closing, all of them return isClosing error and new instance is aquired.

the flush is also locked with same lock so it's safe.

andreas-karlsson · 2025-11-03T10:28:59Z

openfeature-provider/go/resolver_api.go

+func (r *ResolverApi) FlushLogs() error {
+	r.mu.Lock()
+	// Mark instance as closing to prevent new resolves
+	r.isClosing = true


Seems odd that we close on FlushLogs?

no why? we do the same in Java, every time we close we flush logs and create a new instance.

nicklasl

Not really done here but leaving some comments to start.
Also: Needs a full suit of unit tests

openfeature-provider/go/demo/demo

openfeature-provider/go/demo/main.go

openfeature-provider/go/demo/README.md

nicklasl · 2025-11-03T09:18:37Z

openfeature-provider/go/demo/README.md

+# WASM file path (default: ../../../bin/resolver.wasm)
+export CONFIDENCE_WASM_FILE_PATH="/path/to/resolver.wasm"
+


not used.

Suggested change

# WASM file path (default: ../../../bin/resolver.wasm)

export CONFIDENCE_WASM_FILE_PATH="/path/to/resolver.wasm"

openfeature-provider/go/demo/README.md

nicklasl · 2025-11-03T09:59:01Z

openfeature-provider/go/local_resolver_factory.go

+
+		// Do initial state fetch to get initial state for SwapWasmResolverApi
+		if err := stateFetcher.Reload(ctx); err != nil {
+			log.Printf("Initial state fetch failed, using empty state: %v", err)


should throw an error?

nicklasl · 2025-11-03T10:05:21Z

openfeature-provider/go/local_resolver_factory.go

+		// Get account name from token
+		token, err := tokenHolder.GetToken(ctx)
+		if err != nil {
+			log.Printf("Warning: failed to get initial token, account name will be unknown: %v", err)


shouldn't this also be a big problem?

openfeature-provider/go/local_resolver_factory.go

openfeature-provider/go/provider_builder.go

Embed the Confidence resolver WASM module directly into the Go provider binary using Go 1.16+ embed directive. This eliminates runtime file dependencies and ensures version consistency between the Go code and WASM module. ## Changes **Embedding mechanism:** - Add `wasm_data.go` with `//go:embed wasm/confidence_resolver.wasm` - WASM file committed to `openfeature-provider/go/wasm/` as source - Update `provider_builder.go` to use `defaultWasmBytes` instead of `os.ReadFile()` - Remove `os` import (no longer needed) **Version safety with Docker validation:** - Add `openfeature-provider-go.validate-wasm` Docker stage that compares built WASM (from artifact) with committed WASM (from git) - Build fails if files don't match, preventing version drift - Validation runs automatically in CI as part of 'all' target **Developer workflow:** - Add `make sync-wasm-go` target that: 1. Builds WASM using Docker (ensures correct toolchain/dependencies) 2. Extracts artifact to local directory 3. Copies to `openfeature-provider/go/wasm/` 4. Prompts developer to commit - Update `.gitignore` to track embedded WASM while ignoring build artifact **User experience:** - SDK users: `go get` just works, WASM embedded in module - No manual WASM file management required - Version consistency guaranteed by CI **Backwards compatibility:** - `ProviderConfigWithStateProvider.WasmBytes` still supported for testing/advanced use cases - Falls back to embedded WASM if not provided

fabriziodemaria

Feel free to disregard the nit comments, if you don't think they are worth doing now

fabriziodemaria · 2025-11-04T09:53:15Z

openfeature-provider/go/local_resolver_factory.go

+// NewLocalResolverFactory creates a new LocalResolverFactory with gRPC clients and WASM bytes
+// If customStateProvider is not nil, it will be used; otherwise creates a FlagsAdminStateFetcher
+// Exposure logging is automatically enabled with gRPC services, disabled with custom StateProvider
+// accountId is required when using customStateProvider, otherwise it's extracted from the token
+func NewLocalResolverFactory(
+	ctx context.Context,
+	runtime wazero.Runtime,
+	wasmBytes []byte,
+	resolverStateServiceAddr string,
+	flagLoggerServiceAddr string,
+	authServiceAddr string,
+	apiClientID string,
+	apiClientSecret string,
+	customStateProvider StateProvider,
+	accountId string,
+) (*LocalResolverFactory, error) {


nit: We could create two separate constructors for gRPC-based vs. custom state provider, to make it a bit easier to read and prevent errors

On second thought, we probably can avoid exposing the ResolverFactory to the integrator (see comment below): in that case I don't mind keeping these constructors as they are

openfeature-provider/go/local_resolver_factory.go

fabriziodemaria · 2025-11-04T10:01:21Z

openfeature-provider/go/local_resolver_factory.go

+func NewLocalResolverFactory(
+	ctx context.Context,
+	runtime wazero.Runtime,
+	wasmBytes []byte,


nit: let's rename this to ConfidenceResolverWasm or something like that?

fabriziodemaria · 2025-11-04T10:13:31Z

openfeature-provider/go/local_resolver_factory.go

+		if err != nil {
+			log.Printf("Initial state fetch failed, using empty state: %v", err)
+			// TODO should we return an error here?
+			// return nil, fmt.Errorf("failed to get initial state: %w", err)


I think we should throw an error here: missing an initial state would resolve in default values, which is worse than "stale" values in the case a later fetch fails

I would like to have tests to verify that throwing here does not escape through OpenFeature as an exception/error.

I would like to have tests to verify that throwing here does not escape through OpenFeature as an exception/error.

Good point. The OF strategy is to never throw, so really I am not sure what we gain by throwing an error here; thinking in OF terms, I suppose we should set the error code PROVIDER_NOT_READY to all evaluations, but I don't think this is a blocker for getting this PR merged. We can improve on the error codes in a future PR.

In other words, I think we can leave this as is for now

fabriziodemaria · 2025-11-04T10:19:41Z

openfeature-provider/go/swap_wasm_resolver_api.go

+	response, err := instance.Resolve(request)
+
+	// If instance is closed, retry with the current instance (which may have been swapped)
+	if err != nil && errors.Is(err, ErrInstanceClosed) {


We won't swap the instance very often, so it's unlikely we end up in a loop of tries on unloaded instances. I think we can keep this code as is

openfeature-provider/go/state_fetcher.go

fabriziodemaria · 2025-11-04T10:31:11Z

openfeature-provider/go/state_fetcher.go

+		return err
+	}
+
+	f.accountID = response.Account


We have unsynchronized access to f.accountID for read and write, which could cause race conditions. Not sure how dangerous they might be, but perhaps we can avoid this line altogether as we don't expect the account id to change between two fetches within the same session?

openfeature-provider/go/demo/README.md

fabriziodemaria · 2025-11-04T10:40:09Z

openfeature-provider/go/grpc_wasm_flag_logger.go

+			if _, err := stub.WriteFlagLogs(rpcCtx, request); err != nil {
+				log.Printf("Failed to write flag logs: %v", err)
+			} else {
+				log.Printf("Successfully sent flag log with %d entries", len(request.FlagAssigned))


nit: we might need to make the logging levels configurable to not "spam" the integrator's console

fabriziodemaria · 2025-11-04T10:48:18Z

openfeature-provider/go/grpc_wasm_flag_logger.go

+		go func() {
+			defer logger.wg.Done()
+			// Create a context with timeout for the RPC
+			rpcCtx, cancel := context.WithTimeout(context.Background(), 30*time.Second)


probably just a nit: to my understanding is usually better to use ctx rather than creating a new detached background context here. AI suggests me that, if we really want to have a detached context, a slightly better way is detachedCtx := context.WithoutCancel(ctx) (but I haven't tested this)

Disable provenance attestations and set SOURCE_DATE_EPOCH to fix non-deterministic builds in docker/build-push-action@v6. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <[email protected]>

Pin protoc and protobuf-dev to version 29.4-r0 in all Alpine stages to ensure consistent proto code generation across different build environments. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <[email protected]>

fabriziodemaria · 2025-11-04T13:14:56Z

openfeature-provider/go/local_resolver_provider.go

+	case int:
+		return structpb.NewNumberValue(float64(v)), nil
+	case int64:
+		return structpb.NewNumberValue(float64(v)), nil
+	case float64:
+		return structpb.NewNumberValue(v), nil


Should we add other common types like int32, float32 (and maybe uint?)

This reverts commit bff6931.

This reverts commit 47d3fa7.

vahidlazio force-pushed the move-go-provider branch from beac165 to a8f2e5f Compare October 31, 2025 00:35

andreas-karlsson reviewed Nov 3, 2025

View reviewed changes

nicklasl reviewed Nov 3, 2025

View reviewed changes

vahidlazio and others added 15 commits November 4, 2025 11:41

move go local provider to repo

e6bfd8b

Remove generated .pb.go files from git tracking

51797ee

fix: some minor PR fixes

2e8f5b3

fix: add generated protobuf files

2b666a6

build(go): add makefile for go provider

8aaa31a

build(go): add Go provider to docker file

05d4306

fix(go): Make FlagsAdminStateFetcher implement StateProvider interface

7373c44

fix(go): remove disableable exposure logging

bd9e220

fix(go): align error message with other providers

7502378

chore(go): remove dead code

d3a4c91

chore(go): add some comments

dd3efc4

chore(go): pin rust version to make builds property deterministic

dcfbae4

feat(go): setup release please

b6c018e

fix: copy Go provider Cargo.toml in Docker build stages

520c115

nicklasl force-pushed the move-go-provider branch from 66f8cc1 to 520c115 Compare November 4, 2025 10:42

fabriziodemaria reviewed Nov 4, 2025

View reviewed changes

fix(ci): ensure deterministic WASM builds

47d3fa7

Disable provenance attestations and set SOURCE_DATE_EPOCH to fix non-deterministic builds in docker/build-push-action@v6. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <[email protected]>

nicklasl force-pushed the move-go-provider branch from 800e0e3 to 47d3fa7 Compare November 4, 2025 12:13

fabriziodemaria reviewed Nov 4, 2025

View reviewed changes

fix: try to define docker platform

0dde2ea

nicklasl force-pushed the move-go-provider branch from 2aaba8a to 0dde2ea Compare November 4, 2025 13:42

nicklasl added 2 commits November 4, 2025 14:50

Revert "fix: pin protoc to 29.4-r0 for deterministic builds"

d4fc797

This reverts commit bff6931.

Revert "fix(ci): ensure deterministic WASM builds"

c2d46a7

This reverts commit 47d3fa7.

nicklasl changed the title ~~feat: move go local provider~~ feat: add go local provider Nov 4, 2025

fix: declare arm64 on make sync-wasm-go

4d6458b

fix(go): clean up unused poll interval

168a41f

fabriziodemaria approved these changes Nov 4, 2025

View reviewed changes

nicklasl added 2 commits November 4, 2025 15:58

docs(go): add readme

cffb029

chore(go): cleanup unused func

5deb3ab

nicklasl approved these changes Nov 4, 2025

View reviewed changes

nicklasl merged commit 5c0895b into main Nov 4, 2025
10 checks passed

nicklasl deleted the move-go-provider branch November 4, 2025 15:16

github-actions bot mentioned this pull request Nov 4, 2025

chore: release main #102

Open

		# WASM file path (default: ../../../bin/resolver.wasm)
		export CONFIDENCE_WASM_FILE_PATH="/path/to/resolver.wasm"

feat: add go local provider #100

feat: add go local provider #100

Uh oh!

Conversation

vahidlazio commented Oct 31, 2025

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!

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!

nicklasl left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

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!

fabriziodemaria 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!

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!

Choose a reason for hiding this comment

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development