client: read-only client without plugin management or connect (client v0.2.2 adds 1.3 MB to the Docker CLI)
Mantenedores costumam responder em até 1 dia
Ninguém assumiu esta issue ainda.
Avaliação
- Dificuldade
- 5/5
- Tempo estimado
- Mais de uma semana
- Facilidade para iniciantes
- 45/100
- Tipo de issue
- Funcionalidade
- Clareza
- Razoavelmente clara
- Status de atividade
- Ativa
- Stack de tecnologia
- go
- Domínio
- api, backend, performance
Direção de pesquisa
Start by reading the existing client setup and resolver.NewResolverClient implementation, then compare the two proposed options and their measured binary sizes. The issue does not name source or test file paths, so locate the resolver client and Connect server tests in the repository. Done means a read-only client can fetch secrets with the stated error behavior, tests pass against the Connect server, and binary growth matches the intended reduction.
Escrita pelo modelo de indexação a partir do texto da issue.
Descrição
Summary
Using github.com/docker/secrets-engine/client v0.2.2 to read the Docker Hub session (client.New → HubAuth() → GetDefaultSession / GetDefaultProfile) grows the Docker CLI binary by about 3%: +1.27 MB unstripped and +0.85 MB stripped on linux/arm64. Most of that is code a read-only consumer never calls.
The Docker CLI is adding the Docker Desktop session as the first credential source for Docker Hub and dhi.io, as docker/compose does today. The CLI only calls GetSecrets (through dockerhub).
This issue proposes two changes to the client, which together cut that growth by 85%:
- A read-only client package (−53%), non-breaking.
- That package calling
GetSecretswithoutconnect(−85% in total).
Binary growth with client v0.2.2
docker built from docker/cli master, and from the same tree with the client added:
| platform | unstripped | stripped (-s -w) |
|---|---|---|
| linux/arm64 | 40,142,559 → 41,409,437 (+1,266,878, +3.2%) | 27,590,816 → 28,442,784 (+851,968, +3.1%) |
| linux/amd64 | 42,932,536 → 44,250,534 (+1,317,998, +3.1%) | 29,696,160 → 30,593,184 (+897,024, +3.0%) |
| darwin/arm64 | 41,662,322 → 42,970,274 (+1,307,952, +3.1%) | 28,536,418 → 29,426,738 (+890,320, +3.1%) |
| windows/amd64 | 43,581,952 → 44,918,784 (+1,336,832, +3.1%) | 30,532,096 → 31,450,112 (+918,016, +3.0%) |
Modules newly linked: connectrpc.com/connect v1.19.1, github.com/docker/secrets-engine/client v0.2.2, github.com/docker/secrets-engine/x v0.8.5, golang.org/x/mod v0.41.0. Protobuf and net/http were already in the binary.
Symbol sizes added, by package (linux/arm64, go tool nm -size), 291,792 bytes in all:
| package | bytes |
|---|---|
connectrpc.com/connect |
173,748 |
client, client/dockerhub, client/realms |
34,112 |
x/api/plugins/v1, pluginsv1connect |
19,712 |
x/api/resolver, resolver/v1, resolverv1connect |
12,584 |
x/secrets, x/realms, x/api, x/api/health/v1, healthv1connect |
13,728 |
golang.org/x/mod/semver |
2,560 |
| the CLI's own code using the client | 5,684 |
The rest of the growth is Go's per-function and per-type tables for that code (.gopclntab +410 KB, .go.type +163 KB, stripped) and DWARF.
Why a read-only consumer pays for the whole client
client.Newwires six RPCs:GetSecrets,Authorize,ListPlugins,EnablePlugin,DisablePlugin,GetVersion. Each is its ownconnect.NewClient[Req, Res]instantiation, with its own copy of the call path, plus the request and response types and their descriptors.- Every method of
*clientis linked, becauseNewreturns it as an interface. From-ldflags=-dumpdep:
So creating the plugin, version, and authorizer clients lazily inside the methods saves nothing. I measured it: +897 bytes.type:*github.com/docker/secrets-engine/client.client <UsedInIface> -> github.com/docker/secrets-engine/client.(*client).ListPlugins type:*github.com/docker/secrets-engine/client.client <UsedInIface> -> github.com/docker/secrets-engine/client.(*client).Version - Importing
clientalso importsx/api/plugins/v1andx/api/health/v1, whose descriptors are registered at init.
Option 1: a read-only client package
Add a package (name open, for example client/hub) that connects to a socket path and returns a dockerhub.ClientAuth, built on resolver.NewResolverClient only:
- It does not import
x/api/plugins/v1orx/api/health/v1, and does not set up the authorizer. ConnectErrorand its classification move to a leaf package, with an alias inclient, so the new package can still return*ConnectErrorfor "not running", "permission denied", and "timeout".client.New,PluginManagement, andVersionstay as they are, so existing users are not affected.
Measured on linux/arm64: +593,359 unstripped / +393,216 stripped, which saves 53% / 54%.
For comparison, removing only Version and plugin management from client itself (keeping Authorize) gives +767,243 / +524,288, a 39% saving, and breaks existing PluginManagement users. Authorize alone accounts for about 170 KB unstripped.
Option 2: GetSecrets without connect
In that package, call /resolver.v1.ResolverService/GetSecrets with the Connect protocol's unary encoding over net/http. The engine keeps serving with connect. A request is a POST with Content-Type: application/proto, Connect-Protocol-Version: 1, and a protobuf body. A non-200 response carries a JSON {"code", "message"}, mapped to secrets.ErrNotFound and secrets.ErrAccessDenied as resolver.NewResolverClient does today.
The prototype is about 60 lines:
func (r resolverClient) GetSecrets(ctx context.Context, pattern secrets.Pattern) ([]secrets.Envelope, error) {
body, err := proto.Marshal(resolverv1.GetSecretsRequest_builder{Pattern: proto.String(pattern.String())}.Build())
if err != nil {
return nil, err
}
req, err := http.NewRequestWithContext(ctx, http.MethodPost, "http://unix"+resolverv1connect.ResolverServiceGetSecretsProcedure, bytes.NewReader(body))
if err != nil {
return nil, err
}
req.Header.Set("Content-Type", "application/proto")
req.Header.Set("Connect-Protocol-Version", "1")
resp, err := r.httpClient.Do(req)
if err != nil {
return nil, err
}
defer resp.Body.Close()
data, err := io.ReadAll(io.LimitReader(resp.Body, 4<<20))
if err != nil {
return nil, err
}
if resp.StatusCode != http.StatusOK {
var e struct {
Code string `json:"code"`
Message string `json:"message"`
}
_ = json.Unmarshal(data, &e)
switch e.Code {
case "not_found":
return nil, secrets.ErrNotFound
case "permission_denied":
return nil, secrets.ErrAccessDenied
}
return nil, fmt.Errorf("secrets engine: %s: %s %s", resp.Status, e.Code, e.Message)
}
var out resolverv1.GetSecretsResponse
if err := proto.Unmarshal(data, &out); err != nil {
return nil, err
}
return toEnvelopes(out.GetEnvelopes()), nil
}
toEnvelopes is the same conversion resolver.NewResolverClient does today.
The prototype builds the path from a string literal, because resolverv1connect imports connect. The real package would need the procedure path as a constant outside resolverv1connect.
Measured on linux/arm64 with options 1 and 2: +195,400 unstripped / +131,072 stripped, which saves 85% / 85%. What remains is dockerhub, x/secrets, x/api, the resolver/v1 messages, and golang.org/x/mod/semver.
Trade-offs: secrets-engine maintains the unary request and error mapping for this path, and it has no connect interceptors. It needs tests against the connect server in this repository.
Verification
Both prototypes ran in the Docker CLI against a standalone engine on macOS (5 runs per case, 1 for the hung engine):
- Docker Hub and
dhi.iopulls sent the session credentials. - Signed out, a socket with nothing listening, and no socket all fell back to the CLI's own credentials without a warning.
- A hung engine fell back after the CLI's 5 s lookup timeout, with a warning.
How this was measured
docker was built from the same tree with go build -overlay, -tags grpcnotrace, CGO_ENABLED=0, and Go 1.27.1; stripped means -ldflags "-s -w". Stripped sizes move in 64 KiB steps. Release builds also use cgo and -buildmode=pie, which these builds did not.
Related: server-side connect code in clients
connect links 103 server-side symbols (19,584 bytes) into a client-only program. connect.clientConfig is stored in an interface value, which makes the protocol interface's NewHandler method reachable:
type:connectrpc.com/connect.clientConfig <UsedInIface> -> type:connectrpc.com/connect.protocol <UsedInIface>
type:*connectrpc.com/connect.protocolConnect <UsedInIface> -> connectrpc.com/connect.(*protocolConnect).NewHandler
That is for connectrpc, not this repository; option 2 avoids it.
- Linguagem predominante
- Go
- Estrelas
- 97
- Forks
- 23
- Merge médio
- 2h 1min
- PRs com merge (30d)
- 33
Preparar o ambiente
- Inclui um Dockerfile ou arquivo Docker Compose
- Sem modelo de pull request
- Ler o guia de contribuição
Primeiros passos
- Leia a issue inteira e depois o guia de contribuição do projeto.
- Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
- Faça um fork do repositório e trabalhe em uma branch.
- Abra um pull request que referencie o número da issue.
Mais de docker/secrets-engine
-
Dificuldade 3/5 1-2 dias Facilidade para iniciantes 57/100
docker/secrets-engine#739 ·
Mantenedores costumam responder em até 1 dia
-
Non-deterministic wildcard secret selectionTalvez livre de novo Um pull request para esta issue foi fechado sem ser mesclado. Aberta
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 35/100
docker/secrets-engine#723 · 2 comentários ·
Mantenedores costumam responder em até 1 dia
-
Dificuldade 4/5 3-5 dias Facilidade para iniciantes 35/100
docker/secrets-engine#682 · 1 comentário ·
Mantenedores costumam responder em até 1 dia
-
Dificuldade 5/5 Mais de uma semana Facilidade para iniciantes 35/100
docker/secrets-engine#676 · 2 reações ·
Mantenedores costumam responder em até 1 dia
-
Dificuldade 4/5 3-5 dias Facilidade para iniciantes 48/100
docker/secrets-engine#665 ·
Mantenedores costumam responder em até 1 dia
Todas as issues de docker/secrets-engine
Issues semelhantes
-
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 78/100
IBM-Cloud/terraform-provider-ibm#7115 ·
Mantenedores costumam responder em até 2 dias
-
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 86/100
fluxcd/notification-controller#1441 ·
Mantenedores costumam responder em até 1 dia
-
bug
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 66/100
volcengine/volcengine-go-sdk#163 · 1 comentário ·
-
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 76/100
OneBusAway/maglev#2145 ·
Mantenedores costumam responder em até 2 dias
-
priority/medium status/ready type/chore
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 84/100
discobox-ai/discobox#1357 ·
Mantenedores costumam responder em até 1 dia