Files
Enterprise-Knowledge-Base/cli/cmd/chat/chat.go
T
nullkey e236be1ced fix(cli): correct KB id detection, SSE terminal-frame, and CI test isolation
Three defects surfaced during end-to-end RAG verification — the first two
block real chat usage, the third makes Linux CI flaky:

1. KB id detection — `IsKBID` was checking
   `strings.HasPrefix(s, "kb_")`, but WeKnora generates KB ids as bare
   UUIDs (internal/types/knowledge_base.go: `uuid.New().String()` stored
   in a `varchar(36)` column). Real ids therefore fell through to the
   name-resolution path:

     $ weknora chat ... --kb a32a63ff-fb36-4874-bcaa-30f48570a694
     Error: knowledge base not found: a32a63ff-...

   Switched the discriminator to a UUID regex
   (`^[0-9a-fA-F]{8}-…-[0-9a-fA-F]{12}$`). KB names are arbitrary
   user-supplied strings, so the canonical 8-4-4-4-12 form is an
   unambiguous signal. Mirrors gcloud `--project`'s id-vs-name detection.

2. SSE terminal-frame — the accumulator's `Append` was gating
   finalization on `r.Done`, but the server's KnowledgeQAStream protocol
   emits a leading `agent_query` frame with `done=true` to deliver
   session + message metadata *before* the answer fragments arrive:

     event: message
     data: {"response_type":"agent_query","content":"","done":true,…}

     event: message
     data: {"response_type":"answer","content":"你好","done":false}
     …
     event: message
     data: {"response_type":"complete","content":"","done":true}

   The accumulator therefore flipped to `finished=true` on frame #1 and
   discarded every subsequent answer fragment — `weknora chat … --json`
   returned `answer: ""` even though the LLM reported completion_tokens
   > 0. Fixed: terminate only on `response_type == complete`.
   References still captured opportunistically (they may arrive on a
   dedicated `references` event before the terminator).

3. doctor credential_storage CI isolation — the check probes the real
   OS keyring via `secrets.NewBestEffortStore()`: present on macOS dev
   machines → StatusOK; absent on Linux CI runners without libsecret /
   Gnome-Keyring → StatusWarn ("falling back to file store"). That
   host-dependence was leaking into two test classes that assumed
   StatusOK:

     * cmd/doctor/doctor_test.go: TestDoctor_AllOK and
       TestDoctor_NoConfig_StillRunsCredentialStorage already had a
       withCredStoreFactory seam but didn't use it. Added the pin.

     * acceptance/contract/envelope_test.go: doctor.success_offline
       and doctor.error_network golden cases. The contract test runs
       through the cobra tree in-process and shares cmd/doctor's
       package-level credStoreFactory var — but couldn't reach it
       because the existing seam was unexported.

   Fix: export `doctor.SetCredStoreFactoryForTest(fn) (restore func())`
   for out-of-package tests; acceptance/contract/helpers_test.go adds
   a TestMain that pins the factory to a MemStore-returning closure
   for the whole suite (MemStore is neither *FileStore nor a real
   keyring, so doctor's type-switch hits StatusOK). Production stays
   at secrets.NewBestEffortStore — only the test hook is now reachable
   from across packages.

Test fixtures and goldens that used the old `kb_xxx` literals or
`Done: true` terminators were rewritten to use real UUIDs and
`ResponseType: ResponseTypeComplete` respectively. Per-command --help
text and Long descriptions / Examples now show a UUID rather than
`kb_…` so users see the correct shape from the start. New
TestAccumulator_IgnoresAgentQueryDone pins the SSE terminator bug so
it can't regress.

Tests: 24 cli packages green on macOS dev + Linux/macOS/Windows CI
matrix. Verified end-to-end against a live WeKnora server: `weknora
chat "..." --kb <UUID> --no-stream --json` returns the full LLM answer
in the envelope, live token streaming in TTY mode works, and the
credential_storage check renders deterministic envelopes across hosts.
2026-05-12 13:20:42 +08:00

279 lines
10 KiB
Go

// Package chat implements `weknora chat <text>` — the streaming RAG answer
// entry point.
//
// Two output modes share a single SDK call:
//
// - Stream mode (TTY + no --no-stream + no --json): write each
// StreamResponse.Content fragment directly to iostreams.IO.Out as it
// arrives, then print a footer with knowledge references. This is the
// "feels alive" UX a human typing in a terminal expects.
//
// - Accumulate mode (non-TTY, --no-stream, or --json): buffer every
// fragment via sse.Accumulator and emit a single envelope (or a single
// plain-text answer + references block) once Done. Agents and pipes
// get a deterministic single record to parse.
//
// The SDK's KnowledgeQAStream callback contract is invoked sequentially on
// one goroutine, so neither mode needs locking. The runChat core takes a
// chatService interface so tests inject a fake without standing up a real
// SSE server.
package chat
import (
"context"
"errors"
"fmt"
"io"
"strings"
"github.com/spf13/cobra"
"github.com/Tencent/WeKnora/cli/internal/cmdutil"
"github.com/Tencent/WeKnora/cli/internal/format"
"github.com/Tencent/WeKnora/cli/internal/iostreams"
"github.com/Tencent/WeKnora/cli/internal/sse"
sdk "github.com/Tencent/WeKnora/client"
)
// Options captures one `weknora chat` invocation.
type Options struct {
Query string
KBID string
SessionID string
NoStream bool
JSONOut bool
}
// chatService is the narrow SDK surface this command depends on. *sdk.Client
// satisfies it; tests substitute a fake. Compile-time check is at the bottom
// of this file.
type chatService interface {
CreateSession(ctx context.Context, req *sdk.CreateSessionRequest) (*sdk.Session, error)
KnowledgeQAStream(ctx context.Context, sessionID string, req *sdk.KnowledgeQARequest, cb func(*sdk.StreamResponse) error) error
}
// chatData is the success-envelope payload. Mirrors what an agent needs to
// continue a conversation: the answer text, retrieval references, and the
// session pointer to thread follow-ups.
type chatData struct {
Answer string `json:"answer"`
References []*sdk.SearchResult `json:"references"`
SessionID string `json:"session_id"`
AssistantMessageID string `json:"assistant_message_id,omitempty"`
KBID string `json:"kb_id"`
Query string `json:"query"`
}
// NewCmd builds `weknora chat <text>`.
func NewCmd(f *cmdutil.Factory) *cobra.Command {
opts := &Options{}
cmd := &cobra.Command{
Use: `chat <text>`,
Short: "Ask a streaming RAG question against a knowledge base",
Long: `Send a query to the WeKnora knowledge-chat endpoint and stream the
answer back. By default a fresh session is created on first invocation; pass
--session-id to continue an existing conversation.
Modes:
TTY (default): live token streaming + reference footer
Pipe / --no-stream / --json: buffered, emitted once on completion`,
Example: ` weknora chat "What is RRF?" --kb a32a63ff-fb36-4874-bcaa-30f48570a694
weknora chat "Summarise this design doc" --kb my-kb --json
weknora chat "Continue?" --session-id sess_abc`,
Args: cobra.MinimumNArgs(1),
RunE: func(c *cobra.Command, args []string) error {
opts.Query = strings.TrimSpace(strings.Join(args, " "))
if opts.Query == "" {
return cmdutil.NewError(cmdutil.CodeInputInvalidArgument, "query argument cannot be empty")
}
kbID, err := f.ResolveKB(c)
if err != nil {
return err
}
opts.KBID = kbID
cli, err := f.Client()
if err != nil {
return err
}
return runChat(c.Context(), opts, cli)
},
}
cmd.Flags().String("kb", "", "Knowledge base UUID or name (overrides project link / env)")
cmd.Flags().StringVar(&opts.SessionID, "session-id", "", "Continue an existing chat session (skip auto-create)")
cmd.Flags().BoolVar(&opts.NoStream, "no-stream", false, "Buffer the full answer before printing (forces accumulate mode)")
cmd.Flags().BoolVar(&opts.JSONOut, "json", false, "Emit a single JSON envelope (implies --no-stream)")
return cmd
}
// runChat is the testable core: validate, ensure a session, dispatch the
// stream, and route output. Returns a typed error suitable for the envelope.
func runChat(ctx context.Context, opts *Options, svc chatService) error {
if opts.Query == "" {
return cmdutil.NewError(cmdutil.CodeInputInvalidArgument, "query argument cannot be empty")
}
if opts.KBID == "" {
// Defensive: the cobra layer resolves KB before runChat; this guards
// the direct-test entry point.
return cmdutil.NewError(cmdutil.CodeKBIDRequired, "kb id is required")
}
if svc == nil {
return cmdutil.NewError(cmdutil.CodeServerError, "chat: no SDK client available")
}
sessionID := opts.SessionID
autoCreated := false
if sessionID == "" {
sess, err := svc.CreateSession(ctx, &sdk.CreateSessionRequest{Title: "weknora chat"})
if err != nil {
// Map HTTP-shaped failures, but tag generic transport / unknown
// errors as session_create_failed so the dedicated hint fires.
code := cmdutil.ClassifyHTTPError(err)
if code == cmdutil.CodeNetworkError || code == cmdutil.CodeServerError {
code = cmdutil.CodeSessionCreateFailed
}
return cmdutil.Wrapf(code, err, "create chat session")
}
sessionID = sess.ID
autoCreated = true
}
// Decide output mode. Stream mode requires:
// 1. an interactive stdout (tty)
// 2. no --no-stream
// 3. no --json (envelope is single-record by definition)
streamMode := iostreams.IO.IsStdoutTTY() && !opts.NoStream && !opts.JSONOut
// Surface the auto-created session ID up-front so a user who hits ^C
// mid-stream still has the pointer to resume — no need to scroll back
// past tokens. Skipped in JSON mode (it ends up in the envelope) and
// when the caller already supplied --session-id.
if autoCreated && !opts.JSONOut {
fmt.Fprintf(iostreams.IO.Err, "session: %s (use --session-id to continue)\n", sessionID)
}
req := &sdk.KnowledgeQARequest{
Query: opts.Query,
KnowledgeBaseIDs: []string{opts.KBID},
AgentEnabled: false,
WebSearchEnabled: false,
Channel: "api",
}
acc := &sse.Accumulator{}
cb := func(r *sdk.StreamResponse) error {
if streamMode && r != nil && r.Content != "" {
// Best-effort write; if stdout dies the SDK will surface the
// error on the next iteration. No need to bail early.
_, _ = iostreams.IO.Out.Write([]byte(r.Content))
}
acc.Append(r)
return nil
}
streamErr := svc.KnowledgeQAStream(ctx, sessionID, req, cb)
if streamErr != nil {
// Re-surface the auto-created session id on failure so a user who
// missed the start-of-stream notice (it scrolls past mid-stream
// tokens, especially on ^C) can still recover with --session-id.
// Skipped in JSON mode — the envelope carries it in .data.session_id.
if autoCreated && !opts.JSONOut {
fmt.Fprintf(iostreams.IO.Err, "session: %s (resume with --session-id %s)\n", sessionID, sessionID)
}
// Context cancelled (Ctrl-C) → user-aborted, exit 130 lineage.
if errors.Is(streamErr, context.Canceled) || errors.Is(ctx.Err(), context.Canceled) {
return cmdutil.Wrapf(cmdutil.CodeUserAborted, streamErr, "chat cancelled")
}
// Stream began (we observed at least one event) but never reached a
// terminal Done frame: typed as sse_stream_aborted so the hint
// nudges the user toward --no-stream / retry.
if acc.Result() != "" && !acc.Done() {
return cmdutil.Wrapf(cmdutil.CodeSSEStreamAborted, streamErr, "stream aborted before completion")
}
// Pre-stream HTTP / transport failure: route through the canonical
// classifier so 401 / 404 / 5xx still surface their specific codes.
return cmdutil.Wrapf(cmdutil.ClassifyHTTPError(streamErr), streamErr, "knowledge qa stream")
}
// SDK returned nil but we never saw a Done event — server closed the
// connection cleanly mid-stream. Treat as aborted so the user sees the
// truncation rather than a silent partial answer. Includes the empty-body
// case (Done frame never arrived AND no content): better to surface the
// abort than emit ok=true with answer="" — agents can't distinguish the
// model genuinely had nothing to say from the stream getting cut.
if !acc.Done() {
return cmdutil.NewError(cmdutil.CodeSSEStreamAborted, "stream ended without a terminal event")
}
answer := acc.Result()
references := acc.References
if opts.JSONOut {
// Prefer the SDK-echoed session id (acc.SessionID) but fall back to
// our local sessionID — agents must always see a usable pointer.
sid := acc.SessionID
if sid == "" {
sid = sessionID
}
data := chatData{
Answer: answer,
References: references,
SessionID: sid,
AssistantMessageID: acc.AssistantMessageID,
KBID: opts.KBID,
Query: opts.Query,
}
return cmdutil.NewJSONExporter().Write(iostreams.IO.Out, format.Success(data, &format.Meta{KBID: opts.KBID}))
}
// Human / non-JSON paths: streaming mode already wrote the answer body
// via the callback, so we only need to render the trailing references
// (and a closing newline). Accumulate + non-JSON writes the answer here
// for the first time.
out := iostreams.IO.Out
if streamMode {
// Ensure the answer line ends cleanly before the references footer.
if !strings.HasSuffix(answer, "\n") {
fmt.Fprintln(out)
}
} else {
fmt.Fprint(out, answer)
if !strings.HasSuffix(answer, "\n") {
fmt.Fprintln(out)
}
}
renderReferences(out, references)
return nil
}
// renderReferences prints a compact human-readable references block.
// Skipped entirely when the server returned no references — agent-friendly
// silence beats an empty banner.
func renderReferences(w io.Writer, refs []*sdk.SearchResult) {
if len(refs) == 0 {
return
}
fmt.Fprintln(w)
fmt.Fprintln(w, "──── References ────")
for i, r := range refs {
if r == nil {
continue
}
title := r.KnowledgeTitle
if title == "" {
title = r.KnowledgeFilename
}
if title == "" {
title = r.KnowledgeID
}
fmt.Fprintf(w, "[%d] %s", i+1, title)
if r.Score > 0 {
fmt.Fprintf(w, " score=%.3f", r.Score)
}
fmt.Fprintln(w)
}
}
// compile-time check: the production SDK client implements chatService.
var _ chatService = (*sdk.Client)(nil)