Files
ragflow/internal/agent/canvas/compile.go
Zhichang Yu 12787996d1 feat(agent): Go ingestion pipeline progress mirroring and DeepDOC parser hardening (#16795)
feat(ingestion): mirror Go pipeline progress into the document table;
harden resume guards
- pipeline: bind the owning document via WithDocumentID; after each
TrackProgress event aggregate ingestion_task_log progress and mirror
progress/run/progress_msg back into the document table, so GET
/api/v1/datasets/{dataset_id}/documents reflects live Go pipeline
progress without a bespoke endpoint.
- canvas: extend the S3 resume guard to reject legacy no-op nodes (e.g.
ExitLoop) so component_total equals the count of progress-reporting
components and the aggregate percent can reach 100%.
- runtime/canvas: route progress through TrackProgress; add interrupt
test coverage (r3_interrupt_test.go).
- dao/entity: add IngestionTask.DocumentID column and AggregateProgress
support used by the mirror; IngestionTaskLog keeps a Checkpoint column
alongside the progress fields.

feat(deepdoc): cache DocAnalyzer inference results in Redis (1h TTL)
- Redis-backed DocAnalyzerCache decorator over inference.Client; cache
key = "ddoc:cache:<method>:" + sha256 of the JPEG-encoded image bytes
(deterministic).
- TTL = 1h; hits skip the inner HTTP call and return cached JSON; inner
errors are not cached.

refactor(deepdoc): align figure cropping with Python cropout + bounded
page caches
- CropSectionByDLA mirrors Python cropout: best-overlap DLA
figure/equation region, fallback to section bbox per page, vertical
concat on gray background.
- sliding-window page-image cache bounds peak memory to the recent
window instead of the whole PDF.
- rename DLADebug -> DLARegions across parser/chunker/tests.

refactor(parser): drop lib_type selector; align NewXxxParser with
NewPDFParser
- remove config["lib_type"] lookup and the libType param/field/switch
from all nine constructors; surface the CGO-required error at
ParseWithResult time instead of construction time; drop resolveLibType,
its test, and the four lib_type constants.

feat(utility): add a reusable workerpool for bounded concurrent
execution
- internal/utility/workerpool.go (+ tests).

refactor: translate Chinese prose comments to English in non-harness Go
files.

chore: upgrade github.com/cloudwego/eino from v0.9.9 to v0.9.12.
2026-07-10 10:36:10 +08:00

313 lines
13 KiB
Go

// Package canvas — compile entry.
//
// Compile turns a Canvas (DSL) into a CompiledCanvas: a compiled
// compose.Runnable plus the CheckPointID used at this compile. The
// compile-time wiring (state pre/post handlers, checkpoint store,
// serializer) is configured here; the actual run path lives in
// runner.go and the HTTP handler / SSE / RunTracker are wired in
// internal/service and internal/handler.
package canvas
import (
"context"
"fmt"
"strings"
"github.com/cloudwego/eino/compose"
"go.uber.org/zap"
"ragflow/internal/common"
)
// CheckPointStore is the minimal interface Compile needs at compile time.
// RedisCheckPointStore satisfies this; tests can pass any in-memory
// implementation. Matches eino's compose.CheckPointStore (an alias for
// core.CheckPointStore) and adds a Delete method.
type CheckPointStore interface {
Get(ctx context.Context, id string) ([]byte, bool, error)
Set(ctx context.Context, id string, payload []byte) error
Delete(ctx context.Context, id string) error
}
// StateSerializer is the minimal interface Compile needs. The
// CanvasStateSerializer in this package satisfies this. Mirrors
// eino's compose.Serializer (Marshal/Unmarshal, no context).
type StateSerializer interface {
Marshal(v any) ([]byte, error)
Unmarshal(data []byte, v any) error
}
// CompiledCanvas is the compiled runtime representation of a Canvas DSL.
// Workflow is the eino Runnable; CheckPointID is the eino checkpoint
// identifier for this compile.
type CompiledCanvas struct {
Workflow compose.Runnable[map[string]any, map[string]any]
CheckPointID string
}
// CompileOptions bundles the optional collaborators the compile entry needs.
// All fields are optional; nil/zero means "skip that wire".
type CompileOptions struct {
Store CheckPointStore
Serializer StateSerializer
// InterruptBefore / InterruptAfter are passed straight through to
// compose.WithInterruptBeforeNodes / WithInterruptAfterNodes.
InterruptBefore []string
InterruptAfter []string
// CheckPointID is the stable eino checkpoint identifier. Unlike
// eino's compose.WithCheckPointID (a run-time Option applied at
// Workflow.Invoke), this is a compile-time descriptor: Compile cannot
// call compose.WithCheckPointID (the option type is wrong for a
// GraphCompileOption), so it only records the id on the returned
// CompiledCanvas — the caller threads it to Invoke. Use a stable,
// per-task value (e.g. taskID) so re-running the same task hits the
// same Redis checkpoint (agent:cp:{id}). When empty,
// CompiledCanvas.CheckPointID stays empty and the caller must supply
// its own id (or omit it for a fresh per-run checkpoint).
CheckPointID string
// InterruptAfterNonTerminal, when true, makes Compile compute the
// non-terminal node ids internally (components with out-degree > 0)
// and register compose.WithInterruptAfterNodes on them — the caller
// does not enumerate them. UserFillUp nodes are excluded (see §4.2.b)
// because they already emit their own compose.Interrupt;
// double-registering the same node for two interrupt sources would
// break resume. Terminal nodes (no downstream) are excluded so the
// graph does not pause on completion and force an extra, needless
// ResumeWithData round.
InterruptAfterNonTerminal bool
}
// CompileOption mutates a CompileOptions before the compile runs.
type CompileOption func(*CompileOptions)
// WithCheckPointStore attaches a CheckPointStore to the compile.
func WithCheckPointStore(s CheckPointStore) CompileOption {
return func(o *CompileOptions) { o.Store = s }
}
// WithStateSerializer attaches a StateSerializer to the compile.
func WithStateSerializer(s StateSerializer) CompileOption {
return func(o *CompileOptions) { o.Serializer = s }
}
// WithInterruptBefore configures compose.WithInterruptBeforeNodes.
func WithInterruptBefore(nodes []string) CompileOption {
return func(o *CompileOptions) { o.InterruptBefore = nodes }
}
// WithInterruptAfter configures compose.WithInterruptAfterNodes.
func WithInterruptAfter(nodes []string) CompileOption {
return func(o *CompileOptions) { o.InterruptAfter = nodes }
}
// WithCheckPointID sets the stable checkpoint id recorded on the returned
// CompiledCanvas. Unlike eino's compose.WithCheckPointID (a run-time
// Option), this is a compile-time descriptor: Compile stores the id so the
// caller can pass it to Workflow.Invoke. Pass a stable, per-task value
// (e.g. taskID) so re-running the same task loads the same Redis
// checkpoint (agent:cp:{id}).
func WithCheckPointID(id string) CompileOption {
return func(o *CompileOptions) { o.CheckPointID = id }
}
// WithInterruptAfterNonTerminalCpn registers an after-node interrupt on
// every non-terminal component (out-degree > 0) automatically. The set is
// computed inside Compile from the Canvas topology, so callers can't pass
// the wrong list (e.g. all cpnIDs, which would also interrupt terminal
// nodes and force an extra needless ResumeWithData round). UserFillUp
// nodes are excluded (§4.2.b). See computeNonTerminalCpnIDs for the exact
// selection rules.
func WithInterruptAfterNonTerminalCpn() CompileOption {
return func(o *CompileOptions) { o.InterruptAfterNonTerminal = true }
}
// Compile builds the eino Workflow from the Canvas and returns the
// compiled Runnable. State pre/post handlers are wired inside BuildWorkflow
// (see scheduler.go). Checkpoint store + serializer are wired here as
// compile-time options (compose.GraphCompileOption).
//
// IMPORTANT: eino v0.9.2 option split (plan §2.6 fix):
//
// WithStatePreHandler / WithStatePostHandler -> GraphAddNodeOpt (NODE option)
// WithCheckPointStore / WithSerializer -> GraphCompileOption
//
// Mixing them up makes the call fail to compile. We do not accept
// GraphCompileOption from the caller directly — that would let them pass
// the wrong option type. The CompileOption indirection keeps the
// GraphCompileOption surface inside this file.
func Compile(ctx context.Context, c *Canvas, opts ...CompileOption) (*CompiledCanvas, error) {
cfg := CompileOptions{}
for _, o := range opts {
o(&cfg)
}
// Decoder-boundary guard: if the caller handed us a Canvas
// whose `components` still contains LoopItem or IterationItem
// entries, they bypassed dsl.NormalizeForCanvas (the only
// supported decoder path). The fold step never ran, so the
// runtime will see legacy child names and the workflow below
// will misbehave. Surface a visible stderr warning so the
// regression is observable — this is intentionally a log
// rather than a panic, because internal drivers (tests,
// fixtures) may exercise the path with raw components.
if c != nil {
var n int
for _, comp := range c.Components {
switch strings.ToLower(comp.Obj.ComponentName) {
case "loopitem", "iterationitem", "iteration":
n++
}
}
if n > 0 {
common.Info("canvas: Compile received Canvas with legacy LoopItem/IterationItem/Iteration nodes; this path bypassed dsl.NormalizeForCanvas — the fold step is not applied", zap.Int("n", n))
}
}
// S3 (plan §4.2.b 方案 A): ingestion resume mode forbids UserFillUp
// nodes. A UserFillUp node emits its own compose.Interrupt
// (wait-for-user); the pipeline resume loop (pipeline.go runResumable)
// classifies every interrupt via IsInterruptError and auto-resumes with
// nil data — so a UserFillUp pause would be silently skipped instead of
// waiting for a human. Reject at compile time so the mis-classification
// can never occur. The non-terminal-after filter (computeNonTerminalCpnIDs)
// already keeps UserFillUp out of the after-node set; this is a hard
// guard layered on top (plan §8 step 5). Checked before BuildWorkflow so
// the guard fires on DSL content regardless of whether the graph builds.
//
// The same guard also forbids legacy no-op nodes (e.g. "ExitLoop", see
// legacyNoOpNames / isLegacyNoOp). A no-op node routes to an echo body
// that never emits TrackProgress, so it would still be counted in
// ingestion_task.component_total yet never report progress — leaving the
// aggregate percent permanently below 100% (plan §8 "known
// inconsistency"). Forbidding it keeps component_total == "components
// that report progress", so percent can reach 100% and the
// resume/percent invariant holds. Ingestion DSLs must consist solely of
// progress-reporting components.
if cfg.InterruptAfterNonTerminal && c != nil {
var bad []string
bad = append(bad, AutoDiscoverUserFillUpIDs(c)...)
for cpnID, comp := range c.Components {
if isLegacyNoOp(comp.Obj.ComponentName) {
bad = append(bad, cpnID)
}
}
if len(bad) > 0 {
return nil, fmt.Errorf("canvas: Compile: WithInterruptAfterNonTerminalCpn forbids UserFillUp/legacy-no-op nodes %v (plan §4.2.b): ingestion has no user to fill up and no-op nodes do not report progress, breaking the resume/percent invariant", bad)
}
}
wf, err := BuildWorkflow(ctx, c)
if err != nil {
return nil, fmt.Errorf("canvas: build workflow: %w", err)
}
compileOpts := make([]compose.GraphCompileOption, 0, 4)
if cfg.Store != nil {
// eino's compose.WithCheckPointStore expects compose.CheckPointStore
// (no Delete). Our CheckPointStore adds Delete; pass an adapter
// that drops it. RunTracker doesn't call Delete on this
// path — it deletes the agent:cp:* key via a separate Redis call.
compileOpts = append(compileOpts, compose.WithCheckPointStore(checkPointAdapter{cfg.Store}))
}
if cfg.Serializer != nil {
compileOpts = append(compileOpts, compose.WithSerializer(serializerAdapter{cfg.Serializer}))
}
if len(cfg.InterruptBefore) > 0 {
compileOpts = append(compileOpts, compose.WithInterruptBeforeNodes(cfg.InterruptBefore))
}
// Merge the caller-supplied InterruptAfter list with the
// internally-computed non-terminal set (when requested). The
// computed set excludes UserFillUp nodes (§4.2.b); the caller list is
// trusted verbatim. Dedupe so a node isn't registered twice in one
// WithInterruptAfterNodes call.
after := append([]string{}, cfg.InterruptAfter...)
if cfg.InterruptAfterNonTerminal {
after = append(after, computeNonTerminalCpnIDs(c)...)
}
after = dedupeStrings(after)
if len(after) > 0 {
compileOpts = append(compileOpts, compose.WithInterruptAfterNodes(after))
}
runnable, err := wf.Compile(ctx, compileOpts...)
if err != nil {
return nil, fmt.Errorf("canvas: eino compile: %w", err)
}
return &CompiledCanvas{Workflow: runnable, CheckPointID: cfg.CheckPointID}, nil
}
// computeNonTerminalCpnIDs returns the cpnIDs of every component with at
// least one downstream edge (out-degree > 0). These are the nodes the
// "interrupt-after-node" resume strategy must pause on: any node that has
// work after it.
//
// Terminal nodes (no downstream) are intentionally excluded — interrupting
// them would make Invoke return an interrupt error instead of a completion,
// and force an extra needless ResumeWithData round before the graph truly
// finishes.
//
// UserFillUp nodes are excluded (§4.2.b): they already emit their own
// compose.Interrupt and must not be registered for a second, conflicting
// interrupt source. Double-registering the same node breaks resume.
func computeNonTerminalCpnIDs(c *Canvas) []string {
if c == nil {
return nil
}
exclude := make(map[string]bool, len(c.Components))
for _, id := range AutoDiscoverUserFillUpIDs(c) {
exclude[id] = true
}
var ids []string
for cpnID, comp := range c.Components {
if exclude[cpnID] {
continue
}
if len(comp.Downstream) > 0 {
ids = append(ids, cpnID)
}
}
return ids
}
// dedupeStrings returns in with duplicate entries removed, preserving
// first-seen order. Used to merge the computed non-terminal set with the
// caller-supplied InterruptAfter list without registering a node twice in
// the same WithInterruptAfterNodes call.
func dedupeStrings(in []string) []string {
if len(in) == 0 {
return in
}
seen := make(map[string]bool, len(in))
out := make([]string, 0, len(in))
for _, s := range in {
if seen[s] {
continue
}
seen[s] = true
out = append(out, s)
}
return out
}
// checkPointAdapter drops the Delete method that compose.CheckPointStore
// does not declare. The RedisCheckPointStore in this package has
// Delete; eino
// doesn't, so the adapter is a thin passthrough.
type checkPointAdapter struct{ inner CheckPointStore }
func (a checkPointAdapter) Get(ctx context.Context, id string) ([]byte, bool, error) {
return a.inner.Get(ctx, id)
}
func (a checkPointAdapter) Set(ctx context.Context, id string, payload []byte) error {
return a.inner.Set(ctx, id, payload)
}
// serializerAdapter exposes the eino-shaped Serializer (Marshal/Unmarshal,
// no context). The CanvasStateSerializer in this package matches the
// same shape, so
// the adapter is a passthrough.
type serializerAdapter struct{ inner StateSerializer }
func (a serializerAdapter) Marshal(v any) ([]byte, error) { return a.inner.Marshal(v) }
func (a serializerAdapter) Unmarshal(b []byte, v any) error { return a.inner.Unmarshal(b, v) }