feat(knowledge): instrument postprocess subspans + polish timeline UI

Backend: summary, question, and graph-extract async tasks now record
real processing time as subspans hanging off the (closed) postprocess
stage, so the trace viewer no longer caps the postprocess row at the
~10 ms enqueue duration. Carries Attempt through SummaryGeneration /
QuestionGeneration / ExtractChunk payloads so cross-process workers
can resolve the right parent attempt.

Frontend: drawer now uses attach="body" so the secondary 820px detail
drawer escapes the 654px container of the main drawer; timeline
timestamps include date prefix (MM-DD, or YYYY-MM-DD across years);
"updated X ago" caption only shows during live polling.

Tests: 4 new cases covering postprocess subspan attaching under a
closed parent, missing-parent fallthrough, and Attempt JSON round-trip
on the three task payloads.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
wizardchen
2026-05-27 21:33:16 +08:00
committed by lyingbug
parent 5ffc755e76
commit bd029b6d19
11 changed files with 2415 additions and 537 deletions
+429 -17
View File
@@ -64,6 +64,35 @@ mermaid.initialize({
const props = defineProps(["visible", "details", "knowledgeType", "sourceInfo", "canEditKB", "parse_status"]);
const emit = defineEmits(["closeDoc", "getDoc", "questionDeleted"]);
const hasTimelineSpans = ref(false);
const timelineDrawerVisible = ref(false);
const timelineSummary = ref<{ totalMs: number; status: string; stageIndex: number; stageTotal: number; stageLabel: string }>({
totalMs: 0, status: '', stageIndex: 0, stageTotal: 0, stageLabel: '',
});
watch(() => props.details?.id, () => {
hasTimelineSpans.value = false;
timelineDrawerVisible.value = false;
timelineSummary.value = { totalMs: 0, status: '', stageIndex: 0, stageTotal: 0, stageLabel: '' };
});
function formatTimelineDuration(ms: number): string {
if (!ms || ms < 0) return '—';
if (ms < 1000) return `${Math.round(ms)}ms`;
if (ms < 60000) return `${(ms / 1000).toFixed(2)}s`;
const mins = Math.floor(ms / 60000);
const rem = ((ms % 60000) / 1000).toFixed(1);
return `${mins}m${rem}s`;
}
function openTimeline() {
timelineDrawerVisible.value = true;
}
function closeTimeline() {
timelineDrawerVisible.value = false;
}
marked.use({
breaks: true, // 启用单行换行转 <br>
gfm: true, // 启用 GitHub Flavored Markdown
@@ -778,7 +807,7 @@ const handleDetailsScroll = () => {
</script>
<template>
<div class="doc_content" ref="mdContentWrap">
<t-drawer :visible="visible" :zIndex="2000" :closeBtn="true" :footer="false" @close="handleClose">
<t-drawer :visible="visible" :zIndex="2000" size="654px" attach="body" :closeBtn="true" :footer="false" @close="handleClose">
<template #header>
<div class="drawer-header">
<span class="header-title">{{ getDisplayTitle() }}</span>
@@ -788,17 +817,94 @@ const handleDetailsScroll = () => {
</div>
</template>
<!-- 解析阶段时间线(仅在解析中/失败时显示) -->
<div
v-if="details.id"
class="parse_timeline_box"
<!-- 解析阶段时间线 trigger(点击打开二级抽屉) -->
<button
v-if="details.id && hasTimelineSpans"
type="button"
class="kp-trigger"
:class="['kp-trigger-' + (timelineSummary.status || 'unknown')]"
@click="openTimeline"
>
<span class="kp-trigger-status">
<span class="kp-trigger-dot" :class="['kp-trigger-dot-' + (timelineSummary.status || 'unknown')]" />
<span class="kp-trigger-label">{{ $t('knowledgeStages.title') }}</span>
</span>
<span class="kp-trigger-divider" />
<span class="kp-trigger-meta">
<template v-if="timelineSummary.totalMs > 0">
<span class="kp-trigger-meta-key">{{ $t('knowledgeStages.head.duration') }}</span>
<span class="kp-trigger-meta-val kp-trigger-mono">{{ formatTimelineDuration(timelineSummary.totalMs) }}</span>
</template>
<template v-else-if="timelineSummary.stageTotal > 0">
<span class="kp-trigger-meta-key">{{ $t('knowledgeStages.head.stage') }}</span>
<span class="kp-trigger-meta-val kp-trigger-mono">
{{ timelineSummary.stageIndex }}/{{ timelineSummary.stageTotal }}
</span>
<span class="kp-trigger-meta-key kp-trigger-meta-stage">· {{ timelineSummary.stageLabel }}</span>
</template>
</span>
<span class="kp-trigger-tail">
<span
v-if="timelineSummary.status === 'running' || timelineSummary.status === 'processing' || timelineSummary.status === 'pending'"
class="kp-trigger-live"
>
<span class="kp-trigger-live-dot" />
<span class="kp-trigger-live-text">LIVE</span>
</span>
<span class="kp-trigger-cta">{{ $t('knowledgeStages.viewTrace') }}</span>
<t-icon name="chevron-right" size="14px" class="kp-trigger-arrow" />
</span>
</button>
<!-- Hidden mount: keeps the timeline fetching data so trigger summary
stays live even before the user opens the secondary drawer. The
drawer itself shows a separate, fully-interactive instance. -->
<div class="kp-trigger-shadow" aria-hidden="true">
<KnowledgeProcessingTimeline
v-if="details.id"
:knowledge-id="details.id"
:parse-status="details.parse_status"
:compact="true"
@update:has-spans="hasTimelineSpans = $event"
@update:summary="timelineSummary = $event"
/>
</div>
<!-- 二级抽屉:完整 Langfuse-style waterfall -->
<t-drawer
:visible="timelineDrawerVisible"
:zIndex="2100"
size="820px"
attach="body"
:closeBtn="false"
:footer="false"
:header="false"
:showOverlay="true"
:closeOnOverlayClick="true"
placement="right"
class="kp-secondary-drawer"
@close="closeTimeline"
>
<div class="kp-drawer-shell">
<div class="kp-drawer-titlebar">
<div class="kp-drawer-titlebar-left">
<span class="kp-drawer-titlebar-kind">trace</span>
<span class="kp-drawer-titlebar-title">{{ $t('knowledgeStages.title') }}</span>
</div>
<button type="button" class="kp-drawer-titlebar-close" @click="closeTimeline" :aria-label="$t('knowledgeStages.close')">
<t-icon name="close" size="16px" />
</button>
</div>
<div class="kp-drawer-body">
<KnowledgeProcessingTimeline
v-if="details.id && timelineDrawerVisible"
:knowledge-id="details.id"
:parse-status="details.parse_status"
/>
</div>
</div>
</t-drawer>
<!-- 文件类型专属区域 -->
<div v-if="details.type === 'file'" class="doc_box">
<a :href="url" style="display: none" ref="down" :download="details.title"></a>
@@ -1015,18 +1121,12 @@ const handleDetailsScroll = () => {
<style scoped lang="less">
@import "./css/markdown.less";
:deep(.t-drawer .t-drawer__content-wrapper) {
width: min(654px, 85vw) !important; // 减少到85%视口宽度,给左侧留更多空间
max-width: 654px !important;
}
// 在小屏幕上进一步调整
@media (max-width: 768px) {
:deep(.t-drawer .t-drawer__content-wrapper) {
width: 90vw !important; // 小屏幕上使用90%宽度
max-width: none !important;
}
}
/* Drawer widths are now driven by the `:size` prop on each <t-drawer>
(see mainDrawerSize / timelineDrawerSize in <script>). CSS rules with
!important were removed because they fought each other across the
scoped/non-scoped boundary and had no clean specificity ordering in
dev mode (Vite injects scoped <style> tags later than non-scoped,
inverting prod). Inline width via the prop is unambiguous. */
// 代码块样式
:deep(.code-block-wrapper) {
@@ -1108,6 +1208,301 @@ const handleDetailsScroll = () => {
border-radius: 6px;
}
/* ============== Timeline trigger pill ==============
Designed to read like an information row in the doc drawer, not a
loud CTA. Soft 6px corners, neutral border, left strip carries the
status color so the pipeline state is glanceable without dominating
the layout. Mirrors the KBInfoPopover/manual-knowledge-editor card
feel that the rest of the app uses. */
.kp-trigger {
margin: 8px 0 16px;
width: 100%;
display: flex;
align-items: center;
gap: 12px;
padding: 10px 14px;
border: 1px solid var(--td-component-stroke);
border-radius: var(--td-radius-medium);
background: var(--td-bg-color-container);
color: var(--td-text-color-primary);
cursor: pointer;
font-family: var(--app-font-family);
text-align: left;
position: relative;
overflow: hidden;
transition: border-color 150ms ease, background 150ms ease, transform 100ms ease;
}
.kp-trigger::before {
content: "";
position: absolute;
left: 0;
top: 8px;
bottom: 8px;
width: 3px;
border-radius: 0 2px 2px 0;
background: var(--td-component-border);
transition: background 150ms ease;
}
.kp-trigger:hover {
background: var(--td-bg-color-secondarycontainer);
border-color: var(--td-component-border);
}
.kp-trigger:active { transform: translateY(1px); }
.kp-trigger-done::before,
.kp-trigger-completed::before { background: var(--td-success-color); }
.kp-trigger-failed::before { background: var(--td-error-color); }
.kp-trigger-running::before,
.kp-trigger-processing::before,
.kp-trigger-pending::before { background: var(--td-brand-color); }
.kp-trigger-unknown::before { background: var(--td-component-border); }
.kp-trigger-status {
display: inline-flex;
align-items: center;
gap: 8px;
flex-shrink: 0;
}
.kp-trigger-dot {
width: 8px;
height: 8px;
border-radius: 50%;
background: var(--td-text-color-placeholder);
flex-shrink: 0;
}
.kp-trigger-dot-done,
.kp-trigger-dot-completed { background: var(--td-success-color); }
.kp-trigger-dot-failed { background: var(--td-error-color); }
.kp-trigger-dot-running,
.kp-trigger-dot-processing,
.kp-trigger-dot-pending {
background: var(--td-brand-color);
animation: kpTriggerPulse 1.6s ease-in-out infinite;
}
.kp-trigger-dot-unknown { background: var(--td-text-color-placeholder); }
@keyframes kpTriggerPulse {
0%, 100% { box-shadow: 0 0 0 0 var(--td-brand-color-light); }
50% { box-shadow: 0 0 0 4px transparent; }
}
.kp-trigger-label {
font-size: 13px;
font-weight: 600;
color: var(--td-text-color-primary);
}
.kp-trigger-divider {
width: 1px;
height: 14px;
background: var(--td-component-stroke);
flex-shrink: 0;
}
.kp-trigger-meta {
display: inline-flex;
align-items: baseline;
gap: 6px;
flex: 1;
min-width: 0;
overflow: hidden;
white-space: nowrap;
text-overflow: ellipsis;
}
.kp-trigger-meta-key {
color: var(--td-text-color-placeholder);
font-size: 11px;
font-weight: 500;
text-transform: uppercase;
letter-spacing: 0.5px;
}
.kp-trigger-meta-stage {
text-transform: none;
letter-spacing: 0;
font-weight: 400;
color: var(--td-text-color-secondary);
margin-left: 2px;
}
.kp-trigger-meta-val {
color: var(--td-text-color-primary);
font-size: 12px;
font-weight: 500;
}
.kp-trigger-mono {
font-family: var(--app-font-family-mono);
font-size: 12px;
letter-spacing: 0;
}
.kp-trigger-tail {
display: inline-flex;
align-items: center;
gap: 8px;
flex-shrink: 0;
}
/* LIVE badge — pulses while the parser is mid-run, so the user knows
opening the trace will show fresh data. */
.kp-trigger-live {
display: inline-flex;
align-items: center;
gap: 4px;
padding: 2px 7px;
background: var(--td-brand-color-light);
color: var(--td-brand-color);
border-radius: var(--td-radius-default);
font-size: 10px;
font-weight: 600;
letter-spacing: 0.06em;
line-height: 1;
}
.kp-trigger-live-dot {
width: 5px;
height: 5px;
border-radius: 50%;
background: var(--td-brand-color);
animation: kpTriggerLivePulse 1.4s ease-in-out infinite;
}
.kp-trigger-live-text {
font-family: var(--app-font-family-mono);
}
@keyframes kpTriggerLivePulse {
0%, 100% { opacity: 1; transform: scale(1); }
50% { opacity: 0.4; transform: scale(0.8); }
}
.kp-trigger-cta {
font-size: 12px;
color: var(--td-text-color-secondary);
}
.kp-trigger:hover .kp-trigger-cta {
color: var(--td-brand-color);
}
.kp-trigger-arrow {
color: var(--td-text-color-placeholder);
transition: transform 150ms ease, color 150ms ease;
}
.kp-trigger:hover .kp-trigger-arrow {
color: var(--td-brand-color);
transform: translateX(2px);
}
/* Hidden mount keeps fetcher live without showing UI */
.kp-trigger-shadow {
display: none;
}
/* ============== Secondary drawer shell ============== */
.kp-drawer-shell {
display: flex;
flex-direction: column;
height: 100%;
width: 100%;
background: var(--td-bg-color-container);
/* Belt-and-suspenders: even if some inner element forgets a min-width:0
declaration in a flex chain, clip rather than overflow the drawer. */
overflow: hidden;
min-width: 0;
}
.kp-drawer-titlebar {
flex: 0 0 auto;
display: flex;
align-items: center;
justify-content: space-between;
padding: 14px 20px 12px;
border-bottom: 1px solid var(--td-component-stroke);
background: var(--td-bg-color-container);
}
.kp-drawer-titlebar-left {
display: flex;
align-items: center;
gap: 10px;
min-width: 0;
}
.kp-drawer-titlebar-kind {
font-family: var(--app-font-family-mono);
font-size: 10px;
text-transform: uppercase;
letter-spacing: 0.5px;
color: var(--td-text-color-placeholder);
background: var(--td-bg-color-secondarycontainer);
border-radius: var(--td-radius-default);
padding: 2px 6px;
line-height: 1.2;
}
.kp-drawer-titlebar-title {
font-size: 15px;
font-weight: 600;
color: var(--td-text-color-primary);
letter-spacing: -0.005em;
}
.kp-drawer-titlebar-close {
display: inline-flex;
align-items: center;
justify-content: center;
width: 28px;
height: 28px;
border: none;
background: transparent;
cursor: pointer;
color: var(--td-text-color-placeholder);
border-radius: var(--td-radius-default);
transition: background 150ms ease, color 150ms ease;
}
.kp-drawer-titlebar-close:hover {
background: var(--td-bg-color-secondarycontainer);
color: var(--td-text-color-primary);
}
.kp-drawer-body {
flex: 1 1 auto;
min-height: 0;
min-width: 0;
overflow: hidden;
/* Plain block layout — flex was unnecessary and risked the timeline
becoming a flex item with auto-sized basis, leaking past the drawer
bounds when its internal min-widths exceeded the parent. */
display: block;
position: relative;
}
.kp-drawer-body > :deep(.kp-timeline) {
width: 100%;
height: 100%;
}
/* Width is set via the :size prop on the <t-drawer>, not CSS — see the
<script> for mainDrawerSize / timelineDrawerSize. Only padding +
background are still tweaked here so the timeline fills the secondary
drawer cleanly, edge-to-edge. */
:deep(.kp-secondary-drawer .t-drawer__body) {
padding: 0 !important;
}
:deep(.kp-secondary-drawer .t-drawer__content) {
background: var(--td-bg-color-container);
}
// 文档摘要区域
.summary_box {
display: flex;
@@ -1514,3 +1909,20 @@ const handleDetailsScroll = () => {
margin-top: 12px;
}
</style>
<!-- Non-scoped padding/background overrides for the secondary drawer.
Width is now controlled via the :size prop on <t-drawer> (see
timelineDrawerSize in <script>) — that puts width on element.style
rather than fighting !important CSS rules. We only keep these
non-scoped rules because TDesign's default body padding and
content background need to be flushed for the timeline to fill
edge-to-edge. -->
<style lang="less">
.t-drawer.kp-secondary-drawer .t-drawer__body {
padding: 0 !important;
}
.t-drawer.kp-secondary-drawer .t-drawer__content {
background: var(--td-bg-color-container);
}
</style>
File diff suppressed because it is too large Load Diff
+36 -2
View File
@@ -394,21 +394,55 @@ export default {
attempt: 'Attempt {n}',
attemptLatest: 'Attempt {n} (latest)',
retry: 'Retry parsing',
refresh: 'Refresh now',
copy: 'Copy',
copyDetails: 'Copy details',
copied: 'Copied to clipboard',
close: 'Close',
live: 'LIVE',
liveTooltip: 'Parsing in progress, refreshes every 2s (next in {n}s)',
viewTrace: 'View trace',
justNow: 'just now',
secondsAgo: '{n}s ago',
minutesAgo: '{n}m ago',
noActivity: 'No parsing activity yet',
totalDuration: 'Total: {d}',
total: 'Total {d}',
head: {
duration: 'Duration',
stages: 'Stages',
stage: 'Stage',
status: 'Status',
attempt: 'Attempt',
updated: 'Updated',
},
tab: {
overview: 'Overview',
raw: 'Raw JSON',
},
detail: {
started: 'Started',
finished: 'Finished',
duration: 'Duration',
offset: 'Offset',
timing: 'Timing',
identity: 'Identity',
stageBreakdown: 'Stage breakdown',
stageOrder: 'Stage order',
childCount: 'Child spans',
kind: 'Kind',
status: 'Status',
name: 'Name',
input: 'Input',
output: 'Output',
metadata: 'Metadata',
error: 'Error',
showJson: 'Show JSON',
hideJson: 'Hide JSON',
empty: 'No data',
inProgress: 'In progress',
elapsed: 'Elapsed',
placeholderHint: 'This stage has no detailed span record; only the inferred state is shown.',
showJson: 'Expand JSON',
hideJson: 'Collapse JSON',
},
stage: {
docreader: 'Document parsing',
+36 -2
View File
@@ -400,21 +400,55 @@ export default {
attempt: "{n}번째 시도",
attemptLatest: "{n}번째 시도 (최신)",
retry: "다시 파싱",
refresh: "지금 새로고침",
copy: "복사",
copyDetails: "세부정보 복사",
copied: "클립보드에 복사됨",
close: "닫기",
live: "LIVE",
liveTooltip: "파싱 중, 2초마다 자동 새로고침 ({n}s 후)",
viewTrace: "Trace 보기",
justNow: "방금",
secondsAgo: "{n}초 전",
minutesAgo: "{n}분 전",
noActivity: "파싱 활동 없음",
totalDuration: "총 소요시간: {d}",
total: "총 {d}",
head: {
duration: "소요시간",
stages: "단계",
stage: "단계",
status: "상태",
attempt: "시도",
updated: "갱신",
},
tab: {
overview: "개요",
raw: "원본 JSON",
},
detail: {
started: "시작",
finished: "종료",
duration: "소요시간",
offset: "시작 오프셋",
timing: "타이밍",
identity: "식별 정보",
stageBreakdown: "단계 분석",
stageOrder: "단계 순서",
childCount: "하위 span 수",
kind: "유형",
status: "상태",
name: "이름",
input: "입력",
output: "출력",
metadata: "메타데이터",
error: "오류",
showJson: "JSON 표시",
hideJson: "JSON 숨기기",
empty: "데이터 없음",
inProgress: "진행 중",
elapsed: "경과",
placeholderHint: "이 단계에는 상세한 span 기록이 없으며, 추정된 상태만 표시됩니다.",
showJson: "JSON 펼치기",
hideJson: "JSON 접기",
},
stage: {
docreader: "문서 파싱",
+36 -2
View File
@@ -397,21 +397,55 @@ export default {
attempt: "第 {n} 次尝试",
attemptLatest: "第 {n} 次尝试(最新)",
retry: "重新解析",
refresh: "立即刷新",
copy: "复制",
copyDetails: "复制详情",
copied: "已复制到剪贴板",
close: "关闭",
live: "LIVE",
liveTooltip: "解析中,每 2 秒自动刷新({n}s 后)",
viewTrace: "查看 Trace",
justNow: "刚刚",
secondsAgo: "{n} 秒前",
minutesAgo: "{n} 分钟前",
noActivity: "暂无解析记录",
totalDuration: "总耗时:{d}",
total: "总耗时 {d}",
head: {
duration: "总耗时",
stages: "阶段",
stage: "阶段",
status: "状态",
attempt: "尝试",
updated: "更新于",
},
tab: {
overview: "概览",
raw: "原始 JSON",
},
detail: {
started: "开始",
finished: "结束",
duration: "耗时",
offset: "起点偏移",
timing: "时序",
identity: "身份信息",
stageBreakdown: "阶段分解",
stageOrder: "阶段顺序",
childCount: "子 span 数",
kind: "类型",
status: "状态",
name: "名称",
input: "输入",
output: "输出",
metadata: "元数据",
error: "错误",
showJson: "显示 JSON",
hideJson: "隐藏 JSON",
empty: "暂无数据",
inProgress: "进行中",
elapsed: "已用",
placeholderHint: "此阶段未记录详细 span,仅展示推断状态。",
showJson: "展开 JSON",
hideJson: "收起 JSON",
},
stage: {
docreader: "文档解析",
+62 -7
View File
@@ -87,15 +87,21 @@ func NewChunkExtractTask(
tenantID uint64,
chunkID string,
modelID string,
knowledgeID string,
attempt int,
chunkIndex int,
) error {
if strings.ToLower(os.Getenv("NEO4J_ENABLE")) != "true" {
logger.Warn(ctx, "NEO4J is not enabled, skip chunk extract task")
return nil
}
taskPayload := types.ExtractChunkPayload{
TenantID: tenantID,
ChunkID: chunkID,
ModelID: modelID,
TenantID: tenantID,
ChunkID: chunkID,
ModelID: modelID,
KnowledgeID: knowledgeID,
Attempt: attempt,
ChunkIndex: chunkIndex,
}
langfuse.InjectTracing(ctx, &taskPayload)
payload, err := json.Marshal(taskPayload)
@@ -150,6 +156,10 @@ type ChunkExtractService struct {
knowledgeBaseRepo interfaces.KnowledgeBaseRepository
chunkRepo interfaces.ChunkRepository
graphEngine interfaces.RetrieveGraphRepository
// spanTracker records this graph-extract task's subspan under the
// parent attempt's postprocess stage so the trace viewer shows real
// per-chunk graph extraction time rather than the upstream's enqueue.
spanTracker SpanTracker
}
// NewChunkExtractService creates a new chunk extract service
@@ -159,20 +169,25 @@ func NewChunkExtractService(
knowledgeBaseRepo interfaces.KnowledgeBaseRepository,
chunkRepo interfaces.ChunkRepository,
graphEngine interfaces.RetrieveGraphRepository,
spanTracker SpanTracker,
) interfaces.TaskHandler {
// generator := chatpipeline.NewQAPromptGenerator(chatpipeline.NewFormater(), config.ExtractManager.ExtractGraph)
// ctx := context.Background()
// logger.Debugf(ctx, "chunk extract system prompt: %s", generator.System(ctx))
// logger.Debugf(ctx, "chunk extract user prompt: %s", generator.User(ctx, "demo"))
return &ChunkExtractService{
template: config.ExtractManager.ExtractGraph,
modelService: modelService,
knowledgeBaseRepo: knowledgeBaseRepo,
chunkRepo: chunkRepo,
graphEngine: graphEngine,
spanTracker: spanTracker,
}
}
func (s *ChunkExtractService) tracker() SpanTracker {
if s.spanTracker == nil {
return noopSpanTracker{}
}
return s.spanTracker
}
// Handle handles the chunk extraction task
func (s *ChunkExtractService) Handle(ctx context.Context, t *asynq.Task) error {
var p types.ExtractChunkPayload
@@ -184,24 +199,59 @@ func (s *ChunkExtractService) Handle(ctx context.Context, t *asynq.Task) error {
ctx = logger.WithField(ctx, "extract", p.ChunkID)
ctx = context.WithValue(ctx, types.TenantIDContextKey, p.TenantID)
// Open a postprocess subspan keyed by chunk ordinal so the trace
// shows real per-chunk graph extraction time. Skipped silently when
// upstream didn't pass the parent attempt (legacy in-flight tasks)
// or when the postprocess stage span isn't found.
var gSpan *Span
if p.KnowledgeID != "" && p.Attempt > 0 {
parent := s.tracker().LookupStage(ctx, p.KnowledgeID, p.Attempt, types.StagePostProcess)
if parent != nil {
gSpan = s.tracker().BeginSubSpan(ctx, parent,
fmt.Sprintf("postprocess.graph.chunk[%d]", p.ChunkIndex),
types.SpanKindSubSpan,
types.JSONMap{
"chunk_id": p.ChunkID,
"chunk_index": p.ChunkIndex,
"model_id": p.ModelID,
})
}
}
var handleErr error
graphOut := types.JSONMap{}
defer func() {
if gSpan == nil {
return
}
if handleErr != nil {
s.tracker().FailSpan(ctx, gSpan, "GRAPH_EXTRACT_FAILED", handleErr.Error(), handleErr)
} else {
s.tracker().EndSpan(ctx, gSpan, graphOut)
}
}()
chunk, err := s.chunkRepo.GetChunkByID(ctx, p.TenantID, p.ChunkID)
if err != nil {
logger.Errorf(ctx, "failed to get chunk: %v", err)
handleErr = err
return err
}
kb, err := s.knowledgeBaseRepo.GetKnowledgeBaseByID(ctx, chunk.KnowledgeBaseID)
if err != nil {
logger.Errorf(ctx, "failed to get knowledge base: %v", err)
handleErr = err
return err
}
if kb.ExtractConfig == nil {
logger.Warnf(ctx, "failed to get extract config")
graphOut["skipped"] = "no_extract_config"
return err
}
chatModel, err := s.modelService.GetChatModel(ctx, p.ModelID)
if err != nil {
logger.Errorf(ctx, "failed to get chat model: %v", err)
handleErr = err
return err
}
@@ -219,12 +269,14 @@ func (s *ChunkExtractService) Handle(ctx context.Context, t *asynq.Task) error {
extractor := chatpipeline.NewExtractor(chatModel, template)
graph, err := extractor.Extract(ctx, chunk.Content)
if err != nil {
handleErr = err
return err
}
chunk, err = s.chunkRepo.GetChunkByID(ctx, p.TenantID, p.ChunkID)
if err != nil {
logger.Warnf(ctx, "graph ignore chunk %s: %v", p.ChunkID, err)
graphOut["skipped"] = "chunk_disappeared"
return nil
}
@@ -236,8 +288,11 @@ func (s *ChunkExtractService) Handle(ctx context.Context, t *asynq.Task) error {
[]*types.GraphData{graph},
); err != nil {
logger.Errorf(ctx, "failed to add graph: %v", err)
handleErr = err
return err
}
graphOut["nodes_added"] = len(graph.Node)
graphOut["relations_added"] = len(graph.Relation)
return nil
}
+42
View File
@@ -219,6 +219,48 @@ func (s *knowledgeService) skipStage(ctx context.Context, kid, name, reason stri
s.tracker().SkipSpan(ctx, span, reason)
}
// beginPostprocessSubspan opens a subspan beneath the postprocess stage
// span for (kid, attempt). Async post-pipeline tasks (summary, question,
// graph, wiki) call this on entry so their actual processing time shows
// up in the trace tree under postprocess instead of the stage looking
// like an instant ~10ms enqueue.
//
// Returns nil when:
// - attempt <= 0 (legacy in-flight task without span tracking)
// - the postprocess stage span is missing (parse predates tracker, or
// the upstream BeginStage call failed)
//
// Callers must tolerate nil — pair every begin with a deferred
// endPostprocessSubspan / failPostprocessSubspan that no-ops on nil.
func (s *knowledgeService) beginPostprocessSubspan(
ctx context.Context, knowledgeID string, attempt int, name string, input types.JSONMap,
) *Span {
if attempt <= 0 || knowledgeID == "" || name == "" {
return nil
}
parent := s.tracker().LookupStage(ctx, knowledgeID, attempt, types.StagePostProcess)
if parent == nil {
return nil
}
return s.tracker().BeginSubSpan(ctx, parent, name, types.SpanKindSubSpan, input)
}
func (s *knowledgeService) endPostprocessSubspan(ctx context.Context, span *Span, output types.JSONMap) {
if span == nil {
return
}
s.tracker().EndSpan(ctx, span, output)
}
func (s *knowledgeService) failPostprocessSubspan(
ctx context.Context, span *Span, code, msg string, err error,
) {
if span == nil {
return
}
s.tracker().FailSpan(ctx, span, code, msg, err)
}
// getParserEngineOverridesFromContext returns parser engine overrides from tenant in context (e.g. MinerU endpoint, API key).
// Used when building document ReadRequest so UI-configured values take precedence over env.
func (s *knowledgeService) getParserEngineOverridesFromContext(ctx context.Context) map[string]string {
@@ -143,12 +143,12 @@ func (s *KnowledgePostProcessService) Handle(ctx context.Context, task *asynq.Ta
enqueuedSummary := false
enqueuedQuestion := false
if len(textChunks) > 0 {
s.enqueueSummaryGenerationTask(ctx, payload)
s.enqueueSummaryGenerationTask(ctx, payload, attempt)
enqueuedSummary = true
// Question generation only makes sense for RAG indexing (improves chunk recall).
// Skip when only Wiki/Graph is enabled without vector/keyword search.
if kb.NeedsEmbeddingModel() {
enqueuedQuestion = s.enqueueQuestionGenerationIfEnabled(ctx, payload, kb)
enqueuedQuestion = s.enqueueQuestionGenerationIfEnabled(ctx, payload, kb, attempt)
}
}
@@ -156,8 +156,9 @@ func (s *KnowledgePostProcessService) Handle(ctx context.Context, task *asynq.Ta
enqueuedGraph := false
if kb.IsGraphEnabled() {
logger.Infof(ctx, "[KnowledgePostProcess] Spawning Graph RAG extract tasks for %d text-like chunks", len(textChunks))
for _, chunk := range textChunks {
err := NewChunkExtractTask(ctx, s.taskEnqueuer, payload.TenantID, chunk.ID, kb.SummaryModelID)
for i, chunk := range textChunks {
err := NewChunkExtractTask(ctx, s.taskEnqueuer, payload.TenantID, chunk.ID, kb.SummaryModelID,
payload.KnowledgeID, attempt, i)
if err != nil {
logger.Errorf(ctx, "[KnowledgePostProcess] Failed to create chunk extract task for %s: %v", chunk.ID, err)
}
@@ -190,7 +191,7 @@ func (s *KnowledgePostProcessService) Handle(ctx context.Context, task *asynq.Ta
return nil
}
func (s *KnowledgePostProcessService) enqueueSummaryGenerationTask(ctx context.Context, payload types.KnowledgePostProcessPayload) {
func (s *KnowledgePostProcessService) enqueueSummaryGenerationTask(ctx context.Context, payload types.KnowledgePostProcessPayload, attempt int) {
if s.taskEnqueuer == nil {
return
}
@@ -200,6 +201,7 @@ func (s *KnowledgePostProcessService) enqueueSummaryGenerationTask(ctx context.C
KnowledgeBaseID: payload.KnowledgeBaseID,
KnowledgeID: payload.KnowledgeID,
Language: payload.Language,
Attempt: attempt,
}
langfuse.InjectTracing(ctx, &taskPayload)
payloadBytes, err := json.Marshal(taskPayload)
@@ -216,7 +218,7 @@ func (s *KnowledgePostProcessService) enqueueSummaryGenerationTask(ctx context.C
}
}
func (s *KnowledgePostProcessService) enqueueQuestionGenerationIfEnabled(ctx context.Context, payload types.KnowledgePostProcessPayload, kb *types.KnowledgeBase) bool {
func (s *KnowledgePostProcessService) enqueueQuestionGenerationIfEnabled(ctx context.Context, payload types.KnowledgePostProcessPayload, kb *types.KnowledgeBase, attempt int) bool {
if s.taskEnqueuer == nil {
return false
}
@@ -239,6 +241,7 @@ func (s *KnowledgePostProcessService) enqueueQuestionGenerationIfEnabled(ctx con
KnowledgeID: payload.KnowledgeID,
QuestionCount: questionCount,
Language: payload.Language,
Attempt: attempt,
}
langfuse.InjectTracing(ctx, &taskPayload)
payloadBytes, err := json.Marshal(taskPayload)
@@ -887,15 +887,37 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
ctx = context.WithValue(ctx, types.LanguageContextKey, payload.Language)
}
// Open a subspan under the parent attempt's postprocess stage so the
// trace surface shows the real summary-generation duration (LLM call
// + chunk write + index) instead of just the upstream's enqueue time.
// Closes via the deferred handler below — every return path lands in
// the defer, including the early returns ahead.
span := s.beginPostprocessSubspan(ctx, payload.KnowledgeID, payload.Attempt, "postprocess.summary",
types.JSONMap{"language": payload.Language})
var summaryErr error
summaryOut := types.JSONMap{}
defer func() {
if span == nil {
return
}
if summaryErr != nil {
s.failPostprocessSubspan(ctx, span, "SUMMARY_FAILED", summaryErr.Error(), summaryErr)
} else {
s.endPostprocessSubspan(ctx, span, summaryOut)
}
}()
// Get knowledge base
kb, err := s.kbService.GetKnowledgeBaseByID(ctx, payload.KnowledgeBaseID)
if err != nil {
logger.Errorf(ctx, "Failed to get knowledge base: %v", err)
summaryErr = err
return nil
}
if kb.SummaryModelID == "" {
logger.Warn(ctx, "Knowledge base summary model ID is empty, skipping summary generation")
summaryOut["skipped"] = "no_summary_model"
return nil
}
@@ -903,6 +925,7 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
knowledge, err := s.repo.GetKnowledgeByID(ctx, payload.TenantID, payload.KnowledgeID)
if err != nil {
logger.Errorf(ctx, "Failed to get knowledge: %v", err)
summaryErr = err
return nil
}
@@ -927,6 +950,7 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
if err != nil {
logger.Errorf(ctx, "Failed to get chunks: %v", err)
markSummaryFailed()
summaryErr = err
return nil
}
@@ -937,6 +961,7 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
textChunks = append(textChunks, chunk)
}
}
summaryOut["text_chunks"] = len(textChunks)
if len(textChunks) == 0 {
logger.Infof(ctx, "No text chunks found for knowledge: %s", payload.KnowledgeID)
@@ -944,6 +969,7 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
knowledge.SummaryStatus = types.SummaryStatusCompleted
knowledge.UpdatedAt = time.Now()
s.repo.UpdateKnowledge(ctx, knowledge)
summaryOut["skipped"] = "no_text_chunks"
return nil
}
@@ -957,6 +983,7 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
if err != nil {
logger.Errorf(ctx, "Failed to get chat model: %v", err)
markSummaryFailed()
summaryErr = err
return fmt.Errorf("failed to get chat model: %w", err)
}
@@ -974,8 +1001,11 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
knowledge.UpdatedAt = time.Now()
if updateErr := s.repo.UpdateKnowledge(ctx, knowledge); updateErr != nil {
logger.Errorf(ctx, "Failed to mark summary as failed: %v", updateErr)
summaryErr = updateErr
return fmt.Errorf("failed to update knowledge: %w", updateErr)
}
summaryOut["fallback"] = "insufficient_content"
summaryErr = err
return nil
}
// For other errors (LLM API issues etc.), fall back to the first chunk.
@@ -987,6 +1017,7 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
summary = string(runes[:500])
}
}
summaryOut["fallback"] = "first_chunk"
}
}
@@ -994,8 +1025,10 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
knowledge.Description = summary
knowledge.SummaryStatus = types.SummaryStatusCompleted
knowledge.UpdatedAt = time.Now()
summaryOut["summary_chars"] = len([]rune(summary))
if err := s.repo.UpdateKnowledge(ctx, knowledge); err != nil {
logger.Errorf(ctx, "Failed to update knowledge description: %v", err)
summaryErr = err
return fmt.Errorf("failed to update knowledge: %w", err)
}
@@ -1034,6 +1067,7 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
// Save summary chunk
if err := s.chunkService.CreateChunks(ctx, []*types.Chunk{summaryChunk}); err != nil {
logger.Errorf(ctx, "Failed to create summary chunk: %v", err)
summaryErr = err
return fmt.Errorf("failed to create summary chunk: %w", err)
}
@@ -1041,6 +1075,7 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
tenantInfo, err := s.tenantRepo.GetTenantByID(ctx, payload.TenantID)
if err != nil {
logger.Errorf(ctx, "Failed to get tenant info: %v", err)
summaryErr = err
return fmt.Errorf("failed to get tenant info: %w", err)
}
ctx = context.WithValue(ctx, types.TenantInfoContextKey, tenantInfo)
@@ -1049,12 +1084,14 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
ctx, s.retrieveEngine, s.ownership, tenantInfo.ID, kb.VectorStoreID)
if err != nil {
logger.Errorf(ctx, "Failed to init retrieve engine: %v", err)
summaryErr = err
return fmt.Errorf("failed to init retrieve engine: %w", err)
}
embeddingModel, err := s.modelService.GetEmbeddingModel(ctx, kb.EmbeddingModelID)
if err != nil {
logger.Errorf(ctx, "Failed to get embedding model: %v", err)
summaryErr = err
return fmt.Errorf("failed to get embedding model: %w", err)
}
@@ -1070,13 +1107,16 @@ func (s *knowledgeService) ProcessSummaryGeneration(ctx context.Context, t *asyn
if err := retrieveEngine.BatchIndex(ctx, embeddingModel, indexInfo); err != nil {
logger.Errorf(ctx, "Failed to index summary chunk: %v", err)
summaryErr = err
return fmt.Errorf("failed to index summary chunk: %w", err)
}
logger.Infof(ctx, "Successfully created and indexed summary chunk for knowledge: %s", payload.KnowledgeID)
summaryOut["summary_chunk_indexed"] = true
}
logger.Infof(ctx, "Successfully generated summary for knowledge: %s", payload.KnowledgeID)
summaryOut["status"] = "completed"
return nil
}
@@ -1101,6 +1141,12 @@ func (s *knowledgeService) ProcessQuestionGeneration(ctx context.Context, t *asy
indexEntriesPrepared := 0
indexBatchAttempted := false
indexBatchSucceeded := false
// Postprocess subspan for the trace viewer. Opened lazily after we
// unmarshal the payload (so we have payload.Attempt) and closed in
// the defer below alongside the stats log so the span output mirrors
// what we already log to stdout.
var qSpan *Span
var qErr error
defer func() {
logger.Infof(
ctx,
@@ -1125,6 +1171,39 @@ func (s *knowledgeService) ProcessQuestionGeneration(ctx context.Context, t *asy
indexBatchAttempted,
indexBatchSucceeded,
)
if qSpan != nil {
out := types.JSONMap{
"status": exitStatus,
"total_chunks": totalChunks,
"text_chunks": totalTextChunks,
"empty_content_chunks": emptyContentChunks,
"llm_attempts": llmCallAttempts,
"llm_success": llmCallSuccess,
"llm_empty": llmCallEmpty,
"llm_failed": llmCallFailed,
"questions_generated": generatedQuestionsTotal,
"chunk_update_failed": chunkUpdateFailed,
"metadata_set_failed": chunkMetadataSetFailed,
"index_entries_prepared": indexEntriesPrepared,
"index_batch_attempted": indexBatchAttempted,
"index_batch_succeeded": indexBatchSucceeded,
"retry": retryCount,
"max_retry": maxRetry,
}
// Treat any non-success exitStatus as a failed run; the
// existing stats-string already enumerates them. qErr stays
// optional for callers that want to surface a Go error.
if exitStatus != "success" || qErr != nil {
msg := exitStatus
var detailErr error = qErr
if qErr != nil {
msg = qErr.Error()
}
s.failPostprocessSubspan(ctx, qSpan, "QUESTION_FAILED", msg, detailErr)
} else {
s.endPostprocessSubspan(ctx, qSpan, out)
}
}
}()
if err := json.Unmarshal(t.Payload(), &payload); err != nil {
@@ -1135,6 +1214,14 @@ func (s *knowledgeService) ProcessQuestionGeneration(ctx context.Context, t *asy
logger.Infof(ctx, "Processing question generation for knowledge: %s", payload.KnowledgeID)
// Open the postprocess.question subspan now that we have payload.Attempt.
// Closes via the defer above.
qSpan = s.beginPostprocessSubspan(ctx, payload.KnowledgeID, payload.Attempt, "postprocess.question",
types.JSONMap{
"question_count": payload.QuestionCount,
"language": payload.Language,
})
// Set tenant context
ctx = context.WithValue(ctx, types.TenantIDContextKey, payload.TenantID)
if payload.Language != "" {
@@ -1144,7 +1231,8 @@ func (s *knowledgeService) ProcessQuestionGeneration(ctx context.Context, t *asy
if strings.TrimSpace(s.config.Conversation.GenerateQuestionsPrompt) == "" {
exitStatus = "prompt_not_configured"
logger.Errorf(ctx, "GenerateQuestionsPrompt is empty: configure conversation.generate_questions_prompt_id")
return fmt.Errorf("generate questions prompt not configured")
qErr = fmt.Errorf("generate questions prompt not configured")
return qErr
}
// Get knowledge base
@@ -1152,6 +1240,7 @@ func (s *knowledgeService) ProcessQuestionGeneration(ctx context.Context, t *asy
if err != nil {
exitStatus = "kb_not_found"
logger.Errorf(ctx, "Failed to get knowledge base: %v", err)
qErr = err
return nil
}
@@ -1160,6 +1249,7 @@ func (s *knowledgeService) ProcessQuestionGeneration(ctx context.Context, t *asy
if err != nil {
exitStatus = "knowledge_not_found"
logger.Errorf(ctx, "Failed to get knowledge: %v", err)
qErr = err
return nil
}
@@ -2,6 +2,7 @@ package service
import (
"context"
"encoding/json"
"errors"
"testing"
@@ -271,3 +272,140 @@ func TestSpanTracker_FailSpan_CascadesDependentSubspans(t *testing.T) {
assert.Equal(t, types.SpanStatusCancelled, statusBy["embedding.batch[0]"],
"subspan under the cascaded stage must also be cancelled")
}
// TestPostprocessSubspan_AttachesUnderPostProcessStage covers the contract
// that the async post-pipeline tasks (summary, question, graph) rely on:
// after the parsing pipeline closes the postprocess stage span, an
// out-of-band worker can still LookupStage + BeginSubSpan to record its
// real processing time as a child of postprocess. Without this guarantee
// the trace viewer's postprocess row stays at the ~10ms enqueue duration
// even when summary generation takes 20 s.
func TestPostprocessSubspan_AttachesUnderPostProcessStage(t *testing.T) {
tracker, db := setupSpanTrackerTest(t)
ctx := context.Background()
repo := repository.NewKnowledgeSpanRepository(db)
// Set up the parent attempt with a closed postprocess stage — the
// async worker must still find it via LookupStage.
_, attempt, err := tracker.OpenAttempt(ctx, "kid", "lf-trace")
require.NoError(t, err)
post := tracker.BeginStage(ctx, "kid", attempt, types.StagePostProcess, types.JSONMap{
"chunks_total": 20,
})
require.NotNil(t, post)
tracker.EndSpan(ctx, post, types.JSONMap{"enqueued_summary": true})
// Simulate ProcessSummaryGeneration entering: lookup parent +
// BeginSubSpan (the same call shape as beginPostprocessSubspan).
parent := tracker.LookupStage(ctx, "kid", attempt, types.StagePostProcess)
require.NotNil(t, parent, "lookup must succeed even after EndSpan closed the parent")
assert.Equal(t, types.StagePostProcess, parent.Name)
assert.Equal(t, types.SpanKindStage, parent.Kind)
sumSpan := tracker.BeginSubSpan(ctx, parent, "postprocess.summary", types.SpanKindSubSpan,
types.JSONMap{"language": "zh-CN"})
require.NotNil(t, sumSpan)
assert.Equal(t, parent.SpanID, sumSpan.ParentSpanID,
"subspan must hang off the postprocess stage's span_id")
assert.Equal(t, types.SpanKindSubSpan, sumSpan.Kind)
tracker.EndSpan(ctx, sumSpan, types.JSONMap{
"text_chunks": 20,
"summary_chars": 142,
})
// Verify the row landed under the right parent with the right name.
rows, err := repo.ListByAttempt(ctx, "kid", attempt)
require.NoError(t, err)
var found *types.KnowledgeProcessingSpan
for i := range rows {
if rows[i].Name == "postprocess.summary" {
cp := rows[i]
found = &cp
break
}
}
require.NotNil(t, found, "summary subspan row must persist")
assert.Equal(t, parent.SpanID, found.ParentSpanID,
"persisted parent_span_id matches LookupStage result")
assert.Equal(t, types.SpanStatusDone, found.Status)
assert.NotNil(t, found.Output, "EndSpan must record the output map")
}
// TestPostprocessSubspan_MissingParentFallsThrough covers the legacy
// path: an in-flight async task may carry attempt=0 (queued before the
// span-tracking field was added) or hit a knowledge whose postprocess
// stage row is missing (parse predates tracker). LookupStage returning
// nil must NOT crash the handler — the caller is expected to skip span
// recording and continue normal processing.
func TestPostprocessSubspan_MissingParentFallsThrough(t *testing.T) {
tracker, _ := setupSpanTrackerTest(t)
ctx := context.Background()
// No OpenAttempt → no rows for kid. LookupStage must return nil.
parent := tracker.LookupStage(ctx, "kid-without-attempt", 7, types.StagePostProcess)
assert.Nil(t, parent, "missing parent attempt yields nil, not an error")
// Open an attempt but never begin postprocess. Lookup must still nil.
_, attempt, err := tracker.OpenAttempt(ctx, "kid-no-postprocess", "")
require.NoError(t, err)
parent = tracker.LookupStage(ctx, "kid-no-postprocess", attempt, types.StagePostProcess)
assert.Nil(t, parent, "missing postprocess stage row yields nil")
}
// TestChunkExtractPayload_AttemptRoundTrip verifies the new fields
// added to ExtractChunkPayload survive JSON marshal/unmarshal so a
// cross-process asynq worker can recover the parent attempt + chunk
// ordinal on the receiving side. Skipping this would let a typo in
// the JSON tag silently zero the attempt and disable span recording.
func TestChunkExtractPayload_AttemptRoundTrip(t *testing.T) {
in := types.ExtractChunkPayload{
TenantID: 42,
ChunkID: "chunk-x",
ModelID: "m1",
KnowledgeID: "kid-7",
Attempt: 3,
ChunkIndex: 9,
}
bytes, err := json.Marshal(in)
require.NoError(t, err)
var out types.ExtractChunkPayload
require.NoError(t, json.Unmarshal(bytes, &out))
assert.Equal(t, in.KnowledgeID, out.KnowledgeID)
assert.Equal(t, in.Attempt, out.Attempt)
assert.Equal(t, in.ChunkIndex, out.ChunkIndex)
}
// TestSummaryQuestionPayload_AttemptRoundTrip mirrors the above for the
// summary + question payloads to keep the contract documented.
func TestSummaryQuestionPayload_AttemptRoundTrip(t *testing.T) {
sumIn := types.SummaryGenerationPayload{
TenantID: 42,
KnowledgeBaseID: "kb-1",
KnowledgeID: "kid-7",
Language: "zh-CN",
Attempt: 5,
}
sumBytes, err := json.Marshal(sumIn)
require.NoError(t, err)
var sumOut types.SummaryGenerationPayload
require.NoError(t, json.Unmarshal(sumBytes, &sumOut))
assert.Equal(t, 5, sumOut.Attempt)
qIn := types.QuestionGenerationPayload{
TenantID: 42,
KnowledgeBaseID: "kb-1",
KnowledgeID: "kid-7",
QuestionCount: 3,
Language: "zh-CN",
Attempt: 5,
}
qBytes, err := json.Marshal(qIn)
require.NoError(t, err)
var qOut types.QuestionGenerationPayload
require.NoError(t, json.Unmarshal(qBytes, &qOut))
assert.Equal(t, 5, qOut.Attempt)
}
+20
View File
@@ -25,6 +25,16 @@ type ExtractChunkPayload struct {
TenantID uint64 `json:"tenant_id"`
ChunkID string `json:"chunk_id"`
ModelID string `json:"model_id"`
// KnowledgeID + Attempt link the per-chunk extract back to the parent
// parse attempt's postprocess stage so the worker can record a
// postprocess.graph.chunk[i] subspan. 0 / "" means "skip span
// recording" for legacy in-flight tasks.
KnowledgeID string `json:"knowledge_id,omitempty"`
Attempt int `json:"attempt,omitempty"`
// ChunkIndex is the 0-based ordinal of this chunk inside the parent
// knowledge's text-chunk set, used as the subspan name suffix
// ("postprocess.graph.chunk[3]") so the timeline preserves order.
ChunkIndex int `json:"chunk_index,omitempty"`
}
// DocumentProcessPayload represents the document process task payload
@@ -77,6 +87,12 @@ type QuestionGenerationPayload struct {
QuestionCount int `json:"question_count"`
// Language is the request locale (e.g. zh-CN, en-US) when the task was enqueued, used for {{language}} / {{lang}} in templates.
Language string `json:"language,omitempty"`
// Attempt links this task to the parent parse attempt so the worker
// can record a postprocess.question subspan under the right attempt's
// postprocess stage. 0 means "skip span recording" (legacy in-flight
// tasks queued before this field shipped, or callers without a
// tracker).
Attempt int `json:"attempt,omitempty"`
}
// SummaryGenerationPayload represents the summary generation task payload
@@ -86,6 +102,10 @@ type SummaryGenerationPayload struct {
KnowledgeBaseID string `json:"knowledge_base_id"`
KnowledgeID string `json:"knowledge_id"`
Language string `json:"language,omitempty"`
// Attempt links this task to the parent parse attempt so the worker
// can record a postprocess.summary subspan under the right attempt's
// postprocess stage. See QuestionGenerationPayload.Attempt notes.
Attempt int `json:"attempt,omitempty"`
}
// KBClonePayload represents the knowledge base clone task payload