Files
shumai/packages/agent/src/workflows/agent-embedding.ts
T
Yiling-J 176b7f0eab feat(workflow): migrate workflow engine from temporal to dbos (#481)
* feat(workflow): migrate workflow engine from temporal to dbos

Replace Temporal and LocalExecutor with DBOS Transact (@dbos-inc/dbos-sdk) as the unified workflow engine. Background tasks, media transcoding, and AI agent workflows now execute with durable steps and task queues directly against PostgreSQL, eliminating the need for an external Temporal cluster and separate polling executors.

* fix(workflow): submit tasks via DBOSClient to allow enqueueing from steps

Enqueuing workflows using DBOS.enqueueWorkflowWithOptions threw a
DBOSInvalidWorkflowTransitionError when triggered from within a running
step or activity (e.g. creating embedding or autofill tasks during
video transcoding).

Use DBOSClient to submit and cancel workflow tasks directly against the
system database, detaching execution from the ambient workflow/step
context.

* feat(web): support SHUMAI_WORKERS environment variable

* fix(workflow): ensure CLI tasks are enqueued to DBOS and run DBOS migrations in global setup

- Initialize workflows and wait for pending DBOS submissions in shumai regenerate-embeddings CLI
- Remove no-op submit bypass and enqueue tasks via DBOSClient in producer and CLI contexts
- Add DBOS migration runner in test global setup to avoid concurrent migration deadlocks across test forks
- Update changelog and add regression tests

* fix(workflow): ensure unique per-instance executor identity for worker replicas

- Derive unique per-instance DBOS executor IDs across worker replicas

- Support SHUMAI_WORKER_ID, DBOS__VMID, and POD_NAME (Kubernetes Downward API)

- Disambiguate multi-replica bare metal/container deployments with host, PID, and instance suffix

- Prevent startup recovery from re-enqueueing workflows actively executed by other live workers

- Add multi-instance startup recovery tests and update deployment documentation

* fix(workflow): only register listened queues and update concurrency on change

- Only register queues that the executor instance actively listens to

- Use always_update conflict resolution so environment variable changes take effect on restart

- Prevent dedicated workers from overwriting unmanaged queue concurrency configurations

- Add tests verifying concurrency updates and queue isolation across worker roles

* refactor(workflow): simplify activities to use defineStep and direct calls (#482)

* refactor(workflow): simplify activities to use defineStep and direct calls

- Wrap activities using defineStep from @shumai/workflow-core
- Remove obsolete getActivities, registerActivities, dynamic activity proxies, and workerQueue arguments
- Remove executor.ts and registry.ts, consolidating workflow registration into workflow.ts
- Call activities directly across transcode and agent workflows
- Update all corresponding workflow tests and developer guide documentation

* fix(workflow): properly detect DBOSWorkflowCancelledError in runOnCancel fallback

- DBOSWorkflowCancelledError in DBOS SDK does not set name, inheriting 'Error'
- Implement isDbosWorkflowCancelledError checking instanceof, constructor.name, and dbosErrorCode
- Update unit tests to verify real DBOS SDK cancellation error instances

* docs(workflow): update documentation and readme for dbos migration

- Rename docs/configuration/temporal.mdx to docs/configuration/distributed.mdx
- Document DBOS-based distributed worker architecture and task queues
- Remove obsolete docker-compose/temporal/
- Update environment variables documentation in docs/configuration/env-variables.mdx
- Update references in README.md, README_CN.md, docs/docs.json, and related guides

* feat(deploy): implement KEDA ScaledJob and scale-to-zero On-Demand transcoding (#483)

* feat(deploy): add shumai-transcode docker image and k8s autoscaling example with garage s3

* docs(k8s): document port-forwarding and in-cluster s3 domain mapping

* feat(storage): support public s3 endpoint for browser uploads and update k8s example

* fix(k8s): configure s3 bucket cors rules in garage initialization

* fix(k8s): increase transcode worker memory limit to 4Gi for 4K video encoding

* feat(deploy): add AWS EKS deployment with Graviton3 Spot nodegroups and autoscaling

* docs(deploy): redesign EKS documentation, rename k8s to eks, and add scalable deployment feature

* docs: rephrase DBOS feature bullet to Durable Workflows via DBOS

* docs(eks): upgrade default instances to Graviton5 c9g.large and c9g.xlarge

* feat(eks): use On-Demand for general nodegroup and add spot capacity diversification

* docs(eks): fix Architecture diagram to locate all transcode worker pods in Transcode Compute Pool

* feat(deploy): implement KEDA ScaledJob and scale-to-zero On-Demand transcoding

* feat(transcode): support idle-timeout auto-drain for KEDA scaled jobs

* fix(deploy): fix EKS deployment manifests and documentation based on cluster verification
2026-10-19 15:35:08 +08:00

280 lines
8.5 KiB
TypeScript

import type { WorkflowTask } from '@shumai/db'
import {
ApplicationFailure,
updateTaskStatusActivity,
updateTaskUsageActivity,
} from '@shumai/workflow-core'
import { createCommentActivity, updateCommentActivity } from '../activities/agent'
import {
deleteAssetEmbeddingsActivity,
deleteS3ObjectActivity,
generateImageEmbeddingActivity,
generateVideoChunkEmbeddingActivity,
getEmbeddingContextActivity,
saveAssetEmbeddingsActivity,
transcodeVideoChunkActivity,
type ExistingAssetEmbeddingInfo,
} from '../activities/ai'
export interface GeneratedEmbedding {
embedding: number[]
startTime?: number
endTime?: number
}
export type { ExistingAssetEmbeddingInfo }
export async function agentEmbeddingMedia(task: WorkflowTask): Promise<void> {
let placeholderCommentId: string | undefined
try {
// Update status to processing
await updateTaskStatusActivity({
taskId: task.id,
status: 'processing',
})
// 0. Create Placeholder Comment
const payload = task.payload
const placeholder = await createCommentActivity({
assetId: task.assetId,
message: '__EMBEDDING__',
sessionId: payload?.agent?.sessionId || task.id,
agentId: payload?.agent?.agentId || 'default',
})
placeholderCommentId = placeholder.id
if (!task.teamId) {
throw ApplicationFailure.create({ message: 'Task has no teamId', nonRetryable: true })
}
// 1. Fetch Agent Context
const context = (await getEmbeddingContextActivity({
teamId: task.teamId,
assetId: task.assetId,
})) as unknown as {
asset: {
id: string
mediaType: string | null
storageKey?: { key: string } | null
media?: unknown
}
chunkDuration: number
chunkOverlap: number
embeddingProvider: string
existingEmbeddings?: ExistingAssetEmbeddingInfo[]
}
const { asset, chunkDuration, chunkOverlap, embeddingProvider, existingEmbeddings } = context
const isGemmaProvider = embeddingProvider === 'gemma-embedding-2-openai'
const existing = existingEmbeddings || []
// Purge incompatible embeddings (e.g. 1536d when Gemma is used, or 768d when Gemini is used)
const incompatibleIds = existing
.filter((e) => (isGemmaProvider ? e.has1536 || !e.has768 : e.has768 || !e.has1536))
.map((e) => e.id)
if (incompatibleIds.length > 0) {
await deleteAssetEmbeddingsActivity({
assetId: task.assetId,
ids: incompatibleIds,
})
}
const validExistingChunks = existing.filter((e) => !incompatibleIds.includes(e.id))
const mediaInfo = asset.media as PrismaJson.MediaInfo | null
const is3dWithViews =
mediaInfo?.proxyType === '3d' && (mediaInfo.imageTranscodes?.length ?? 0) > 0
const isImage = asset.mediaType?.startsWith('image/') || is3dWithViews
const isVideo = asset.mediaType?.startsWith('video/')
if (!isImage && !isVideo) {
throw ApplicationFailure.create({
message: `unsupported media type for embeddings: ${asset.mediaType}`,
nonRetryable: true,
})
}
// Resolve the S3 key to use for embedding (preferring transcoded version)
let embeddingKey = asset.storageKey?.key
let embeddingMediaType = asset.mediaType
if (asset.media) {
// eslint-disable-next-line @typescript-eslint/no-explicit-any
const media = asset.media as any
if (isVideo) {
if (media.videoTranscodes && media.videoTranscodes.length > 0) {
const transcode = media.videoTranscodes[0] as PrismaJson.VideoTranscode
if (transcode?.key) {
embeddingKey = transcode.key
}
}
} else if (isImage) {
if (media.imageTranscodes && media.imageTranscodes.length > 0) {
const transcode = media.imageTranscodes[0] as PrismaJson.ImageTranscode
if (transcode?.key) {
embeddingKey = transcode.key
if (transcode.format) {
embeddingMediaType = `image/${transcode.format}`
}
}
}
}
}
if (!embeddingKey) {
throw ApplicationFailure.create({ message: 'asset has no key', nonRetryable: true })
}
const usage = {
model: isGemmaProvider ? 'gemma-embedding-2-openai' : 'gemini-embedding-2',
inputTokens: 0,
outputTokens: 0,
}
if (isImage) {
const alreadyEmbedded = validExistingChunks.some(
(e) => e.startTime === null && e.endTime === null,
)
if (!alreadyEmbedded) {
const result = await generateImageEmbeddingActivity({
teamId: task.teamId,
assetKey: embeddingKey,
mediaType: embeddingMediaType || 'image/jpeg',
})
await saveAssetEmbeddingsActivity({
assetId: task.assetId,
embeddings: [{ embedding: result.embedding }],
})
if (result.usage) {
usage.inputTokens += result.usage.inputTokens || 0
usage.outputTokens += result.usage.outputTokens || 0
}
}
} else if (isVideo) {
// Get video duration
// eslint-disable-next-line @typescript-eslint/no-explicit-any
const duration = (asset.media as any)?.duration || 0
const limit = chunkDuration || 60.0
const overlap = chunkOverlap ?? 5.0
const step = Math.max(1, limit - overlap)
for (let start = 0.0; start < duration; start += step) {
let end = start + limit
if (end >= duration) {
end = duration
}
// Check if this chunk is already completed
const chunkExists = validExistingChunks.some(
(c) =>
c.startTime !== null &&
c.endTime !== null &&
Math.abs(c.startTime - start) < 0.1 &&
Math.abs(c.endTime - end) < 0.1,
)
if (chunkExists) {
// If this chunk reached the end of the video, stop
if (end >= duration) {
break
}
continue
}
// Step A: Slice chunk on transcode worker and upload to S3
const chunkRes = await transcodeVideoChunkActivity({
assetId: asset.id,
assetKey: embeddingKey,
startTime: start,
endTime: end,
format: isGemmaProvider ? 'frames' : 'video',
})
const { chunkKey } = chunkRes
try {
// Step B: Generate embedding on agent worker
const chunkResult = await generateVideoChunkEmbeddingActivity({
teamId: task.teamId,
chunkKey,
})
// Step C: Save chunk embedding to database immediately
await saveAssetEmbeddingsActivity({
assetId: task.assetId,
embeddings: [
{
embedding: chunkResult.embedding,
startTime: start,
endTime: end,
},
],
})
if (chunkResult.usage) {
usage.inputTokens += chunkResult.usage.inputTokens || 0
usage.outputTokens += chunkResult.usage.outputTokens || 0
}
} finally {
// Step D: Delete temporary chunk from S3
await deleteS3ObjectActivity({
key: chunkKey,
})
}
// If this chunk reached the end of the video, stop to avoid redundant overlapping chunks
if (end >= duration) {
break
}
}
}
// Update Usage
await updateTaskUsageActivity({
taskId: task.id,
inputTokens: usage.inputTokens,
outputTokens: usage.outputTokens,
model: usage.model,
})
// 7. Update Placeholder Comment
if (placeholderCommentId) {
await updateCommentActivity({
commentId: placeholderCommentId,
message: 'Embedding completed successfully.',
})
}
// Update status to completed
await updateTaskStatusActivity({
taskId: task.id,
status: 'completed',
})
} catch (err) {
console.error(`AgentEmbeddingMedia failed for task ${task.id}:`, err)
// Update placeholder comment with error message
if (placeholderCommentId) {
try {
await updateCommentActivity({
commentId: placeholderCommentId,
message: `Embedding failed: ${err instanceof Error ? err.message : String(err)}`,
})
} catch (commentErr) {
console.error('Failed to update error comment:', commentErr)
}
}
// Update status to failed
await updateTaskStatusActivity({
taskId: task.id,
status: 'failed',
output: { error: err instanceof Error ? err.message : String(err) },
})
throw err
}
}
export const agentEmbeddingWorkflow = agentEmbeddingMedia