chore(): Workflow engine timers and notification improvements (#13434)

RESOLVES CORE-1177

**What**
main changes are:
- not blocking execution when notifying
- timers management
- race condition checks improvements
This commit is contained in:
Adrien de Peretti
2025-09-08 20:19:55 +02:00
committed by GitHub
parent b776fd55dc
commit fc4d5f0ac9
14 changed files with 464 additions and 213 deletions

View File

@@ -186,6 +186,42 @@
"unique": false,
"expression": "CREATE INDEX IF NOT EXISTS \"IDX_workflow_execution_run_id\" ON \"workflow_execution\" (run_id) WHERE deleted_at IS NULL"
},
{
"keyName": "IDX_workflow_execution_workflow_id_transaction_id",
"columnNames": [],
"composite": false,
"constraint": false,
"primary": false,
"unique": false,
"expression": "CREATE INDEX IF NOT EXISTS \"IDX_workflow_execution_workflow_id_transaction_id\" ON \"workflow_execution\" (workflow_id, transaction_id) WHERE deleted_at IS NULL"
},
{
"keyName": "IDX_workflow_execution_state_updated_at",
"columnNames": [],
"composite": false,
"constraint": false,
"primary": false,
"unique": false,
"expression": "CREATE INDEX IF NOT EXISTS \"IDX_workflow_execution_state_updated_at\" ON \"workflow_execution\" (state, updated_at) WHERE deleted_at IS NULL"
},
{
"keyName": "IDX_workflow_execution_retention_time_updated_at_state",
"columnNames": [],
"composite": false,
"constraint": false,
"primary": false,
"unique": false,
"expression": "CREATE INDEX IF NOT EXISTS \"IDX_workflow_execution_retention_time_updated_at_state\" ON \"workflow_execution\" (retention_time, updated_at, state) WHERE deleted_at IS NULL AND retention_time IS NOT NULL"
},
{
"keyName": "IDX_workflow_execution_updated_at_retention_time",
"columnNames": [],
"composite": false,
"constraint": false,
"primary": false,
"unique": false,
"expression": "CREATE INDEX IF NOT EXISTS \"IDX_workflow_execution_updated_at_retention_time\" ON \"workflow_execution\" (updated_at, retention_time) WHERE deleted_at IS NULL AND retention_time IS NOT NULL AND state IN ('done', 'failed', 'reverted')"
},
{
"keyName": "workflow_execution_pkey",
"columnNames": [

View File

@@ -0,0 +1,19 @@
import { Migration } from '@mikro-orm/migrations';
export class Migration20250908080305 extends Migration {
override async up(): Promise<void> {
this.addSql(`CREATE INDEX IF NOT EXISTS "IDX_workflow_execution_workflow_id_transaction_id" ON "workflow_execution" (workflow_id, transaction_id) WHERE deleted_at IS NULL;`);
this.addSql(`CREATE INDEX IF NOT EXISTS "IDX_workflow_execution_state_updated_at" ON "workflow_execution" (state, updated_at) WHERE deleted_at IS NULL;`);
this.addSql(`CREATE INDEX IF NOT EXISTS "IDX_workflow_execution_retention_time_updated_at_state" ON "workflow_execution" (retention_time, updated_at, state) WHERE deleted_at IS NULL AND retention_time IS NOT NULL;`);
this.addSql(`CREATE INDEX IF NOT EXISTS "IDX_workflow_execution_updated_at_retention_time" ON "workflow_execution" (updated_at, retention_time) WHERE deleted_at IS NULL AND retention_time IS NOT NULL AND state IN ('done', 'failed', 'reverted');`);
}
override async down(): Promise<void> {
this.addSql(`drop index if exists "IDX_workflow_execution_workflow_id_transaction_id";`);
this.addSql(`drop index if exists "IDX_workflow_execution_state_updated_at";`);
this.addSql(`drop index if exists "IDX_workflow_execution_retention_time_updated_at_state";`);
this.addSql(`drop index if exists "IDX_workflow_execution_updated_at_retention_time";`);
}
}

View File

@@ -38,4 +38,21 @@ export const WorkflowExecution = model
on: ["run_id"],
where: "deleted_at IS NULL",
},
{
on: ["workflow_id", "transaction_id"],
where: "deleted_at IS NULL",
},
{
on: ["state", "updated_at"],
where: "deleted_at IS NULL",
},
{
on: ["retention_time", "updated_at", "state"],
where: "deleted_at IS NULL AND retention_time IS NOT NULL",
},
{
on: ["updated_at", "retention_time"],
where:
"deleted_at IS NULL AND retention_time IS NOT NULL AND state IN ('done', 'failed', 'reverted')",
},
])

View File

@@ -9,11 +9,13 @@ import {
import {
ContainerLike,
Context,
Logger,
MedusaContainer,
} from "@medusajs/framework/types"
import {
isString,
MedusaError,
promiseAll,
TransactionState,
} from "@medusajs/framework/utils"
import {
@@ -101,6 +103,7 @@ export class WorkflowOrchestratorService {
private subscribers: Subscribers = new Map()
private container_: MedusaContainer
private inMemoryDistributedTransactionStorage_: InMemoryDistributedTransactionStorage
readonly #logger: Logger
constructor({
inMemoryDistributedTransactionStorage,
@@ -113,6 +116,10 @@ export class WorkflowOrchestratorService {
this.container_ = sharedContainer
this.inMemoryDistributedTransactionStorage_ =
inMemoryDistributedTransactionStorage
this.#logger =
this.container_.resolve("logger", { allowUnregistered: true }) ?? console
inMemoryDistributedTransactionStorage.setWorkflowOrchestratorService(this)
DistributedTransaction.setStorage(inMemoryDistributedTransactionStorage)
WorkflowScheduler.setStorage(inMemoryDistributedTransactionStorage)
@@ -673,46 +680,49 @@ export class WorkflowOrchestratorService {
}
private notify(options: NotifyOptions) {
const {
eventType,
workflowId,
transactionId,
errors,
result,
step,
response,
state,
} = options
// Process subscribers asynchronously to avoid blocking workflow execution
setImmediate(() => this.processSubscriberNotifications(options))
}
private async processSubscriberNotifications(options: NotifyOptions) {
const { workflowId, transactionId, eventType } = options
const subscribers: TransactionSubscribers =
this.subscribers.get(workflowId) ?? new Map()
const notifySubscribers = (handlers: SubscriberHandler[]) => {
handlers.forEach((handler) => {
handler({
eventType,
workflowId,
transactionId,
step,
response,
result,
errors,
state,
})
const notifySubscribersAsync = async (handlers: SubscriberHandler[]) => {
const promises = handlers.map(async (handler) => {
try {
const result = handler(options) as void | Promise<any>
if (result && typeof result === "object" && "then" in result) {
await (result as Promise<any>)
}
} catch (error) {
this.#logger.error(`Subscriber error: ${error}`)
}
})
await promiseAll(promises)
}
const tasks: Promise<void>[] = []
if (transactionId) {
const transactionSubscribers = subscribers.get(transactionId) ?? []
notifySubscribers(transactionSubscribers)
if (transactionSubscribers.length > 0) {
tasks.push(notifySubscribersAsync(transactionSubscribers))
}
if (options.eventType === "onFinish") {
if (eventType === "onFinish") {
subscribers.delete(transactionId)
}
}
const workflowSubscribers = subscribers.get(AnySubscriber) ?? []
notifySubscribers(workflowSubscribers)
if (workflowSubscribers.length > 0) {
tasks.push(notifySubscribersAsync(workflowSubscribers))
}
await promiseAll(tasks)
}
private buildWorkflowEvents({

View File

@@ -284,36 +284,26 @@ export class WorkflowsModuleService<
} as any)
}
@InjectSharedContext()
async subscribe(
args: {
workflowId: string
transactionId?: string
subscriber: Function
subscriberId?: string
},
@MedusaContext() context: Context = {}
) {
async subscribe(args: {
workflowId: string
transactionId?: string
subscriber: Function
subscriberId?: string
}) {
return this.workflowOrchestratorService_.subscribe(args as any)
}
@InjectSharedContext()
async unsubscribe(
args: {
workflowId: string
transactionId?: string
subscriberOrId: string | Function
},
@MedusaContext() context: Context = {}
) {
async unsubscribe(args: {
workflowId: string
transactionId?: string
subscriberOrId: string | Function
}) {
return this.workflowOrchestratorService_.unsubscribe(args as any)
}
@InjectSharedContext()
async cancel<TWorkflow extends string | ReturnWorkflow<any, any, any>>(
workflowIdOrWorkflow: TWorkflow,
options: WorkflowOrchestratorCancelOptions,
@MedusaContext() context: Context = {}
options: WorkflowOrchestratorCancelOptions
) {
return await this.workflowOrchestratorService_.cancel(
workflowIdOrWorkflow,

View File

@@ -69,6 +69,24 @@ function parseNextExecution(
return result
}
const invokingStatesSet = new Set([
TransactionStepState.INVOKING,
TransactionStepState.NOT_STARTED,
])
const compensatingStatesSet = new Set([
TransactionStepState.COMPENSATING,
TransactionStepState.NOT_STARTED,
])
function isInvokingState(step: TransactionStep) {
return invokingStatesSet.has(step.invoke?.state)
}
function isCompensatingState(step: TransactionStep) {
return compensatingStatesSet.has(step.compensate?.state)
}
export class InMemoryDistributedTransactionStorage
implements IDistributedTransactionStorage, IDistributedSchedulerStorage
{
@@ -87,8 +105,9 @@ export class InMemoryDistributedTransactionStorage
config: SchedulerOptions
}
> = new Map()
private retries: Map<string, unknown> = new Map()
private timeouts: Map<string, unknown> = new Map()
private retries: Map<string, NodeJS.Timeout> = new Map()
private timeouts: Map<string, NodeJS.Timeout> = new Map()
private pendingTimers: Set<NodeJS.Timeout> = new Set()
private clearTimeout_: NodeJS.Timeout
@@ -113,12 +132,46 @@ export class InMemoryDistributedTransactionStorage
async onApplicationShutdown() {
clearInterval(this.clearTimeout_)
for (const timer of this.pendingTimers) {
clearTimeout(timer)
}
this.pendingTimers.clear()
for (const timer of this.retries.values()) {
clearTimeout(timer)
}
this.retries.clear()
for (const timer of this.timeouts.values()) {
clearTimeout(timer)
}
this.timeouts.clear()
// Clean up scheduled job timers
for (const job of this.scheduled.values()) {
clearTimeout(job.timer)
}
this.scheduled.clear()
}
setWorkflowOrchestratorService(workflowOrchestratorService) {
this.workflowOrchestratorService_ = workflowOrchestratorService
}
private createManagedTimer(
callback: () => void,
delay: number
): NodeJS.Timeout {
const timer = setTimeout(() => {
this.pendingTimers.delete(timer)
callback()
}, delay)
this.pendingTimers.add(timer)
return timer
}
private async saveToDb(data: TransactionCheckpoint, retentionTime?: number) {
const isNotStarted = data.flow.state === TransactionState.NOT_STARTED
const isFinished = [
@@ -453,38 +506,43 @@ export class InMemoryDistributedTransactionStorage
)
}
// Predefined states for quick lookup
const invokingStates = [
TransactionStepState.INVOKING,
TransactionStepState.NOT_STARTED,
]
const compensatingStates = [
TransactionStepState.COMPENSATING,
TransactionStepState.NOT_STARTED,
]
const isInvokingState = (step: TransactionStep) =>
invokingStates.includes(step.invoke?.state)
const isCompensatingState = (step: TransactionStep) =>
compensatingStates.includes(step.compensate?.state)
const currentFlowLastInvokingStepIndex =
currentFlowSteps.findIndex(isInvokingState)
const latestUpdatedFlowLastInvokingStepIndex = !latestUpdatedFlow.steps
let latestUpdatedFlowLastInvokingStepIndex = !latestUpdatedFlow.steps
? 1 // There is no other execution, so the current execution is the latest
: latestUpdatedFlowSteps.findIndex(isInvokingState)
: -1
const reversedCurrentFlowSteps = [...currentFlowSteps].reverse()
const currentFlowLastCompensatingStepIndex =
reversedCurrentFlowSteps.findIndex(isCompensatingState)
if (latestUpdatedFlow.steps) {
for (let i = 0; i < latestUpdatedFlowSteps.length; i++) {
if (isInvokingState(latestUpdatedFlowSteps[i])) {
latestUpdatedFlowLastInvokingStepIndex = i
break
}
}
}
const reversedLatestUpdatedFlowSteps = [...latestUpdatedFlowSteps].reverse()
const latestUpdatedFlowLastCompensatingStepIndex = !latestUpdatedFlow.steps
let currentFlowLastCompensatingStepIndex = -1
for (let i = currentFlowSteps.length - 1; i >= 0; i--) {
if (isCompensatingState(currentFlowSteps[i])) {
currentFlowLastCompensatingStepIndex = currentFlowSteps.length - 1 - i
break
}
}
let latestUpdatedFlowLastCompensatingStepIndex = !latestUpdatedFlow.steps
? -1 // There is no other execution, so the current execution is the latest
: reversedLatestUpdatedFlowSteps.findIndex(isCompensatingState)
: -1
if (latestUpdatedFlow.steps) {
for (let i = latestUpdatedFlowSteps.length - 1; i >= 0; i--) {
if (isCompensatingState(latestUpdatedFlowSteps[i])) {
latestUpdatedFlowLastCompensatingStepIndex =
latestUpdatedFlowSteps.length - 1 - i
break
}
}
}
const isLatestExecutionFinishedIndex = -1
const invokeShouldBeSkipped =
@@ -536,8 +594,16 @@ export class InMemoryDistributedTransactionStorage
interval: number
): Promise<void> {
const { modelId: workflowId, transactionId } = transaction
const key = `${workflowId}:${transactionId}:${step.id}`
const inter = setTimeout(async () => {
const existingTimer = this.retries.get(key)
if (existingTimer) {
clearTimeout(existingTimer)
this.pendingTimers.delete(existingTimer)
}
const timer = this.createManagedTimer(async () => {
this.retries.delete(key)
const context = transaction.getFlow().metadata ?? {}
await this.workflowOrchestratorService_.run(workflowId, {
transactionId,
@@ -551,8 +617,7 @@ export class InMemoryDistributedTransactionStorage
})
}, interval * 1e3)
const key = `${workflowId}:${transactionId}:${step.id}`
this.retries.set(key, inter)
this.retries.set(key, timer)
}
async clearRetry(
@@ -562,9 +627,10 @@ export class InMemoryDistributedTransactionStorage
const { modelId: workflowId, transactionId } = transaction
const key = `${workflowId}:${transactionId}:${step.id}`
const inter = this.retries.get(key)
if (inter) {
clearTimeout(inter as NodeJS.Timeout)
const timer = this.retries.get(key)
if (timer) {
clearTimeout(timer)
this.pendingTimers.delete(timer)
this.retries.delete(key)
}
}
@@ -575,8 +641,16 @@ export class InMemoryDistributedTransactionStorage
interval: number
): Promise<void> {
const { modelId: workflowId, transactionId } = transaction
const key = `${workflowId}:${transactionId}`
const inter = setTimeout(async () => {
const existingTimer = this.timeouts.get(key)
if (existingTimer) {
clearTimeout(existingTimer)
this.pendingTimers.delete(existingTimer)
}
const timer = this.createManagedTimer(async () => {
this.timeouts.delete(key)
const context = transaction.getFlow().metadata ?? {}
await this.workflowOrchestratorService_.run(workflowId, {
transactionId,
@@ -590,8 +664,7 @@ export class InMemoryDistributedTransactionStorage
})
}, interval * 1e3)
const key = `${workflowId}:${transactionId}`
this.timeouts.set(key, inter)
this.timeouts.set(key, timer)
}
async clearTransactionTimeout(
@@ -600,9 +673,10 @@ export class InMemoryDistributedTransactionStorage
const { modelId: workflowId, transactionId } = transaction
const key = `${workflowId}:${transactionId}`
const inter = this.timeouts.get(key)
if (inter) {
clearTimeout(inter as NodeJS.Timeout)
const timer = this.timeouts.get(key)
if (timer) {
clearTimeout(timer)
this.pendingTimers.delete(timer)
this.timeouts.delete(key)
}
}
@@ -614,8 +688,16 @@ export class InMemoryDistributedTransactionStorage
interval: number
): Promise<void> {
const { modelId: workflowId, transactionId } = transaction
const key = `${workflowId}:${transactionId}:${step.id}`
const inter = setTimeout(async () => {
const existingTimer = this.timeouts.get(key)
if (existingTimer) {
clearTimeout(existingTimer)
this.pendingTimers.delete(existingTimer)
}
const timer = this.createManagedTimer(async () => {
this.timeouts.delete(key)
const context = transaction.getFlow().metadata ?? {}
await this.workflowOrchestratorService_.run(workflowId, {
transactionId,
@@ -629,8 +711,7 @@ export class InMemoryDistributedTransactionStorage
})
}, interval * 1e3)
const key = `${workflowId}:${transactionId}:${step.id}`
this.timeouts.set(key, inter)
this.timeouts.set(key, timer)
}
async clearStepTimeout(
@@ -640,9 +721,10 @@ export class InMemoryDistributedTransactionStorage
const { modelId: workflowId, transactionId } = transaction
const key = `${workflowId}:${transactionId}:${step.id}`
const inter = this.timeouts.get(key)
if (inter) {
clearTimeout(inter as NodeJS.Timeout)
const timer = this.timeouts.get(key)
if (timer) {
clearTimeout(timer)
this.pendingTimers.delete(timer)
this.timeouts.delete(key)
}
}
@@ -726,11 +808,8 @@ export class InMemoryDistributedTransactionStorage
throwOnError: false,
})
// Only schedule the next job execution after the current one completes successfully
const timer = setTimeout(async () => {
setImmediate(() => {
this.jobHandler(jobId)
})
const timer = this.createManagedTimer(() => {
this.jobHandler(jobId)
}, nextExecution)
// Prevent timer from keeping the process alive