Mastra Workflows, Enhanced

Mastra Workflows are getting some major boosts this summer (stay tuned). And we just took our first big "step" (pun intended).

We recently released our Temporal integration. Now you can still use Mastra's workflow framework while getting all the benefits of Temporal.

Now you get all the benefits of Mastra's workflows, but supercharged by Temporal to become their own, persistent system.

Here's what that unlocks for you.

More Durable Workflows

At Mastra we know that workflows are the backbone of many agentic systems. Thus, they should be treated as their own durable systems.

In Mastra's framework, workflow snapshots provide pause and resume resilience. Every step of a workflow persists as a data snapshot with all step context. Any failure can be restarted from the last step, providing workflow stability.

To help minimize failures in the first place, Temporal deploys workflows as clusters on distributed servers.

These workflows can survive failures, network outages, data center issues, etc. And no need for manual intervention. This means less retries and a more smooth UX.

Advanced Retries

Workflows often interact with external services that may result in transient errors. Every viable system needs built-in retries.

Mastra currently allows for automated retries at the step or workflow level to help ensure these are mitigated. You can add a lot of controls to a RetryConfig:

interface RetryConfig {
  /** Maximum retry attempts before the task is marked failed. */
  maxRetries?: number;
 
  /** Delay between retries in milliseconds. */
  retryDelayMs?: number;
 
  /** Multiplier applied to retryDelayMs on each subsequent attempt. */
  backoffMultiplier?: number;
 
  /** Upper bound on the retry delay regardless of backoff. */
  maxRetryDelayMs?: number;
 
  /** Predicate decides if a given error should be retried. */
  retryableErrors?: (error: Error) => boolean;
}

Temporal adds some additional control.

With Temporal, retries can target specific actions/calls within steps. So, rather than retrying an entire step, you can just retry the specific failure point.

There are also some additional features built into Temporal:

• Jitter: retries spread at random intervals to prevent cascade failures from multiple simultaneous retries
• Max retry time: Bound retries by time rather than just number
• Separate server: Retries, like workflows, are managed by Temporal server so can survive crashes

More Complex Scheduling

Mastra workflows support built-in cron scheduling with one line:

const workflow = createWorkflow({
  id: 'daily-report-workflow',
  inputSchema: z.object({ reportType: z.string() }),
  outputSchema: z.object({ generated: z.boolean() }),
  steps: [generateReportStep],
  // Run daily at midnight
  cron: '0 0 * * *',
})

Quick and easy. You can also introduce things like concurrency, rate limiting, and priority.

With Temporal, scheduling is treated as its own definition:

// Schedule a workflow to run at specific times
const handle = await client.schedule.create({
  scheduleId: 'daily-report',
  spec: {
    intervals: [{
      every: Duration.ofDays(1)
    }]
  },
  action: {
    type: 'StartWorkflow',
    workflowType: 'dailyReport',
    input: { reportType: 'daily' }
  }
})

If you have more complex needs, Temporal offers additional scheduling options such as timezone-awareness and calendar based runs. You also can explore schedules within Temporal's UI.

Increased Observability

We know observability is a first-class feature. That's why we just announced Mastra's observability as a standalone part of Mastra's platform. This is no less important for workflows.

Mastra's workflows use things like snapshots, suspend/resume, and time travel to allow for a ton of debugging. Workflow runs can also be individually tested in Studio/Observability, and every run trace can be explored.

Integrating Temporal allows you to utilize their UI to get an even deeper look into workflows: Query execution history, search workflows by attributes, worker and queue monitoring.

For a large-scale, distributed workflow system, having these extra features helps get a more granular view of your workflows for debugging and development.

Get Started

At their core, Mastra workflows are best suited for application-level task sequences. For long-running distributed systems, an integration with Temporal provides a more comprehensive, durable workflow experience (although we have exciting updates here coming very soon). Give it a shot and let us know how it goes:

npm install @mastra/temporal @temporalio/client @temporalio/worker @temporalio/envconfig

Refer to the documentation for any questions.