Update: Search checkpoint across all Runs in Job for --resume-from

[FL4TLiN3]

Background

In #78, the --resume-from option currently searches for a checkpoint only in the most recent Run of a Job:

const run = await getMostRecentRunInJob(jobId)
checkpoint = await getCheckpointById(jobId, run.runId, input.resumeFrom)

This fails when the specified checkpointId exists in a previous Run within the same Job.

Proposed Solution

Add a findCheckpointInJob(jobId, checkpointId) function that searches across all Runs in a Job:

checkpoint = await findCheckpointInJob(jobId, input.resumeFrom)

Acceptance Criteria

• Add findCheckpointInJob(jobId: string, checkpointId: string) to run-manager.ts
• Update context.ts to use the new function
• Add tests for checkpoint search across multiple Runs

[github-actions]

I've analyzed issue #79 and the related PR #78. Here's a comprehensive answer:

Current Implementation Problem

In PR #78 (feat/job-concept branch), the --resume-from option searches for checkpoints only in the most recent Run of a Job:

In context.ts (lines 35-39):

const jobId = input.continueJob ?? (await getMostRecentCheckpoint()).jobId
const run = await getMostRecentRunInJob(jobId)
checkpoint = await getCheckpointById(jobId, run.runId, input.resumeFrom)

This fails when the specified checkpointId exists in a previous (older) Run within the same Job.

Proposed Solution

Add a new function findCheckpointInJob to run-manager.ts that searches across all Runs in a Job:

export async function findCheckpointInJob(
  jobId: string,
  checkpointId: string
): Promise<Checkpoint> {
  const runs = await getRunsByJobId(jobId)
  
  for (const run of runs) {
    try {
      const checkpoint = await getCheckpointById(jobId, run.runId, checkpointId)
      return checkpoint
    } catch {
      // Checkpoint not found in this run, continue to next
    }
  }
  
  throw new Error(`Checkpoint ${checkpointId} not found in job ${jobId}`)
}

Implementation Steps

1. Add the function to run-manager.ts:

   • Place it after the existing getCheckpointById function
   • It should iterate through all runs in the job (already sorted by updatedAt descending)
   • Return the first matching checkpoint found

2. Update context.ts:
   Replace:

   const run = await getMostRecentRunInJob(jobId)
   checkpoint = await getCheckpointById(jobId, run.runId, input.resumeFrom)

   With:

   checkpoint = await findCheckpointInJob(jobId, input.resumeFrom)

3. Add tests:

   • Test finding a checkpoint in the most recent run
   • Test finding a checkpoint in an older run
   • Test error handling when checkpoint doesn't exist in any run

This change makes --resume-from more flexible by allowing users to resume from any checkpoint in the Job's history, not just from the latest Run.

---

🤖 Bot Activity

✅ Done
---

[FL4TLiN3]

Resolved in PR #83. Checkpoint path now uses direct construction: jobs/{jobId}/checkpoints/{id}.json, eliminating the need to search across Runs.