Skip to content

Runs: a shard that has not written its job yet is a launch of its own #171

Closed

Description

@harenome

Priority

high

Components

  • app/nullius (backend)
  • app/web (routes)
  • templates and static files
  • docs and the site
  • catalogue (benchmarks, agents, capabilities)
  • the standard (ports, kits, relay, runs)
  • CI, workflows and skills

Evidence

A standard launch sharded over 3 machines starts 3 run records, one per shard (<name>-shard-1of3, -2of3, -3of3). While the shards prepare or start, Runs → By launch shows one launch block per shard. When every shard has written its job, the blocks become one.

Seeded on main (411b985f) with the plans that the builder's Run compiles, all 3 records running:

shards with a written job: none
  /runs?view=launch: 3 launch block(s)
shards with a written job: [1]
  /runs?view=launch: 3 launch block(s)
shards with a written job: [1, 2, 3]
  /runs?view=launch: 1 launch block(s)

Cause, in app/nullius/runs/launches.py group: a run record joins another record's launch only by the launch ids of its written jobs (Job.launch_id, from the job's config.json). A record whose job is not written yet has no launch id, so it becomes a launch of its own. The run record has no launch id to give: standard/recipe.py puts --launch <id> on each shard's command, but the plan's job entries (Plan.jobs) hold only name, agent and dataset.

Screenshot

Not visual beyond the block count above.

Reproduction

  1. Compile a standard recipe with 3 machines in Where (standard.recipe.compile), and make one run record per shard, as web/execute.standard_plans does.
  2. Write the job of shard 1 only (or of none).
  3. Open Runs → By launch: 3 launch blocks.

Expected

The shards of one launch are one launch block from the moment their run records exist, whatever each shard has written.

Acceptance

  • A compiled sharded plan records each shard job's launch id in Plan.jobs.
  • launches.group joins a run record to its launch by that id, with none, some or all of its shards' jobs written.
  • Runs → By launch shows one block for a sharded launch whose shards have not all written their jobs.
  • Launches that are not sharded group as before.

Dependencies and related

Related: #136 and #137 (a launch's shards folded into one row per participant), #155 and #156 (delete a launch, which deletes every run of its group). Run records made before the fix carry no launch id in their plan: their shards stay apart until they write their jobs, as now.

Activity

  1. added
    bugSomething isn't working
    P2Planned: limited in scope, has a way around it, or a missing feature
    on Nov 16, 2026
  2. added theissue type on Nov 16, 2026
  3. added a commit that references this issue on Nov 17, 2026
    c4c082a
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    P2Planned: limited in scope, has a way around it, or a missing featurebugSomething isn't working

    Type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions