• Home
  • Features
  • Pricing
  • Docs
  • Announcements
  • Sign In

run-llama / llama-agents / 39347013761
85%

Build:
DEFAULT BRANCH: main
Ran 20 Oct 2026 12:20AM UTC
Jobs 0
Files 0
Run time –
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst

pending completion
  cancel
39347013761

push

github

web-flow
Don't report idle while sent events are undelivered (#817)

Fixes run-llama/llama_index#23583

When a step fans out with `ctx.send_event()`, the sent events are
recorded in `BrokerState.deliveries` until their keyed `TickAddEvent` is
reduced (#791). `_check_idle_state` only looked at step queues and
running workers, so the run reported `WorkflowIdleEvent` as soon as the
sending step finished, before the sent events had reached their steps. A
client that answers or saves the run on the first idle event then has
only part of the input requests.

`_broker_busy` now treats pending deliveries as work, which covers both
the idle event and the idle check behind idle release.

Output of the script from the issue (fork → two steps that each
`wait_for_event`):

Before:
```
idle: requests seen [], steps waiting [], undelivered events 2
asked left
idle: requests seen ['left'], steps waiting ['left'], undelivered events 1
asked right
idle: requests seen ['left', 'right'], steps waiting ['left', 'right'], undelivered events 0
result [1, 1]
```

After:
```
asked left
asked right
idle: requests seen ['left', 'right'], steps waiting ['left', 'right'], undelivered events 0
result [1, 1]
```

Tests:
- `test_check_idle_state_pending_delivery_is_pending_work`: a recorded
delivery keeps the state non-idle.
- `test_control_loop_idle_event_waits_for_sent_events`: the issue's
fan-out; the first `WorkflowIdleEvent` comes after both
`InputRequiredEvent`s.

Both fail without the change. The `llama-index-workflows` tests pass
apart from `test_snapshot_preserves_true_first_attempt_at`, which fails
intermittently on `main` too on my machine. In the DBOS package,
`test_idle_release_e2e_sqlite` is also intermittent here with and
without the change, and the other DBOS failures I see on Windows happen
on `main` as well.

I used Claude Code for this PR and reviewed the change and the results.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Source Files on build 39347013761
Detailed source file information is not available for this build.
  • Back to Repo
  • Github Actions Build #39347013761
  • cdc529fa on github
  • Prev Build on main (#38455839589)
  • Next Build on main (#39347029776)
  • Delete
STATUS · Troubleshooting · Open an Issue · Sales · Support · CAREERS · ENTERPRISE · START FREE TRIAL · SCHEDULE DEMO
ANNOUNCEMENTS · TWITTER · TOS & SLA · Supported CI Services · What's a CI service? · Automated Testing

© 2026 Coveralls, Inc