Skip to content

v2.44.0

Choose a tag to compare

View all tags
@m-aebrer m-aebrer released this 23 Jul 16:06
· 54 commits to master since this release
v2.44.0
39ea93a
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Fast Mobile Fleet Updates

The dashboard now pushes compact, debounced fleet snapshots over SSE instead of repeatedly fetching the full cross-project fleet after lifecycle events. This substantially reduces bandwidth and server work on mobile connections while keeping fleet cards current.

Improved

  • SSE-pushed fleet state. Live runtime cards update from event-derived snapshots built from in-memory state, with no child RPC calls or disk inventory scans in the lifecycle path.
  • Narrow inventory refreshes. Create, resume, stop, and delete actions refresh only on-disk sessions rather than reloading the full fleet.
  • Single-call session drill-in. Opening a session now hydrates state, messages, and background agents through one HTTP request and one ordered RPC snapshot.
  • Bounded stats polling. Fleet stats refresh at most every 30 seconds while visible, with per-runtime freshness, request timeouts, single-flight polling, and last-good value retention.
  • Client-side activity previews. Fleet cards use hydrated transcript entries for recent assistant text, retaining authoritative initial/resync values as fallback.
  • Reliable local actions. Successful stops update local state immediately and converge idempotently with the later SSE removal event.

Mobile Measurements

  • The measured five-runtime workload returned a 15.6 MB authoritative fleet response and required roughly 0.9 seconds of local server processing.
  • After the change, 12 coalesced fleet snapshots totaled about 137 KB over a 60-second workload.
  • Chromium acceptance coverage verifies one initial fleet request, no lifecycle-triggered fleet reads, card updates within the throttled-mobile threshold, and one-request session hydration.

Reliability and Tooling

  • Adds barrier-aware hydration replay, stale-response protection, bounded hydration queues, authoritative follow-up snapshots, disk-inventory convergence across tabs, and deterministic fleet ordering.
  • Adds a payload-free mobile transport profiler for fleet and SSE timing, size, event-type, and burst measurements.
  • Adds --no-live-api test guarding across provider suites, including Bedrock and OAuth credential resolution, so local hooks cannot unexpectedly make external API calls.
  • Expands dashboard race, timeout, browser, streaming-byte, and user-action regression coverage.

Implemented in PR 388.

Assets 2
Loading