Repository navigation
Simpler Runs and Results with Restart; sync without gaps; credential show/hide - #489
Merged
Merged
Conversation
The catalog sync deleted the contracts of every piece first, then upserted the pieces one request at a time. For the ~30 s that took, a flow check found pieces with no inputs or outputs and refused runs with FLOW_WIRE_TARGET, FLOW_WIRE_SOURCE and FLOW_OUTPUT. Each piece's delete and upsert are now one mutation, which Hasura runs as one transaction.
The Runs step table had one column per adjustable parameter, so flows with many settings produced wide tables with horizontal and vertical scroll bars nested inside the step pane. Parameters that differ between the listed runs keep a column; the rest are summarised once above the table as 'Same for every run'. The table no longer has its own vertical scroll box.
A pasted key with a space at either end reached the model server as a key that does not start with sk-, and was refused with a 401 though it looked right. The Saved credentials form now has a show/hide button for each typed value and warns when a value starts or ends with a space or a quote mark. The adapter stores every field with spaces at either end removed, and the form trims too. Stored values are still never read back.
…ically' into fix/sync-runs-credentials
…sync-runs-credentials
…nto fix/sync-runs-credentials
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This was referenced Nov 20, 2026
A plan for pre-model adapter steps encoded only the thread, the model and the saved adapter plan ids, so planning the same thread again gave the same plan id. Submitting it returned the earlier run, which meant a failed run could never be started again. Add a short nonce to such plans, as flow and post-model plans already differ on every planning. Submitting one plan twice still returns the one run.
Runs: one collapsible card per model (status, duration, one next-step sentence). A flow's steps are drawn once, its task ids sit under a collapsed Technical details, and a paused run puts its edit controls first. A single run drops the counters, run configuration box, pager and table for Log, Files and a collapsed Settings used; several runs keep compact counters, the table trimmed to what differs, and a pager only past one page. A failed or cancelled run shows its reason and a Restart button (also on each failed table row) that sends the model again and is disabled while the send is in flight. Results: one collapsible card per model with a single Refresh. Each output of a single run is its own block with type, Download, Copy location and an inline preview (Markdown report, first CSV rows, indented JSON), read only when open and capped at 200 KB, with a quiet fallback. How this was made is collapsed. A run that has not finished says so, with a link to Runs or a Restart. Several runs keep a trimmed results table. Update the facilitator guide pages to match.
…kind A plan id is the encoded plan, and submitting an id that already has a record returns that record. Pre-model steps, post-model plans and flow plans now all carry a nonce, so planning again after a failed run is a new run even if the adapter returns the same plan id, while submitting one plan twice still returns one run. Tests plan, fail, re-plan and submit for each kind (pre-model, post- model, flow, plain job) with an adapter that repeats its ids, and check the plan id size limit with the nonce.
Restart: one control shared by the Runs and Results steps. It carries the earlier run's maximum time and queue (or shows the default to change), pre- selects the catalogs the earlier plan published to, is offered only once every run of the model is over, asks to confirm for a model run several times (restart all N runs, successes included), and is disabled while a send of that model is in flight in either step. The per-row button is gone. Previews: only Tapis files are previewed, never a web address. The body is read as a stream and cancelled at 200 KB; a Markdown report opens by itself only when the listing says it is under 100 KB. Markdown input is bounded (line length, line count, bounded patterns) with a notice, links show their host, CSV rows and cells are capped, JSON is indented only below a size and depth limit, and a cut off JSON or CSV is described as cut off.
This was referenced Nov 20, 2026
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Combines #486, #487 and #488 with the Runs and Results redesign, so they build and deploy together.
1. Runs and Results steps, simplified (new)
Reported on dev: the two steps were "confusing with too much going on".
model_result) and pre-selects the earlier publishing choice.MarkdownLite, CSV first rows (capped columns and cells), JSON pretty-printed (with size and depth limits).unifiedExecutionService.ts): every adapter-backed plan (pre-model steps, post-model, flow) gets a random nonce inside its signedplan_id. A re-plan after a failure therefore creates a new run instead of returning the failed record. The same plan submitted twice still returns one run.unifiedExecutionRestart.test.tscovers each plan kind.2. Sync replaces a piece's contracts in the same transaction as its upsert (#486)
This fixes the transient
FLOW_WIRE_TARGET/FLOW_WIRE_SOURCE/FLOW_OUTPUTrefusals that happened while a catalog sync was running.3. Run tables show only the settings that differ between runs (#487)
This removes the nested scroll bars in each card.
4. Credentials: show a typed value on request; values saved trimmed (#488)
Reviews
QA and security reviews ran on the redesign, and every finding is fixed:
Checks
tsc -band prettier are clean; eslint shows 0 errors.test_demo_apifailures (git-ignored.env) remain.Known limits
🤖 Generated with Claude Code