[Task] listRuns() filters by workflow name: Temporal visibility cannot match a name with a backslash

[gplanchat]

Split from #264 point 3 (S5). The in-memory, DBAL and Illuminate filters (exact match) and the conformance case are parked in commit 1c60c18 on branch feat/run-catalog-workflow-filter, without a PR.

What blocks it

On Temporal, no visibility query matches a WorkflowType that contains a backslash. Tested on a private Server 1.25.2 (Postgres) and on the CLI 1.9.1 dev server CI uses: one to four backslashes, single or double quotes, and STARTS_WITH "App\\" all return nothing, while WorkflowType = "Plain" matches. Durable workflow names can be FQCNs, so the filter cannot run on the server.

Options (human decision)

• (a) Temporal filters in PHP, page by page. That brings back [plugin] Split the dashboard into a list route and a run route #264's empty-page problem.
• (b) A capability flag on the port, like tellsWaitingForWorker: Temporal declares it cannot filter by name, and the UI hides the filter.
• (c) Temporal filters on something Durable writes at start, a memo or a search attribute. That would come with OpenSpec change: backend-data-parity — make the four backends expose the same thing #269's durableExecutionId memo work.

Done when

• The option is chosen and recorded here.
• listRuns() takes the filter on all four catalogues, or Temporal declares it cannot, and the conformance suite covers whichever it is.
• UPGRADE note if the port signature changes.

Refs #264, #269, #514, #557. Filter UI: #383 slice B.

[gplanchat]

Decision (the user, 2026-09-26): option (c), custom search attributes.

• Durable writes a DurableWorkflowName Keyword search attribute at start, holding a normalized name with no backslash (the query parser rejects them; the value is ours to choose). The catalogue filters on it, and normalizes the searched name the same way.
• The same work adds a DurableExecutionId Keyword attribute, so that findRun($executionId) and [Task] listRuns() filters by run id fragment, on every catalogue #557's filter can query Temporal directly ([Task] On Temporal the run list shows the server's run id, not the execution id Durable assigned #514).
• Registration per namespace (temporal operator search-attribute create) is an install step: documented, and done by the bench and CI. SQL visibility needs Server ≥ 1.20; 10 Keyword attributes per namespace, 255 characters each.
• Runs started before the change carry no attribute: documented.

First step before any code: check on a private Server 1.25.2 (Postgres) that such a Keyword filters, with = and STARTS_WITH. Assigned to alice. Refs #514, #557, #269.

[gplanchat]

Step 0: Keyword search attributes on Temporal Server 1.25.2 (Postgres)

Setup. temporalio/auto-setup:1.25.2 (serverVersion reports 1.25.2) with postgres:16 (DB=postgres12), private network, default namespace. DurableWorkflowName and DurableExecutionId were registered as Keyword with temporal operator search-attribute create.

Five runs carried both attributes. A sixth, legacy-no-attr, carried neither and stands in for a run started before the change.

What matches

Query	Result
DurableWorkflowName = 'App.Workflow.OrderWorkflow' (single or double quotes)	exactly the 2 runs with that name
DurableWorkflowName = 'App/Workflow/OrderWorkflow', = 'App:Workflow:OrderWorkflow'	the matching run
DurableWorkflowName STARTS_WITH 'App.Workflow.Order'	OrderWorkflow ×2 and OrderRefund
DurableWorkflowName STARTS_WITH 'App.Workflow.OrderW'	OrderWorkflow ×2 only
DurableExecutionId = '01J9ZK3Q-order-1'	that run
DurableExecutionId STARTS_WITH '01J9ZK3Q'	the 2 runs whose id starts with it
DurableExecutionId STARTS_WITH 'order'	nothing (a prefix, not a substring)
STARTS_WITH '%', STARTS_WITH '_pp'	nothing: % and _ are matched as literal characters
DurableExecutionId IN ('…', '…')	both runs
DurableWorkflowName = '…' AND ExecutionStatus = 'Running'	the 2 runs
DurableWorkflowName = '…' AND DurableExecutionId STARTS_WITH '…'	the expected run
DurableWorkflowName IS NULL	legacy-no-attr only
DurableExecutionId = 'quote\'run', STARTS_WITH 'quote\''	the run whose id contains a '

What is rejected or does not match

• LIKE: rejected with invalid operator 'like'. There is no substring match on 1.25.2, so [Task] listRuns() filters by run id fragment, on every catalogue #557 can offer exact or prefix matching only.
• ORDER BY: rejected with operation is not supported: 'order by' clause. The catalogue query must not use it.
• A backslash in the value still matches nothing, as jane found in 1c60c18. That is why the name is normalized before it is written.

Install-time delay

For a few seconds after search-attribute create, a start that sets the attribute fails:

Namespace default has no mapping defined for search attribute DurableExecutionId

It succeeds once the namespace cache refreshes. The install step, whether in the bench, in CI or run by hand, has to wait or retry after registering.

Noise

The CLI hit a few one-off context deadline exceeded errors, and each of those queries succeeded when retried. They are client timeouts on a cold server, not query rejections.