feat(paymodels): add auto-select-single-paymodel to survive workspace terminate

[jawadqur]

Users on vpodc.data-commons.org are currently unable to launch workspaces after terminating one. This adds an opt-in fix.

The bug

Terminating a workspace fires this goroutine (hatchery.go:629):

go func() {
    for { ... poll until status == "Not Found" ... }
    err = resetCurrentPaymodel(userName)
}()

resetCurrentPaymodelInDB sets current_pay_model = false on every row for that user. On the next launch, getCurrentPayModel hits this branch (paymodels.go:108):

if pm == nil || len(*pm) == 0 {                    // nothing with current=true
    activePayModels, _ := payModelsFromDatabase(userName, false)
    if activePayModels != nil && len(*activePayModels) > 0 {
        return nil, nil                            // <-- here
    }
    pm, err := getDefaultPayModel()                // unreachable in this case
    return pm, nil
}

The row is still request_status: "active", so it returns nil, and hatchery.go:516 responds 500 Current Paymodel is not set. Launch forbidden. The default-pay-model fallback is unreachable — it only fires for users with zero rows.

Observed in vadcprod over two days across three users:

2026/09/23 21:39:18 Current Paymodel is not set. Launch forbidden for user craigrbarnes@uchicago.edu
2026/09/23 21:50:27 Current Paymodel is not set. Launch forbidden for user shawnoconnor@uchicago.edu
2026/09/24 15:11:37 Current Paymodel is not set. Launch forbidden for user aartiv@uchicago.edu

GET /paymodels for an affected user:

{
  "current_pay_model": null,
  "all_pay_models": [
    { "bmh_workspace_id": "3813c208-...", "request_status": "active",
      "current_pay_model": false, "hard-limit": 5, "soft-limit": 4 }
  ]
}

The intended recovery is re-selecting via POST /setpaymodel, but on a commons using default-pay-model as a blanket trial that choice is never surfaced, so users are simply stuck.

The fix

New opt-in key "auto-select-single-paymodel". When set, a user whose only active pay model is not flagged current gets that row returned as current.

Returning the user’s own row — rather than substituting DefaultPayModel — is deliberate. DefaultPayModel has no bmh_workspace_id, so pods would be annotated with an empty paymodel_type; handlePodDeleted would then pass podPaymodelID == "" and skip updatePayModelCost entirely, silently breaking cost tracking. Keeping the real row preserves the workspace ID, its hard-limit/soft-limit, and accrued total-usage.

Users with 2+ pay models are untouched: they have a real choice and the flag must not make it for them.

Tests

Three cases added to Test_GetCurrentPayModel, all passing; the 5 existing cases still pass unchanged.

case	flag	rows	expected
AutoSelectSingle_SingleActiveNotCurrent	on	1	returns the row
AutoSelectSingle_MultipleActiveNotCurrent	on	2	nil (unchanged)
AutoSelectSingleDisabled_SingleActiveNotCurrent	off	1	nil (unchanged)

Verified the new tests fail without the paymodels.go change and pass with it. go build ./..., go vet, and gofmt are clean.

Note: TestCreateLogGroup fails on this branch, but it also fails on clean master — pre-existing and unrelated.

Rollout

Default is off; no commons changes behavior until it opts in. For vpodc, add to the hatchery JSON in gen3-gitops:

"auto-select-single-paymodel": true,

Existing stuck rows self-heal on the next launch — no DynamoDB surgery needed.

Not addressed here

• The reset goroutine polls every 5s with no timeout; if the pod is rolled mid-loop the reset never lands and the flag is left stale true. Opposite inconsistency, worth a separate look.
• Separately, vpodc has a "priciing" typo zeroing all workspace costs — fixed in fix(vpodc): correct "priciing" typo so hatchery cost tracking works gen3-gitops#2124.

[github-actions]

Integration Tests

filepath	passed	SUBTOTAL
tests/test_discoverypage.py	1	1
tests/test_workspace.py	1	1
TOTAL	2	2

Please find the detailed integration test report here

Please find the Github Action logs here

[github-actions]

Integration Tests

filepath	passed	SUBTOTAL
tests/test_discoverypage.py	1	1
tests/test_workspace.py	1	1
TOTAL	2	2

Please find the detailed integration test report here

Please find the Github Action logs here