A swap of the active site is stored as two single-site requests held together by naming conventions #184

Closed
opened 2026-10-06 21:25:18 +00:00 by cgalo5758 · 5 comments
Owner

A swap parks one site and activates another. core.lifecycle_requests holds one request per instance, so a swap is two set_status rows under one workflow id, each naming the other site in its parameters. Readers rebuild it from conventions: a second swap is refused by workflow-id shape (IsSwapRequestWorkflowID), a dismissal pairs rows by workflow id, the slot count in internal/integrations/fedwiki/store/queries/sites.sql skips rows with swap_park_site_id, an unstarted swap deletes its rows in separate statements, and FedWiki answers a repeated swap itself (sameRowRequest) because core would compare the park site the console chose.

The published protocol (#171) will carry lifecycle requests and FedWiki moves onto it (#174), so the model must say first whether a request can cover several instances, and if not, what holds the rows together.

Done when: docs/models/provider-integration.md states it, #171 carries it, and no code finds a swap by workflow id or parameter key.

A swap parks one site and activates another. `core.lifecycle_requests` holds one request per instance, so a swap is two `set_status` rows under one workflow id, each naming the other site in its parameters. Readers rebuild it from conventions: a second swap is refused by workflow-id shape (`IsSwapRequestWorkflowID`), a dismissal pairs rows by workflow id, the slot count in `internal/integrations/fedwiki/store/queries/sites.sql` skips rows with `swap_park_site_id`, an unstarted swap deletes its rows in separate statements, and FedWiki answers a repeated swap itself (`sameRowRequest`) because core would compare the park site the console chose. The published protocol (#171) will carry lifecycle requests and FedWiki moves onto it (#174), so the model must say first whether a request can cover several instances, and if not, what holds the rows together. Done when: `docs/models/provider-integration.md` states it, #171 carries it, and no code finds a swap by workflow id or parameter key.
cgalo5758 added the
kind
design
area/fedwikiarea/integrations
labels 2026-10-06 21:25:18 +00:00
Author
Owner

#179, #181 and #184 share one gap: an action that writes several rows has no record of its own, so readers rebuild it from prefixes, a borrowed request id, or the shape of a workflow id. They take one convention, decided by whichever is worked first and followed by the others. #184 comes first, because the published protocol (#171) needs its answer. #131 asks the same of subscription changes.

#179, #181 and #184 share one gap: an action that writes several rows has no record of its own, so readers rebuild it from prefixes, a borrowed request id, or the shape of a workflow id. They take one convention, decided by whichever is worked first and followed by the others. #184 comes first, because the published protocol (#171) needs its answer. #131 asks the same of subscription changes.
cgalo5758 added this to the Public launch milestone 2026-10-07 00:16:56 +00:00
Author
Owner

The order changes: #187 sets the convention for an action that writes several rows, because its plan-wide acts need it first. A tier removal, a default change or a reorder will record one act and one row per organization it reaches, each pointing at the act. That design is researched against every action that will follow it: this issue's swap, #179's tier changes, #181's rule-change batches, #131's subscription changes and #171's acts, so each fits it. This issue follows the convention, and still answers before #171 starts whether a lifecycle request can cover several instances.

The order changes: #187 sets the convention for an action that writes several rows, because its plan-wide acts need it first. A tier removal, a default change or a reorder will record one act and one row per organization it reaches, each pointing at the act. That design is researched against every action that will follow it: this issue's swap, #179's tier changes, #181's rule-change batches, #131's subscription changes and #171's acts, so each fits it. This issue follows the convention, and still answers before #171 starts whether a lifecycle request can cover several instances.
Author
Owner

Two more things the swap's design has to cover.

Core's rule for a repeated request (sameLifecycleContent in internal/integration/lifecycle.go: the same operation, the same requesting person, the same parameters as JSON values) is written twice more in FedWiki, as sameRowRequest (internal/integrations/fedwiki/web/requests.go) and sameCreate (internal/integrations/fedwiki/web/api.go). For every parameter FedWiki writes, the three agree except the swap's swap_park_site_id and swap_activate_site_id, which sameRowRequest ignores because the console chooses the site to park under the lock. sameCreate has no such reason. Once the site to park leaves a request's parameters, core's rule answers every repeat and both copies can go. Core reads an address through one function that the recording and the lookup share, so a lookup that also applies the content rule is a small addition there.

A swap recorded as one act takes two instance locks. Today it records the activated site's request first and the parked site's second (swapActiveSite in internal/integrations/fedwiki/web/site_actions.go), not in instance order. That is safe only because both run under the workspace lock and a site belongs to one workspace. An act that records requests on several instances without that outer lock has to take their locks in ascending instance order.

Two more things the swap's design has to cover. Core's rule for a repeated request (`sameLifecycleContent` in `internal/integration/lifecycle.go`: the same operation, the same requesting person, the same parameters as JSON values) is written twice more in FedWiki, as `sameRowRequest` (`internal/integrations/fedwiki/web/requests.go`) and `sameCreate` (`internal/integrations/fedwiki/web/api.go`). For every parameter FedWiki writes, the three agree except the swap's `swap_park_site_id` and `swap_activate_site_id`, which `sameRowRequest` ignores because the console chooses the site to park under the lock. `sameCreate` has no such reason. Once the site to park leaves a request's parameters, core's rule answers every repeat and both copies can go. Core reads an address through one function that the recording and the lookup share, so a lookup that also applies the content rule is a small addition there. A swap recorded as one act takes two instance locks. Today it records the activated site's request first and the parked site's second (`swapActiveSite` in `internal/integrations/fedwiki/web/site_actions.go`), not in instance order. That is safe only because both run under the workspace lock and a site belongs to one workspace. An act that records requests on several instances without that outer lock has to take their locks in ascending instance order.
Author
Owner

The slot holds FedWiki now keeps for creates and activations in flight key on the lifecycle request, not on the workflow id a swap's two requests share, and a swap takes no hold: it takes its slot inside its record transaction. The slot count no longer reads the swap's park-site parameter, so one reader of the two-request convention is gone.

The slot holds FedWiki now keeps for creates and activations in flight key on the lifecycle request, not on the workflow id a swap's two requests share, and a swap takes no hold: it takes its slot inside its record transaction. The slot count no longer reads the swap's park-site parameter, so one reader of the two-request convention is gone.
cgalo5758 referenced this issue from a commit 2026-10-22 18:40:40 +00:00
Author
Owner

Fixed in 153b2812.

A lifecycle request now covers exactly one instance, and docs/models/provider-integration.md states this as an invariant.

When one action changes several instances, the console records one request per instance in one transaction and groups them under one act. An act is a row in core.acts that records who acted, when and why.

A swap of the active site records two requests under the new cause instances_swapped:

  • the member's request, for the site they chose (role asked);
  • the console's request, for the site it parks (role consequent). This request names no person.

Only the database function core.record_lifecycle_act writes the link from a request to its act. It runs with its owner's rights, and it links only requests that its own transaction inserted, so a request can never be moved into an act later.

Nothing finds a swap by its workflow id or a parameter key any more:

  • IsSwapRequestWorkflowID, sameRowRequest and the swap_park_site_id and swap_activate_site_id parameters are gone.
  • The notice list, dismissal, and the settle that runs when the console starts all read the act.
  • A swap whose workflow cannot start loses both its requests in one statement.
  • A repeated request is answered by core's own rule. sameCreate stays until FedWiki moves onto the protocol (#174).

#171 has the grouping rule for the protocol. The commit message lists the upgrade steps for a deployment that has a swap in flight.

Fixed in 153b2812. A lifecycle request now covers exactly one instance, and `docs/models/provider-integration.md` states this as an invariant. When one action changes several instances, the console records one request per instance in one transaction and groups them under one act. An act is a row in `core.acts` that records who acted, when and why. A swap of the active site records two requests under the new cause `instances_swapped`: - the member's request, for the site they chose (role `asked`); - the console's request, for the site it parks (role `consequent`). This request names no person. Only the database function `core.record_lifecycle_act` writes the link from a request to its act. It runs with its owner's rights, and it links only requests that its own transaction inserted, so a request can never be moved into an act later. Nothing finds a swap by its workflow id or a parameter key any more: - `IsSwapRequestWorkflowID`, `sameRowRequest` and the `swap_park_site_id` and `swap_activate_site_id` parameters are gone. - The notice list, dismissal, and the settle that runs when the console starts all read the act. - A swap whose workflow cannot start loses both its requests in one statement. - A repeated request is answered by core's own rule. `sameCreate` stays until FedWiki moves onto the protocol (#174). #171 has the grouping rule for the protocol. The commit message lists the upgrade steps for a deployment that has a swap in flight.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: wiki-cafe/member-console#184