Changelog
User-visible changes only. Internal refactors and infrastructure work omitted.
v0.3.2 (2026-10-19)#
- ScaiQueue's own API keys are gone. The
/tenants/{tenant_id}/api-keysendpoints and the admin page are removed. Thesesq_…keys could be created but never authenticated a request. For a service, use a ScaiGrid API key of a dedicated service user and give that user a scope role; see Scopes as realms.
v0.3.1 (2026-10-15)#
- Claim one message by id:
POST /scopes/{scope_id}/messages/{msg_id}/claim. - Lapsed claims come back. A claim whose visibility timeout lapsed returns to
pending(ordead_letteredonce its attempts are used up) and can be claimed again. Before, it stayedclaimedand could no longer be claimed. - Queue counters are right. Completing, dead-lettering or cancelling a claimed message lowered
depth_pendinginstead ofdepth_claimed, sodepth_claimedstayed raised. Existing counters were recounted. - Deleting a queue that still holds messages answers
409 SCAIQUEUE_QUEUE_NOT_EMPTYwith the counts, instead of429 SCAIQUEUE_QUEUE_FULL.?purge=truedeletes the queue with its messages. - All timestamps are UTC with an explicit offset.
created_atand other server-set times were stored in the database server's local time; they are converted.
v0.3 (2026-10-15)#
- Acting on a user's behalf is recorded. When an application exchanges a user's ScaiKey token for a ScaiGrid one (RFC 8693) and claims or completes messages with it, ScaiQueue records the user as the actor and the application as
via. See Reviewing through another application. - Message representations now include
claimed_byandcompleted_by. - Claims, completions, failures and releases are now written to the audit log.
- The
scaiqueue.message.completedevent carriescompleter_via. - Admin UI: the messages list shows who handled each message, and through which application.
v0.2.2 (2026-10-14)#
- Deleting a scope that isn't archived now answers
409 SCAIQUEUE_INVALID_TRANSITION("Archive the scope before deleting it"), instead of a misleading404.
v0.2.1 (2026-10-14)#
- New permission
scaiqueue:create_scopes: create scopes and own them, and retire the scopes you own, without tenant-wide rights. For services that provision a realm per customer. - Deleting a scope now removes everything in it (queues, messages, archived messages and attachments, grants, ACL entries, trusts, rules, subscriptions, streams, schemas, HITL patterns) and reports the counts. Before, queues and grants were left behind. The audit log is kept.
v0.2 — Scopes as realms (2026-10-14)#
- Scope roles are enforced:
owner,operator,producer,consumer,viewer, for users and groups, with optional expiry. A role gives access to one scope without any tenant-wide ScaiQueue permission. See Scopes as realms. - Restricted scopes:
restricted: trueon create, orPATCH /scopes/{id}by a tenant admin. Only members and tenant admins can reach a restricted scope; everyone else gets404. GET /scopeslists only the scopes the caller can reach, each withrestricted,my_roleandcapabilities.- New
PATCH /scopes/{id}for name, description andrestricted. - Access management moves to the scope owner: grants, queue ACL entries and trusted scopes need the
ownerrole (tenant admins qualify). Grant roles are nowowner/operator/producer/consumer/viewer; the oldwriter/readerare rejected. Queue ACL entries are validated (queue in the scope, known action and effect, member in the tenant). - Group rules work for API-key callers: group membership for grants and queue ACL entries is read from ScaiGrid when the token carries no groups claim.
- Tenant isolation fixes: message, routing-rule, schema, HITL-pattern, audit and subscription routes now check that the scope belongs to the caller's tenant and that every id in the path belongs to that scope.
GET /scopes/{id}/messages?correlation_id=returns only that scope's messages. - A same-tenant trust is activated at once only when the caller owns both scopes; otherwise the target's owner accepts it.
v0.1 — Initial release#
First public iteration.
- Scopes and queues with FIFO, priority, and deadline ordering;
competingandbroadcastconsumer modes. - Message lifecycle: publish, claim, complete, fail, release, extend, cancel — with at-least-once delivery, idempotency keys, and per-message TTL / deadline.
- Routing engine with priority-ordered rules, conditions, actions, dry-run, and a 5-hop circuit breaker.
- Streams with chunk assembly (default
concatenate). - Subscriptions with Redis-backed processed-message dedup.
- HITL spec on messages plus a per-scope pattern registry with parameter expansion.
- JSON-Schema registry per scope with payload validation.
- ACL: scope grants, per-queue ACL entries, and cross-scope trust grants.
- ScaiQueue-specific API keys with rotation, revocation, and configurable grace periods.
- Audit log per scope with correlation-id trace and event-type stats.
- System agents: visibility-timeout enforcer, expiry enforcer, message archiver, dead-letter monitor, priority-aging worker.
- GDPR erasure and export endpoints.
- Auto-seeded system queues per scope:
_dead_letter,_events,_audit,_integrity,_corrections. scaiqueue.message.completedevent published on ScaiGrid's event bus carrying the originalcorrelation_id.