RecordDiff.Key() gives a stable normalized identifier ("TYPE name.") for
every diff kind, exposed as recordView.Key. ApplyRequest now takes
Updates/Prunes key lists instead of two booleans, so callers can apply a
subset of records. service.Apply builds the applied set with selected
prunes (Delete) added before selected updates (Add/Update) — an
invariant, not an option — since the provider rejects an Add/Update
whose name still conflicts with an existing record (e.g. a CNAME cannot
be created while an A on the same name still exists).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BwxdSt4reTm7Dj1oxRvpP3
handleCheck's error branch wrote last_check_status via an id-only UPDATE, so
an authenticated caller's own valid project id paired with a foreign domain
id in the URL could flip a stranger's domain to "error" even though Check
itself is project-scoped and would 404/error out first. Add project_id to
the WHERE clause (queries/domains.sql + generated db/domains.sql.go), thread
projectID through Store/TenantStore/SchedStore SetDomainStatus, and pass pid
from context at both call sites in handleCheck plus the scheduler.
Also collapse checkDomain's inline status derivation in scheduler.go into a
call to service.DeriveStatus, the same helper handleCheck already uses, so
there's a single source of truth for "drift vs in_sync" instead of two
copies that could drift apart.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BwxdSt4reTm7Dj1oxRvpP3
Manual domain checks (Recheck button / diff page load) never wrote
domains.last_check_status - only the scheduler did, leaving a
newly-templated domain stuck at "unknown" until the next scheduled run.
Extract status derivation into internal/service (single source of truth):
StatusUnknown/InSync/Drift/Error constants and DeriveStatus(diff.Changeset).
The scheduler now aliases these constants instead of duplicating them.
handleCheck persists the derived status (or StatusError on failure) via
TenantStore.SetDomainStatus after every manual check - status/history only,
no notification, which remains the scheduler's job.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BwxdSt4reTm7Dj1oxRvpP3
toChangesetResponse initialises updates/prunes/readOnly so a zone matching
its template exactly (e.g. right after 'create template from zone') marshals
arrays, not null. DiffView/DomainDiffPage also normalise null defensively.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BwxdSt4reTm7Dj1oxRvpP3
LoadDomain requires a template, so a zone without one could never be
viewed or snapshotted. Adds a template-free path: store.LoadZone /
service.ZoneRef / DomainService.ZoneRecords read a zone's live records
straight from the provider (no diff, no template). GET
/domains/{did}/records exposes read-only viewing; POST
/domains/{did}/template-from-zone snapshots only managed record types
(NS/SOA excluded) into a new template and auto-attaches it to the domain.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BwxdSt4reTm7Dj1oxRvpP3
vasyansk
