Skip to content

1a.8 Open Decisions — next-session prep

Created: 2026-06-29 (end of the 1a.8 doc-builder session, before a 1–2 day break; pickup ~2026-06-30 / 07-01). Purpose: the single resume point for Sprint 1 task 1a.8 (Solr write/sync). Each decision links to its detailed deferred note; this page is the queue + status.

Where 1a.8 stands

  • ✅ Doc-builder built and validated. mandala_kmassets_sync produces the flat kmassets "golden doc"; diffed against the 3 golden fixtures → 114 fields match, 0 builder-logic defects (PR #18). The builder is the bulk of 1a.8 and is done.
  • ⏳ Sinks not started — direct-to-master + file→S3 (gated by decisions B1/B2).
  • ⏳ Track A (Dave coordination) — Yuji owns the comms; gates B2.

Nothing below is a builder defect. Every item is either an external dependency, a migration-data gap, or a deliberate fidelity choice.

Blocking decisions

These gate completing 1a.8 (writing migrated content to the real index, and faithful KMaps faceting). Resolve before/alongside the sinks.

B1 — kmassets uid identity across the D7→D11 nid reassignment · High

Owner: architecture (Yuji) · Blocks: any write to the existing index · Detail: deferred/kmassets-uid-identity-across-migration

D11 reassigns node IDs (1028396 → 5), so uid = images-{nid} no longer matches the ~111k live docs or external references (reindeer_x, React app). Single-site nid collisions across the five former sites make blanket nid-preservation impossible.

  • Options: (a) carry the D7 nid on the node (e.g. field_legacy_nid / via field_other_ids) and compose id/uid from it — recommended; (b) preserve nids per-asset-type where ranges don't collide — fragile; (c) accept new identity + full reindex + consumer migration — high blast radius.
  • This is the most consequential decision — until it's made, the write path can't target the existing index.

B2 — Direct writer to the Solr master: allowed, and how does Drupal reach it? · High

Owner: Dave (Cloud Infra) · Blocks: the direct-to-master sink · Detail: [project-staging-solr-write-path memory] + doc contract §3

The staging master (mandala-solr-master-staging-private:8080/solr) takes plain HTTP gated by security-group + source-IP — no credentials. So this is a network-access question, not a creds question.

  • Options: (a) open the SG path from the Drupal subnet → master :8080 so the direct sink works; (b) route everything through the S3 → batch pipeline and drop the direct sink. (a) keeps the fast diagnostic loop; (b) is simpler but loses incremental immediacy.
  • Already confirmed from terraform-infrastructure: batch sizing BLOCK_COUNT=25, the -unprocessed/-email-notify failure path, and the existing json-transform stage — so only the access question is truly open. (Yuji handling the Dave comms.)

B3 — kmapid ancestor-ID resolution · High

Owner: architecture + migration · Blocks: faithful kmapid faceting/discovery · Detail: deferred/kmassets-kmapid-ancestor-id-resolution

The migrated KMaps path column holds breadcrumb names ({{Earth}}{{Asia}}…), not numeric ancestor IDs, so the builder's ancestor expansion (kmapid, kmapid_is) is starved. Directly-tagged kmapid_strict is correct.

  • Options: (a) resolve at migration time — populate path with ancestor_id_path via the existing KmapsPathResolver — recommended, no build-time Solr dependency; (b) resolve at build time (live kmterms-Solr call per term, cacheable but slower).
  • Plus: fix KMaps field ordering so kmapid_strict/_strict_ss/idfacets match D7's order (goldens are places → subjects → terms).

Non-blocking decisions

The builder works without these; decide before the write path is finalized.

# Decision Pri Detail
N1 Description text-format fidelity — match D7's auto-<p> filtering vs. accept the markup diff (text content already matches). Med deferred note
N2 url_html/url_ajax/url_json single-site URL scheme — currently config-templated to the D7 multisite shape; the canonical D11 scheme is undecided (affects React API-compat). Med doc contract §7
N3 projects_ss producer field — trace the D7 source of "tibet" first, then map-or-drop. Med deferred note
N4 IIIF thumb ^! vs ! — bug-for-bug (edit shared IiifUrlBuilder) vs. canonical. Low deferred note

Suggested next-session sequence

  1. Land B1 + B3 decisions (architecture) — they shape what the builder emits and whether re-migration is needed.
  2. B2 outcome from Dave → choose the sink topology.
  3. Build the sinks: direct-to-master first (fast validation loop) if B2 allows, else file→S3 first.
  4. Fold in N1–N4 as the write path firms up.
  5. Then 1a.9 (rollback story) closes Step 1a → 1b auth increment.

Reference: PR #18, doc contract, deferred index, project-1a8-doc-builder-status memory.