Phase 0: PostgreSQL 16 + Apache AGE 1.6.0 + pgvector 0.8.2 compatibility spike - #1
Merged
noelsaw1 merged 4 commits intoJul 21, 2026
Conversation
7-phase serial chain (p0→p6) with per-phase briefs, swarm/concurrency analysis, and a Phase-0 swarm-preflight contract. Reviewed to Approved via a 4-round headless Codex relay (all blockers resolved): phase write-sets widened to match briefs, pytest gate standardized, swarm analysis corrected to an honest post-P0 per-phase decomposition, and p0's scope widened to permit its success + failure evidence. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Built via a headless Codex marathon turn (autonomously fired, per
operator's prior authorization), then hand-verified against the exit
gate: image builds; both extensions coexist in one PostgreSQL 16
database (age 1.6.0, vector 0.8.2); the container starts healthy.
Gate result: 3/4 pytest markers pass, including the plan's single
highest-risk item — an AGE session survives connection-pool checkout
and reinitializes correctly (test_age_pool_checkout_reinitializes_session).
The remaining failure (test_combined_vector_candidate_feeds_curated_age_query)
and the corresponding docker/postgres/test/{age,combined}-smoke.sql are
a known limitation, not a compatibility gap: AGE's cypher() forbids a
parameterized 3rd argument, so labels/relationship types can't be bind
params — a curated Cypher template is needed, per the plan's own
guidance on this exact restriction. Next: rewrite that one query
without parameterized labels and re-run the gate for a clean pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
noelsaw1
added a commit
that referenced
this pull request
Jul 21, 2026
phases/reflex-hybrid-analytics-starter-p0--p0/{RELAY.md,ESCALATION.md}
were auto-committed by marathon-drive.sh's own turn protocol and
landed in main via PR #1 unintentionally — they're build-tooling
transcripts, not application deliverables. Untrack them and ignore
phases/, relay-system/, and .tick/ going forward so future marathon
phases don't repeat this.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
noelsaw1
added a commit
that referenced
this pull request
Jul 21, 2026
…ests The PR #1 merge landed Phase 0 at an honest 3/4 pytest pass with one known failure. Closing it out surfaced four real, independently confirmed AGE/SQLAlchemy findings (not one, as originally scoped): 1. SQLAlchemy's text() treats any ":word" as its own bind-parameter marker, colliding with Cypher's ":Label" syntax. Fixed by escaping literal colons (\:Document). 2. cypher()'s 1st argument (graph name) must be a literal in the query text, never a bind parameter of any kind -- AGE resolves it at parse time to look up label/table OIDs, before a genuine protocol param is bound. create_graph()/drop_graph() are plain functions and accept a bind param fine; only cypher()'s own graph-name argument doesn't. 3. cypher()'s 3rd argument (its native parameter map) must be a real Postgres Param/Const, not a value computed per-row in the same statement (a CROSS JOIN LATERAL column, or even a non-correlated scalar subquery). Fixed in Python with two round trips: fetch the candidate client-side, pass it as a genuine top-level bind param. Fixed in SQL with PL/pgSQL EXECUTE format() -- genuine dynamic SQL, exactly what PROJECT.md's own database section anticipated. 4. count(*) aggregated directly over a cypher() result (even via a CTE) is unreliable for graph data created earlier in the same session -- returns 0 rows where an identical non-aggregate SELECT correctly returns 1. Root cause not fully isolated; PL/pgSQL's aggregate-into- variable form is reliable and used throughout instead. Findings 2 and 3 both bear on PROJECT.md Phase 5's "one CTE, no round-trip" bridge design -- verify there whether that goal is achievable as stated before committing to it. Also fixed an unrelated pre-existing bug in scripts/verify-extensions: every psql call runs INSIDE the container via `compose exec`, but was passed the HOST-mapped port (55432) instead of the container's own (5432) -- meaning verify-extensions could never have passed as originally written, independent of any AGE-level fix. Also fixed age-smoke.sql's CREATE: three comma-separated patterns don't reuse nodes from earlier patterns in the same clause (Cypher CREATE never matches existing nodes), so the KNOWS edge was (harmlessly, but confusingly) connecting two accidental duplicate nodes rather than the two named people. Chained CREATE clauses with shared variables fix this. Verified: uv run pytest -m 'database or integration' -> 4 passed. ./scripts/verify-extensions -> all 4 SQL smoke tests pass, extension check passes, "Phase 0 extension smoke checks passed." Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
8 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Phase 0 of the Reflex Hybrid Analytics Starter build plan — the PostgreSQL 16 + Apache AGE + pgvector compatibility spike that gates everything downstream.
docker/postgres/Dockerfilebuilding PostgreSQL 16.14 + AGE 1.6.0 + pgvector 0.8.2 from source-pinned, SHA-verified checkouts.tests/test_database_smoke.py),scripts/verify-extensions,compose.test.yaml, a minimaldatabase-image.ymlCI workflow, anddocs/architecture/database.md+docs/upgrade-matrix.md.PROJECT/2-WORKING/PROJECT-marathon/) that produced this was QA'd to Approved over 4 headless-Codex relay rounds before firing.Result — honest status: 3/4 gate, one known limitation
age+vectorextensions coexist in one PG16 DBage 1.6.0,vector 0.8.2)test_age_pool_checkout_reinitializes_sessionThe one failure (
test_combined_vector_candidate_feeds_curated_age_query, and the correspondingage-smoke.sql/combined-smoke.sql) is a known AGE limitation, not a compatibility gap:cypher()'s third argument must be a literal, so Cypher node labels / relationship types can't be bind parameters. The generated query tried to parameterize them. Fix is a curated Cypher template with literal labels — already the approach the plan calls for; not yet applied here.Test plan
docker compose -f compose.test.yaml build postgres— succeedsdocker compose -f compose.test.yaml up -d postgres— healthypsql -c '\dx'— confirmsage 1.6.0+vector 0.8.2both presentdocker/postgres/test/relational-smoke.sql— passesdocker/postgres/test/vector-smoke.sql— passesdocker/postgres/test/age-smoke.sql— fails (known limitation above)docker/postgres/test/combined-smoke.sql— fails (same root cause)uv run pytest -m 'database or integration'— 3 passed, 1 failed (same known failure)Next
Rewrite the combined query's Cypher without parameterized labels, re-run the gate for a clean 4/4, then release Phase 1.
🤖 Generated with Claude Code