observer/.github/workflows/org-postgres.yml at main · superbasedapp/observer · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
# org-postgres.yml - run the org-server suites that are GATED on a live
# PostgreSQL DSN, against a real postgres:16 service container, and FAIL the
# job if any of those live-PG tests silently skipped.
#
# WHY THIS EXISTS
# The stateless-collectors / Postgres-data-plane arc moved the org server off
# SQLite onto Postgres. Its riskiest bug class is a lost-update under Postgres
# MVCC (SQLite's _txlock=immediate serialized every writer, so the SQLite half
# of a two-writer test proves the final STATE, not that the Postgres
# SELECT ... FOR UPDATE / advisory-lock guard actually excludes a concurrent
# writer). Those proofs live in dbtest.ForEachEngine "postgres" subtests,
# controltest.Dialects "postgres" subtests, and TestLivePG* tests - every one
# of which t.Skip()s when OBSERVER_CONTROL_STORE_DSN is unset. ci.yml runs
# `go test ./...` with NO Postgres, so today the entire Postgres data-plane
# regression surface (plan acceptance criteria 7 + 9) skips silently on every
# PR. This job closes that gap: it points the DSN env vars the test helpers
# read at a real service container and turns a live-PG skip into a failure.
#
# REQUIRED-IN-SPIRIT: this job SHOULD gate merges to the packages below (it is
# the only automated proof the MVCC guards hold). It is intentionally NOT wired
# into branch protection here - promoting it to a required check is a repo-admin
# action, done outside this file.
#
# NATS: not needed. The named suite below touches Postgres, not the durable
# log. internal/telemetrylog's natslog tests embed an in-process nats-server
# and already run in ci.yml's `go test ./...`; the kafkalog tests are BYO and
# gate on OBSERVER_KAFKA_TEST_BROKERS (not provided anywhere). telemetrylog is
# a TRIGGER path because a change to the log semantics must re-run the Postgres
# ingest/apply/receipt suite (./internal/orgserver/ingest/...), which is in the
# set below - not because this job runs telemetrylog's own tests.

name: org-postgres

on:
  pull_request:
    paths:
      # Same paths teams-golden-path.yml watches ...
      - 'internal/orgclient/**'
      - 'internal/orgserver/**'
      - 'internal/orgcontract/**'
      - 'internal/store/orgpush*'
      - 'cmd/observer-org/**'
      - 'cmd/observer/org.go'
      - 'deploy/observer-org/**'
      # ... plus the durable log (feeds the ingest/apply/receipt suite) ...
      - 'internal/telemetrylog/**'
      - 'internal/aigateway/gwstore/**'
      # ... and this job's own wiring.
      - 'scripts/ci/org-postgres-suite.sh'
      - '.github/workflows/org-postgres.yml'
  workflow_dispatch: {}

permissions:
  contents: read

jobs:
  live-pg:
    name: live-postgres suites
    runs-on: ubuntu-latest
    timeout-minutes: 30

    services:
      postgres:
        image: postgres:16-alpine
        env:
          # The service superuser. NewPostgres (dbtest) needs CREATEDB to make
          # its per-test disposable databases; the POSTGRES_USER is a superuser
          # so that holds. This password is a fixed non-secret dev value (it
          # only ever exists inside this ephemeral runner + its service
          # container), so it is inlined rather than pulled from a secret.
          POSTGRES_USER: observer
          POSTGRES_PASSWORD: dev-postgres-password-change-me
          POSTGRES_DB: observer_org
        ports:
          - 5432:5432
        options: >-
          --health-cmd "pg_isready -U observer -d observer_org"
          --health-interval 5s
          --health-timeout 3s
          --health-retries 20

    env:
      # Every DSN env var the test helpers read (grepped from
      # internal/orgserver/db/dbtest, .../controlstore, .../leader and the
      # pgmigrations tests). OBSERVER_POSTGRES_URL is the alternate name
      # controlstore/leader/controltest also accept; OBSERVER_CONTROL_STORE_LIVE
      # is set by the helpers themselves, so it is deliberately NOT set here.
      # OBSERVER_DBTEST_ENGINE is deliberately NOT set: we want ForEachEngine to
      # run BOTH engines, not force a single-engine suite.
      OBSERVER_CONTROL_STORE_DSN: postgres://observer:dev-postgres-password-change-me@localhost:5432/observer_org?sslmode=disable
      OBSERVER_DATA_STORE_DSN: postgres://observer:dev-postgres-password-change-me@localhost:5432/observer_org?sslmode=disable
      CGO_ENABLED: '0'

    steps:
      - uses: actions/checkout@v5

      - uses: actions/setup-go@v5
        with:
          # Pins the toolchain the same way ci.yml's go job does.
          go-version-file: 'go.mod'
          cache: true

      - name: run live-Postgres org-server suites (fail if any live-PG test skipped)
        run: bash scripts/ci/org-postgres-suite.sh