WIP: MariaDB baseline rescue and CI handoff - #107
Draft
thedancingdeveloper wants to merge 9 commits into
Draft
Conversation
Preservation commit. A previous agent performed most of P1's database work and crashed before committing any of it; 94 files were sitting unstaged on this branch with zero commits ahead of main. Nothing here is reviewed. Committed as-is, unsplit, so the work survives. State at time of rescue, measured against origin/main (81c865b): migrations 18 files deleted, replaced by 001_baseline.sql (907 lines, 60 tables, MariaDB 11.4 dialect) sqlx feature postgres -> mysql PgPool refs 153 -> 0 $n placeholders 1420 -> 1 ON CONFLICT 77 -> 0 RETURNING 78 -> 8 crate rename stackarr-postgres -> stackarr-mariadb Two things need a decision before any of this becomes a PR: 1. The rename dropped the embedded-database subsystem. stackarr-postgres was 1,216 LOC of Postgres provisioning (download binaries, initdb, supervise a child process, in external/managed/embedded modes). stackarr-mariadb is 98 LOC; config.rs, lifecycle.rs and provision.rs were deleted rather than ported. Whether StackArr still ships a self-provisioning database is a product decision, not a task detail. 2. This spans T21-T25 in one commit. It must be split per item before review, since the repository requires one item per branch and PR. Refs #59, #60, #61, #62, #64
The rescued swap renamed crates/stackarr-postgres to crates/stackarr-mariadb in Cargo.toml but left AGENTS.md naming the old crate, so the doc-drift guard landed by T29 failed: $ python3 scripts/check_workspace_docs.py workspace documentation drift: Cargo.toml has [... 'crates/stackarr-mariadb'], AGENTS.md has [... 'crates/stackarr-postgres'] exit=1 Now green: "workspace documentation verified: 17 members". Also drops the sentence announcing the rename as future work, since it has happened. This is the guard doing exactly the job §P0.3 gave it. Refs #60
The rescued swap converted Rust code but left every surface around it on
PostgreSQL, so the binary spoke the sqlx mysql driver while the deployment
stack still started postgres:17 and handed it a postgresql:// URL. P1's exit
criterion — a fresh docker run reaching /health from an empty database — could
not have passed.
Runtime and deployment:
docker/docker-compose.yml postgres:17 -> mariadb:11.4, mysql:// URL
docker/docker-compose.dev.yml likewise; now publishes 3306, matching the
TEST_DATABASE_URL default the tests already
assumed
docker/docker-compose.prod.yml likewise; volume path and the required env
var change (see below)
docker/docker-compose.test.yml likewise
docker/run-tests.sh pg_isready -> healthcheck.sh
tests/e2e/*.yml, *.toml five stacks and five configs
docker/mariadb-init/ new: grants the dev account server-wide
CREATE/DROP, which TestDb needs to make and
drop a database per test
CI (T19, the MariaDB-service part):
- mariadb:11.4 service container on the test job, with a health gate
- TEST_DATABASE_URL pointing at it
- a second test step running the #[ignore]d tests. Those 31 tests are the
only ones that touch a real server, so without this the service container
would have proved nothing.
Source and config:
- 34 stale #[ignore = "requires running postgres"] reasons
- two connect_lazy() URLs still using the postgresql:// scheme
- config.example.toml and stackarr.toml still advertising port 5432
- no "postgres" left in any .rs file
Docs:
- docs/DATABASE.md rewritten for MariaDB and the single baseline. It had
described 18 migrations that no longer exist and a PgPool API.
- docs/UNIFIED-ARR-PLAN.md: corrected two statements that said the swap was
concentrated in stackarr-postgres and amounted to a rename. That crate
held zero PgPool and zero sqlx references.
Deployment-affecting, needs a human decision before deploy:
- prod requires STACKARR_MARIADB_PASSWORD, not STACKARR_POSTGRES_PASSWORD
- the prod data volume moves to .../stackarr/mariadbdata
- docker/Dockerfile.standalone builds --features managed-postgres, which the
rename deleted. Left in place and flagged in-file rather than rewritten or
removed, because whether StackArr still ships a self-provisioning database
is a product decision. Nothing in CI builds it.
- the indexarr sidecar keeps its own postgres service; D8 (#29) owns whether
it merges in, so this change does not decide it.
Gates, locally, with no server present:
cargo fmt --all -- --check pass
cargo clippy --workspace --all-features -Dwarn pass
cargo test --workspace --all-features 1010 passed, 0 failed, 31 ignored
scripts/check_workspace_docs.py pass
scripts/check_dependency_sources.py pass
The 31 ignored tests have still never executed against MariaDB — no docker
daemon on this host. The new CI step is the first thing that will run them.
Refs #57, #58, #59, #60, #61, #62, #63, #64
Commits the working-tree change that was sitting unstaged, plus the handoff issue register. NOT complete T19. Still missing, and raised as #105: coverage ratcheting (the referenced coverage-watchdog is a monitoring web application, not a Rust coverage-ratchet command, so a tool decision and a recorded baseline are needed first), and musl, which is not representable as a Docker platform. Committed rather than left in the working tree because that is how work gets destroyed -- this checkout already lost most of a day's rescue to being uncommitted once. Deliberately NOT pushed: the register makes pushing conditional on T20 (#58) acceptance. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Collaborator
Author
|
Harness execution update: T19 was selected after human-decision/epic rows were blocked as queue cleanup, but the Fleet worker exited with the item claimed and session orphaned (tracked in agent-harness#87). T19 was re-queued and no application fix was applied. The PR remains draft/dirty pending careful ci.yml conflict resolution and upstream harness lifecycle repair. |
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.
Status
Rescues the in-flight MariaDB baseline work and records the handoff issues. This is intentionally a draft.
Blocking review
Verification
cargo fmt --all -- --checkcargo clippy --workspace --all-features --locked -- -D warningsSee
docs/2026-08-03-handoff-issues.mdfor the evidence register.