[aersam]

Why

The on-prem CI agents (BMS_OnPrem, 4 GB RAM / 2 GB swap each) run out of memory and go down. pgdevkit's shared pgdevkit-postgres container has no memory cap by default, and PGDEVKIT_TESTDB_MEM_LIMIT alone left the swap limit unbounded, so a container "limited" to e.g. 1g could still spill into swap and thrash the agent.

Changes

• PGDEVKIT_TESTDB_MEM_LIMIT now also sets memswap_limit to the same value (the --memory-swap=<same> idiom). New PGDEVKIT_TESTDB_MEMSWAP_LIMIT overrides it (-1 = unlimited swap, the old behaviour). Same for PGDEVKIT_TESTDB_MSSQL_*. No limit is applied when MEM_LIMIT is unset, as before.
• New PGDEVKIT_TESTDB_PG_SETTINGS (e.g. shared_buffers=128MB max_connections=50): whitespace-separated key=value pairs (values may contain commas, e.g. search_path=a,b, but no spaces) appended as -c key=value to the Postgres start command, after the defaults.
• MEMSWAP_LIMIT without a MEM_LIMIT is ignored (Docker/Podman need both).
• Version bump 0.16.1 -> 0.17.0 (pyproject.toml, uv.lock).
• README + tests.

Behaviour change

Anyone who already sets *_MEM_LIMIT now also gets a matching swap cap. Set *_MEMSWAP_LIMIT=-1 to opt out.

Not verified

• Rootless podman only enforces memory limits when the cgroup v2 memory controller is delegated to the user. Unit tests cover the kwargs passed to the Docker API, not enforcement on the agents; check on a test branch that the container still starts on devops_onprem-*.
• As before, limits only apply when pgdevkit creates the container: remove a stale pgdevkit-postgres on the agents (podman rm -f pgdevkit-postgres) for them to take effect.

🤖 Generated with Claude Code