feat: scale Postgres memory settings with instance RAM

[tonychang04]

Problem

docker-init/db/postgresql.conf ships nano-safe memory values (shared_buffers=32MB, work_mem=2MB, maintenance_work_mem=16MB, effective_cache_size=128MB) and every instance tier runs them — #98 made max_connections/pool scale with RAM, but the memory settings stayed fixed, so a large (8G) or xl (16G) instance runs nano-tuned Postgres and leaves most of its RAM unused by the planner and buffer cache.

Change

Same mechanism as #98:

• auto-scale-memory.sh picks per-tier values (nano tier = current conf values, unchanged) and writes PG_SHARED_BUFFERS / PG_EFFECTIVE_CACHE_SIZE / PG_WORK_MEM / PG_MAINTENANCE_WORK_MEM to .env
• compose passes them as -c flags (command line wins over config_file); the -c defaults equal the conf values, so hosts without the env vars behave exactly as before

Tier	shared_buffers	effective_cache_size	work_mem	maintenance_work_mem
nano (~0.5G)	32MB	128MB	2MB	16MB
micro (~1G)	64MB	256MB	4MB	32MB
small (~2G)	128MB	512MB	4MB	64MB
medium (~4G)	256MB	1GB	8MB	128MB
large (~8G)	512MB	2GB	8MB	256MB
xl (~16G)	1GB	4GB	16MB	512MB
2xl (~32G)	2GB	8GB	16MB	1GB

shared_buffers stays ~15–25% of the postgres container limit (not host RAM) since the host is shared with the API; work_mem is bounded per tier against worst-case work_mem × max_connections.

Verification

• bash -n + real run of auto-scale-memory.sh (24G host → xl tier written to .env correctly)
• docker compose config renders the -c flags for both the scaled case and the no-env default case (defaults = conf values, no behavior change)
• Nano values are the exact config just validated under load (see docs: load-test evidence for the nano Postgres memory settings #99): survives 90-conn storm fail-fast, 0 OOM

Notes

• Applies wherever auto-scale-memory.sh runs (provision + instance resize). Existing running instances are untouched until their next auto-scale + compose up.
• Companion evidence doc: docs: load-test evidence for the nano Postgres memory settings #99

🤖 Generated with Claude Code

https://claude.ai/code/session_01J9NtiFkU34HQQgRSffNb4k

---

Summary by cubic

Scale Postgres memory settings with instance RAM so larger tiers don’t run nano defaults and can better use available memory. docker-compose.yml now passes these settings via -c flags with safe defaults, so behavior is unchanged until auto-scale runs.

• New Features

  • auto-scale-memory.sh writes PG_SHARED_BUFFERS, PG_EFFECTIVE_CACHE_SIZE, PG_WORK_MEM, PG_MAINTENANCE_WORK_MEM to .env based on RAM tiers.
  • docker-compose.yml applies them with -c flags; defaults equal current postgresql.conf values.
  • Tiers mirror connection scaling; shared_buffers ~15–25% of the Postgres container limit; work_mem bounded per tier.

• Migration

  • No action needed. Changes apply on next provision/resize or after running auto-scale-memory.sh and docker compose up. Existing instances keep current behavior until then.

^{Written for commit 419f78b. Summary will update on new commits.}

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: a83a801e-3c27-4587-9b0c-340cc32e2cff

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/scale-pg-memory-settings

---

_{Comment @coderabbitai help to get the list of available commands.}