fix(db): restore elastic postgres pool

[izadoesdev]

Summary

• Restore the elastic pg.Pool in the shared PostgreSQL client.
• Keep the pool lazy/dynamic (min remains 0), cap each replica at 50 connections, close idle clients after 30 seconds, and bound acquisition at 10 seconds.
• Preserve PgBouncer-compatible URL normalization, query timing, error handling, warmup, and graceful shutdown.

Production evidence

• PostgreSQL/PgBouncer itself is not saturated: PgBouncer was 23/400, direct PostgreSQL 10/50, and database CPU was 8%.
• The current production runtime emits Calling client.query() when the client is already executing a query, proving concurrent requests are sharing one pg.Client.
• A 39-request /links/create burst from one API key produced p95 6.49s latency and 28 cache-command timeouts; every slow request had one PostgreSQL query and 0 HTTP 5xx.
• Staging already runs this exact pool implementation with clean health checks and no PostgreSQL acquisition errors.

Validation

• bun run lint
• bun run check-types (34/34 tasks)
• bun run test (25/25 tasks; 3,905 AI-package tests passed, 0 failed)
• Pre-push test hook passed
• Independent focused review requested before merge

Rollout and rollback

• After merge, verify production health plus Axiom error rate, PostgreSQL timing, and /links/create p95 under the next real burst.
• Rollback is limited to reverting these two commits if pool acquisition or connection counts regress.

Issue

Emergency maintainer production-incident hotfix; no public issue contains the operational details.

AI disclosure

OpenAI Codex assisted with incident investigation, the minimal cherry-pick, validation, and PR preparation. The implementation was already exercised in staging and was verified with the full local suite.

---

Summary by cubic

Restore an elastic pg.Pool for the shared PostgreSQL client to fix concurrent query contention and reduce timeouts. The pool is lazy (min 0), defaults to 50 max connections per replica, closes idle after 30s, and bounds acquisition to 10s.

• Bug Fixes
  • Replaced pg.Client with pg.Pool and moved query timing to the pool.
  • Preserved PgBouncer-compatible URL normalization, warmup, error reporting, and graceful shutdown.
  • Added DB_POOL_MAX env support to override the pool ceiling.

^{Written for commit 6d65b6f. Summary will update on new commits.}

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
dashboard	Ready	Preview	Aug 1, 2026 6:10pm
databuddy-status	Ready	Preview	Aug 1, 2026 6:10pm
documentation	Ready	Preview	Aug 1, 2026 6:10pm

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. 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: Repository UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 10f22c09-52f8-4661-bad4-91d105cb9ea4

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

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[greptile-apps]

Greptile Summary

The PR restores a lazily initialized, bounded PostgreSQL pool in place of the shared single client while retaining query instrumentation, error handling, warmup, and shutdown behavior.

• Configures up to 50 pooled connections with 30-second idle eviction and a 10-second acquisition timeout.
• Preserves PostgreSQL URL normalization and application naming.
• Updates warmup and graceful shutdown to acquire, release, and close pooled clients.

Confidence Score: 5/5

The PR appears safe to merge, with no concrete changed-code failure identified in the restored pool implementation.

The shared database path now uses the Promise-based pg pool API expected by Drizzle, while the configured acquisition bounds and acquire-release/end lifecycle preserve the existing public behavior.

Important Files Changed

Filename	Overview
packages/db/src/client.ts	Replaces the process-wide PostgreSQL Client with a bounded lazy Pool and adapts query timing, warmup, error handling, and shutdown without an accepted changed-code defect.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Caller[Application query] --> Proxy[Lazy db proxy]
  Proxy --> Drizzle[Drizzle ORM]
  Drizzle --> Pool[Shared pg.Pool]
  Pool --> PgBouncer[PgBouncer / PostgreSQL]
  Warmup[warmPostgres] --> Acquire[Acquire and release client]
  Acquire --> Pool
  Shutdown[shutdownPostgres] --> End[pool.end]
  End --> Pool

Loading

_{Reviews (1): Last reviewed commit: "fix(db): raise postgres pool ceiling" | Re-trigger Greptile}

[izadoesdev]

Closing because this PR was mistakenly opened against main. Project PRs target staging, and the two PostgreSQL pool changes are already present on staging (0fe9cf895, 24ed27490). Retargeting this divergent hotfix branch would include unrelated changes, so there is no valid staging diff to review here.