Parallelize pipelined cluster Redis reads across slots in runPipelinedByKey and mGetClusterRaw

[vijaygupta18]

🤖 Argus proposed fix

Fix confidence: HIGH (score 0.85)
Reasons: RCA high-confidence, exact line identified, confirmed pattern matched, localized diff (1f/70l)

Root cause

Commit e297d90aa8 introduced shared-kernel pipelined cluster-Redis helpers (runPipelinedByKey, mGetClusterRaw, zAddIfPossibleMany) into the always-on driver-pool reservation path. These helpers group keys by Redis cluster slot but then execute all slot groups sequentially inside a single runHedis action. For driver-pool batches that touch many slots, this serializes many Redis round-trips, causing BPP request handling to slow down, containers to restart, and 5xx/503 responses, which delays the allocator's SendSearchRequestToDriver job pickups and triggers AllocatorJobPickupDelayHigh.

Evidence

see RCA

Rollback

Revert the shared-kernel change, or roll back the BPP deployment to the previous stable image (e.g. beckn-driver-offer-bpp-production-af1f94v1 or 220c1ev2) until the fix is deployed.

---

Draft PR — review and merge manually. Generated by Argus.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

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.}