Files
notifkit/drizzle/0006_message_logs_redundant_indexes.sql
coo1estguy 880fe32b03 perf: rebuild profiling harness and reduce per-message pipeline cost
- Rebuild profiling suite with compose CLI, in-network k6 load generator, and production-shaped tiers
- Add fused pipeline mode for in-memory single-user stage hand-offs
- Route directly by contact ID and defer template rendering to delivery
- Add Redis connection sharing, auto-pipelining, stream read coalescing, and chunked provider throttling
- Switch hot message_logs and delivery_outbox inserts to unnest bulk SQL queries
- Drop redundant indexes on message_logs
- Optimize worker batching, flush concurrency, and dataloader windowing
2026-09-29 00:35:01 +05:30

8 lines
476 B
SQL

-- Both indexes are strict prefixes of others on the same table:
-- message_logs_project_idx (project_id) <- every composite leading with project_id
-- task_idx (task_id) <- task_channel_attempt_uidx (task_id, ...)
-- message_logs takes two rows per delivered notification, so each index is
-- paid for on the hottest insert path in the system.
DROP INDEX IF EXISTS "message_logs_project_idx";--> statement-breakpoint
DROP INDEX IF EXISTS "task_idx";