Views

Recern samples each open connection every 5 seconds and keeps the last hour in memory, so charts have history from the moment you connect. Views refresh on their own; see Shortcuts and settings to change how often.

Findings

Recern turns each server's statistics into findings, rated critical, warning or info: a session blocked by another, a long open transaction, an unused or duplicate index, a Redis server with no memory limit, a vector collection full of deleted records. Each finding explains itself and links to the view that shows the details.

PostgreSQL

• Overview: transactions per second, connections against max_connections, cache hits, blocked sessions, activity charts and findings.
• Sessions: client sessions by state, what each one is running, how long its transaction has been open, and whom it blocks.
• Locks: wait chains (the blocked statement and the session holding its lock) and every lock held, with transaction id locks hidden by default.
• Queries (with pg_stat_statements): statements ranked by total time, mean time, calls, rows, disk reads or temporary files.
• Tables: size with indexes and TOAST, dead rows, sequential scans, writes, and when each table was last vacuumed and analyzed.
• Indexes: unused, duplicate and redundant indexes with the space they take, and foreign keys without an index.
• Console: see Safe Mode and the console.

Redis, Valkey and Dragonfly

• Overview: operations per second, memory, hit ratio, clients and findings.
• Memory: where memory goes (data, key overhead, scripts, client buffers), peak, fragmentation and used memory over the last hour.
• Clients: connections by client name, library or address; idle, blocked and pub/sub clients.
• Slow log: logged commands on a timeline against the threshold, and total time by command.
• Keys: keys matching a pattern, read page by page with SCAN, with their type, time to live and approximate memory.
• Console: see Safe Mode and the console.

Recern Vector

• Overview: vectors, file size, memory, deleted records, collections and findings.
• Index: for each collection, nodes per HNSW layer, links per node against what m allows, unreachable records and memory.
• Recall: approximate search compared with exact search at several values of ef, with latency, so you can see what raising ef_search buys and costs.