Dasha 1.10: Patroni auto-discovery, wait analysis and replication slot forecasts for PostgreSQL

Dasha is an open source performance dashboard for PostgreSQL fleets. It connects to your clusters with a read-only role, shows what the databases are doing right now, and explains what to do about it. Nothing is installed on the database hosts: no agent, and no extension beyond the ones you already have.

One instance serves many clusters and all of their replicas. Most PostgreSQL statistics are per-instance and are not replicated, so Dasha reads every host of a cluster and reasons over the combined picture. An index that looks unused on the primary may be serving the entire read workload on a standby.

New since the last announcement

Patroni clusters from etcd v3 or Consul. Dasha reads the keys Patroni keeps in its DCS and turns every Patroni scope into a cluster: the members become hosts and the databases are listed from PostgreSQL. New clusters, lost replicas, created and dropped databases show up within the refresh interval. The DCS is only read, never written. User, password, port and database filters can be overridden per scope, and TLS with client certificates is supported for both etcd and Consul. Member names appear in the host selector next to the address.
Standby cluster links. A Patroni standby cluster shows the cluster it replicates from, and the primary shows its standbys, on the Summary and Replication pages. Clusters are matched by the PostgreSQL system identifier, so a standby behind a VIP is found too.
Activity & Waits (PostgreSQL 14+). A live chart of active sessions by class: CPU, wait event type (IO, Lock, LWLock, ...) and idle in transaction, sampled from pg_stat_activity every 1-10 seconds. Below it are the top statements, sessions, databases, users, applications and wait events; a click on a value adds a filter, and a statement links to its query report and the lock tree. Statements are grouped by query_id. No extension is needed, and each Dasha instance reads a host's pg_stat_activity at most once per second however many people watch the page.
Replication slot invalidation forecast. Dasha measures the WAL rate over a chosen window (30 seconds to 5 minutes) and shows, for every slot, the WAL it retains, the remaining max_slot_wal_keep_size reserve and the time until the slot is lost. The measurement takes two LSN samples by the server clock and holds no connection in between.
COPY progress (PostgreSQL 14+). Table, source type, bytes and rows processed, rows excluded and skipped, and a percentage with ETA when the total is known: the file size for COPY FROM a server-side file, the row estimate for COPY TO.
Health Score from the stock exporters. The time-series mode of the Health Score now works with plain postgres_exporter, node_exporter and pgbouncer_exporter, so an existing Prometheus or VictoriaMetrics setup is enough. Metrics of Patroni clusters are mapped automatically by scope and member name.
Cold tables are not penalized. Tables with no writes for longer than an inactivity threshold (7 days by default) no longer lower the score for dead rows and missing vacuum. A separate recommendation lists those that need a single VACUUM (FREEZE, ANALYZE), and the Maintenance page can filter all, active or cold tables.

Features

Health Score: a composite 0-100 score per instance across eight categories, with prioritized recommendations tied to a specific database. Optionally computed from Prometheus/VictoriaMetrics time series, with a trend and a seasonal baseline.
Query analysis: top queries by execution time and WAL, running and blocked queries with wait events, and pg_stat_statements snapshots. Snapshots can be saved to a storage database, shared by URL, and diffed against another snapshot or against live data.
Index advisor and analysis: missing B-tree indexes proposed from pg_stat_statements of every host, optionally backed by auto_explain plans from the log. Bloat, duplicates, invalid indexes, and an answer to "may I drop this index?" that accounts for statistics resets, per-host counters and partitioned indexes rather than trusting idx_scan = 0.
Schema checks: 17 checks for structural defects, such as sequences close to exhaustion, tables without a primary key or a public schema anyone can create objects in, each with a severity and remediation SQL.
I/O analysis built on pg_stat_io (PostgreSQL 16+), with WAL on PostgreSQL 18.
Log insights: log search in OpenSearch, Elasticsearch, VictoriaLogs and Yandex Managed PostgreSQL, event categories, auto_explain plan trees with findings, and plan regressions between two time windows.
Automatic snapshots: a separate daemon captures query and instance-wide lock-contention snapshots on activity spikes or primary/replica role changes. The evidence is there the next morning when someone asks what happened at 03:00.
Tables and maintenance: size with a TOAST breakdown, per-table describe, and autovacuum and wraparound thresholds computed with PostgreSQL's own formula. Also progress of VACUUM, ANALYZE, CREATE INDEX, CLUSTER, base backups and COPY, and hot tables and indexes derived from scheduled delta snapshots.
Cluster discovery: a static list, Patroni clusters from etcd or Consul, or Yandex Managed Service for PostgreSQL; databases are discovered from the cluster itself.
Access control: three auth modes (open, static API keys, or OIDC with encrypted session cookies), Casbin RBAC, personal access tokens and per-identity rate limiting.
MCP connector: a read-only MCP server that gives AI assistants the fleet diagnostics. It has 35 tools, including index recommendations, schema checks, I/O, log search, plan regressions and live activity sampling. It also has 5 prompts, per-user token passthrough and an embedded knowledge base of the scoring rules.

The interface is available in English, Russian and German.

Requirements

PostgreSQL 14 or newer (tested on 14-19b) for the monitored clusters, reachable with a role holding pg_monitor and CONNECT on every monitored database.
pg_stat_statements (or pgpro_stats on Postgres Pro) for query statistics; pgstattuple is optional.
Activity & Waits needs compute_query_id set to on, or to auto with pg_stat_statements or pgpro_stats preloaded.
The I/O page requires PostgreSQL 16 or newer.
Optional: etcd v3 or Consul read access for Patroni discovery.
Optional: a log source (OpenSearch, Elasticsearch, VictoriaLogs or Yandex Managed PostgreSQL) for log search and plan insights, with auto_explain enabled for plan analysis.
Optional: a separate PostgreSQL database for snapshots, hot-object history and personal access tokens. Everything else works without it.