Dasha 1.10 reads Patroni clusters from etcd or Consul and forecasts replication slot loss
Dasha, an open-source performance dashboard for fleets of PostgreSQL servers, has released version 1.10, announced on the PostgreSQL news site on 9 November. The tool connects with a read-only role and installs nothing on the database hosts, neither an agent nor an extra extension. One instance watches many clusters and all their replicas, and because most PostgreSQL statistics are kept per instance and not replicated, it reads every host and combines the results: an index that looks idle on the primary can carry the whole read load on a standby.
The biggest addition is Patroni discovery. Dasha reads the keys Patroni keeps in its distributed configuration store, etcd v3 or Consul, and turns each Patroni scope into a cluster whose members become hosts. New clusters, lost replicas and created or dropped databases appear within the refresh interval. The store is only read, and TLS with client certificates works for both backends. Standby clusters are linked to the cluster they replicate from by matching the PostgreSQL system identifier, so a standby behind a virtual IP is still found.
For PostgreSQL 14 and later there is an Activity and Waits view that samples pg_stat_activity every one to ten seconds and charts active sessions by CPU, wait event type and idle-in-transaction. Statements are grouped by query_id, and however many people watch the page, each Dasha instance queries a host's activity view at most once per second.
A new forecast estimates when each replication slot will be invalidated. Dasha measures the WAL rate over a window between 30 seconds and five minutes, then shows per slot the WAL retained, the remaining max_slot_wal_keep_size reserve and the time left. Other changes add COPY progress with an estimated finish time, let the time-series Health Score run on plain postgres_exporter, node_exporter and pgbouncer_exporter, and stop penalising tables with no writes for seven days for dead rows and missed vacuums; those are listed separately for a single freeze-and-analyse pass.

Why it matters
A slot that silently retains WAL until the disk fills is one of the classic PostgreSQL outages. Turning the reserve into a time-to-loss figure gives operators a deadline instead of a disk alarm.