Every database sends its engine metrics to its project in Observability, free and
automatically: there’s nothing to install or turn on, and they aren’t billed or counted against
Observability’s limits. They arrive every 30 seconds, in the project’s production environment, with
the database’s ID as the database label.
Dashboards
The database’s Monitoring tab in the console shows a built-in dashboard for its engine:
connections, transactions or queries, cache, locks, replication, storage, vacuum or InnoDB, and
statements. Duplicate as dashboard saves a copy you can edit, as an ordinary
dashboard. Viewing it needs obs.metrics.query.
Metrics
Metrics come from the widely used postgres_exporter and mysqld_exporter collectors, with their
names as PromQL shows them. They carry the labels database and role. Take the maximum over
role, for example max(sum by (role) (…)), so a query keeps working through failovers.
| Engine | Examples |
|---|---|
| PostgreSQL | pg_stat_activity_count, pg_settings_max_connections, pg_locks_count, pg_replication_lag_seconds, pg_database_size_bytes, pg_stat_database_xact_commit_total, pg_stat_database_xact_rollback_total, pg_stat_database_blks_hit_total, pg_stat_database_blks_read_total, pg_stat_database_deadlocks_total, pg_stat_database_temp_bytes_total, pg_stat_user_tables_n_dead_tup |
| Percona Server, MariaDB | mysql_global_status_threads_connected, mysql_global_status_threads_running, mysql_global_variables_max_connections, mysql_global_status_questions_total, mysql_global_status_commands_total, mysql_global_status_slow_queries_total, mysql_global_status_innodb_row_ops_total, mysql_global_status_innodb_buffer_pool_read_requests, mysql_global_status_innodb_buffer_pool_reads, mysql_global_status_innodb_row_lock_waits, mysql_global_status_innodb_deadlocks, mysql_binlog_size_bytes |
# Connections to one database
max(sum by (role) (pg_stat_activity_count{database="dbi_4k2m9x0a7qpe"}))
# Buffer cache hit ratio over 5 minutes
max(
sum by (role) (rate(pg_stat_database_blks_hit_total{database="dbi_4k2m9x0a7qpe"}[5m]))
/ (sum by (role) (rate(pg_stat_database_blks_hit_total{database="dbi_4k2m9x0a7qpe"}[5m]))
+ sum by (role) (rate(pg_stat_database_blks_read_total{database="dbi_4k2m9x0a7qpe"}[5m])))
)pg_stat_user_tables_* cover the postgres database.
Insights
The database’s Insights tab shows its top statements by time and by calls, and what sessions waited on, over the last 24 hours. It’s built from:
| Metric | Labels | Meaning |
|---|---|---|
db_digest_seconds_total |
digest |
Time spent running the statement |
db_digest_calls_total |
digest |
Times the statement ran |
db_wait_seconds_total |
event |
Time sessions waited, by wait event |
Each statement’s normalized text, with no literal values, is a log record in Obs with the service
ezgh.databases, the resource attribute database, and record attributes digest and text.
Logs
Engine logs are off until you choose which to export. Exported logs go to the project’s Observability logs and are billed as log ingest.
| Log | Holds |
|---|---|
error |
The engine’s error log. |
slow |
Statements slower than log_min_duration_statement (PostgreSQL) or long_query_time (Percona Server, MariaDB), 1 second unless you set it. |
audit |
Schema, user, role and privilege changes, and on Percona Server and MariaDB, connections. On PostgreSQL, connections go to the error log while log_connections is on. |
ezgh databases update dbi_4k2m9x0a7qpe --log-export error --log-export slow
ezgh databases update dbi_4k2m9x0a7qpe --no-log-exportsIn the API, it’s logExports, for example ["error", "slow"]. A change applies with no
interruption.
Events
A database’s events list its failovers and recoveries, maintenance, blue/green switchovers and failures, backups, and full storage, newest first:
ezgh databases events dbi_4k2m9x0a7qpe