diff --git a/deployment/query-exporter/config.yaml b/deployment/query-exporter/config.yaml index bb73663be8..bca54ec3b5 100644 --- a/deployment/query-exporter/config.yaml +++ b/deployment/query-exporter/config.yaml @@ -36,7 +36,7 @@ metrics: pg_connections_stuck: type: gauge - description: Count of connections with state of idle in transaction + description: Count of connections that have been idle in transaction for more than 1 minute pg_long_running_queries_count: type: gauge @@ -54,7 +54,7 @@ metrics: pg_dead_tuple_percent: type: gauge labels: [table_name] - description: Table ratio of dead tuples to live ones a ratio > 10-20% indicates not aggressive enough autovacuum + description: Percentage of dead tuples in tables with at least 10,000 dead tuples (0 otherwise); a sustained value > 20-30% indicates autovacuum is not keeping up pg_last_autovacuum_hours: type: gauge @@ -70,6 +70,14 @@ metrics: type: gauge description: DB size in MB + pg_asset_raw_data_size: + type: gauge + description: Asset table heap size in MB (steady growth without a matching asset count increase indicates bloat) + + pg_asset_disk_size: + type: gauge + description: Asset table total size including TOAST and indexes in MB + pg_datapoint_raw_data_size: type: gauge description: Asset datapoint table raw uncompressed size in MB @@ -238,7 +246,10 @@ queries: FROM (SELECT current_setting('max_connections')::int AS max_limit) max_conn, (SELECT count(*) AS total_used FROM pg_stat_activity) active, - (SELECT count(*) AS idle_in_xact_count FROM pg_stat_activity WHERE state = 'idle in transaction') idle_xact; + -- Ignore transactions that are only briefly idle between statements + (SELECT count(*) AS idle_in_xact_count FROM pg_stat_activity + WHERE state = 'idle in transaction' + AND now() - state_change > interval '1 minute') idle_xact; long_running_queries: interval: 60 # 1 minute @@ -265,8 +276,13 @@ queries: THEN ROUND((n_tup_hot_upd::numeric / n_tup_upd) * 100, 2) ELSE 0.00 END AS pg_hot_update_percent, - -- Calculate the percentage of the table that is dead space - ROUND((n_dead_tup::numeric / NULLIF(n_live_tup + n_dead_tup, 0)) * 100, 2) AS pg_dead_tuple_percent, + -- Calculate the percentage of the table that is dead space; report 0 for tables with fewer than 10,000 dead + -- tuples, as small tables naturally sit at a high percentage (autovacuum waits for 50 + 20% of live tuples) + CASE + WHEN n_dead_tup < 10000 + THEN 0.00 + ELSE ROUND((n_dead_tup::numeric / NULLIF(n_live_tup + n_dead_tup, 0)) * 100, 2) + END AS pg_dead_tuple_percent, -- Calculate hours since last autovacuum ROUND((EXTRACT(EPOCH FROM (now() - last_autovacuum)) / 3600.0)::numeric, 1) AS pg_last_autovacuum_hours, -- Calculate hours since last autoanalyze @@ -281,6 +297,15 @@ queries: sql: | SELECT ROUND(pg_database_size(current_database()) / 1048576.0, 0) as pg_db_disk_size; + asset_table_size: + interval: 3600 # Hourly + databases: [openremote] + metrics: [pg_asset_raw_data_size, pg_asset_disk_size] + sql: | + SELECT + ROUND(pg_relation_size('openremote.asset') / 1048576.0, 1) AS pg_asset_raw_data_size, + ROUND(pg_total_relation_size('openremote.asset') / 1048576.0, 1) AS pg_asset_disk_size; + datapoint_table_size: interval: 3600 # Hourly databases: [openremote]