Merge pull request #488 from cognitedata/add-timeline-to-wal-metric

Metrics: add "wal.timeline" column to be able to detect failovers

pgwatch2/metrics/wal/10/metric.sql

@@ -8,5 +8,11 @@ select /* pgwatch2_generated */
     end as xlog_location_b,
   case when pg_is_in_recovery() then 1 else 0 end as in_recovery_int,
   extract(epoch from (now() - pg_postmaster_start_time()))::int8 as postmaster_uptime_s,
-  system_identifier::text as tag_sys_id
+  system_identifier::text as tag_sys_id,
+  case
+    when pg_is_in_recovery() = false then
+      ltrim(pg_walfile_name(pg_current_wal_lsn())::char(8), '0')::int
+    else
+      (select min_recovery_end_timeline::int from pg_control_recovery())
+    end as timeline
 from pg_control_system();

pgwatch2/metrics/wal/10/metric_su.sql

@@ -8,5 +8,11 @@ select /* pgwatch2_generated */
     end as xlog_location_b,
   case when pg_is_in_recovery() then 1 else 0 end as in_recovery_int,
   extract(epoch from (now() - coalesce((pg_stat_file('postmaster.pid', true)).modification, pg_postmaster_start_time())))::int8 as postmaster_uptime_s,
-  system_identifier::text as tag_sys_id
+  system_identifier::text as tag_sys_id,
+  case
+    when pg_is_in_recovery() = false then
+      ltrim(pg_walfile_name(pg_current_wal_lsn())::char(8), '0')::int
+    else
+      (select min_recovery_end_timeline::int from pg_control_recovery())
+    end as timeline
 from pg_control_system();

pgwatch2/sql/config_store/metric_definitions.sql

@@ -2638,7 +2638,13 @@ select
     end as xlog_location_b,
   case when pg_is_in_recovery() then 1 else 0 end as in_recovery_int,
   extract(epoch from (now() - pg_postmaster_start_time()))::int8 as postmaster_uptime_s,
-  system_identifier::text as tag_sys_id
+  system_identifier::text as tag_sys_id,
+  case
+    when pg_is_in_recovery() = false then
+      ltrim(pg_walfile_name(pg_current_wal_lsn())::char(8), '0')::int
+    else
+      (select min_recovery_end_timeline::int from pg_control_recovery())
+    end as timeline
 from pg_control_system();
 $sql$,
 $sql$
@@ -2652,7 +2658,13 @@ select
     end as xlog_location_b,
   case when pg_is_in_recovery() then 1 else 0 end as in_recovery_int,
   extract(epoch from (now() - coalesce((pg_stat_file('postmaster.pid', true)).modification, pg_postmaster_start_time())))::int8 as postmaster_uptime_s,
-  system_identifier::text as tag_sys_id
+  system_identifier::text as tag_sys_id,
+  case
+    when pg_is_in_recovery() = false then
+      ltrim(pg_walfile_name(pg_current_wal_lsn())::char(8), '0')::int
+    else
+      (select min_recovery_end_timeline::int from pg_control_recovery())
+    end as timeline
 from pg_control_system();
 $sql$,
 '{"prometheus_gauge_columns": ["in_recovery_int", "postmaster_uptime_s"]}'