Skip to content

Commit a24991e

Browse files
BupycHukmatejkubinec
BupycHuk
and
matejkubinec
authored
PMM-13868 Add custom PostgreSQL queries for replication and monitoring (#320)
* PMM-13868 Add custom PostgreSQL queries for replication and monitoring This commit introduces YAML configurations for various PostgreSQL metrics, including replication slots, WAL receiver, database sizes, and activity monitoring. These queries enhance the ability to collect and track replication performance and database activity. * PMM-13868 Set 'master' flag to false for pg_custom_database_size_custom The 'master' flag was changed to false in the pg_custom_database_size_custom query. This ensures the query no longer runs exclusively on the master database, aligning it with the desired database configuration and behavior. --------- Co-authored-by: Matej Kubinec <[email protected]>
1 parent a477908 commit a24991e

File tree

1 file changed

+231
-0
lines changed

1 file changed

+231
-0
lines changed

‎queries-hr.yml

Lines changed: 231 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,231 @@
1+
pg_custom_replication_slots:
2+
query: |
3+
SELECT
4+
slot_name,
5+
plugin,
6+
slot_type,
7+
active,
8+
restart_lsn::pg_lsn - '0/0'::pg_lsn AS restart_lsn_bytes,
9+
confirmed_flush_lsn::pg_lsn - '0/0'::pg_lsn AS confirmed_flush_lsn_bytes
10+
FROM pg_replication_slots;
11+
metrics:
12+
- slot_name:
13+
usage: "LABEL"
14+
description: "Name of the replication slot."
15+
- plugin:
16+
usage: "LABEL"
17+
description: "Plugin associated with the replication slot."
18+
- slot_type:
19+
usage: "LABEL"
20+
description: "Type of replication slot (e.g., logical, physical)."
21+
- active:
22+
usage: "GAUGE"
23+
description: "Whether the slot is active (1 for active, 0 for inactive)."
24+
- restart_lsn_bytes:
25+
usage: "GAUGE"
26+
description: "Restart LSN in bytes for the replication slot."
27+
- confirmed_flush_lsn_bytes:
28+
usage: "GAUGE"
29+
description: "Confirmed flush LSN in bytes for the replication slot."
30+
31+
pg_custom_stat_wal_receiver:
32+
query: |
33+
SELECT
34+
status,
35+
sender_host AS primary_host,
36+
written_lsn::pg_lsn - '0/0'::pg_lsn AS written_lsn_bytes,
37+
latest_end_lsn::pg_lsn - '0/0'::pg_lsn AS latest_end_lsn_bytes,
38+
latest_end_lsn::pg_lsn - pg_last_wal_replay_lsn()::pg_lsn AS lag_bytes,
39+
EXTRACT(EPOCH FROM latest_end_time) AS latest_end_time_seconds,
40+
EXTRACT(EPOCH FROM (now() - latest_end_time)) AS lag_time_seconds,
41+
EXTRACT(EPOCH FROM last_msg_send_time) AS last_msg_send_time_seconds,
42+
EXTRACT(EPOCH FROM last_msg_receipt_time) AS last_msg_receipt_time_seconds
43+
FROM pg_stat_wal_receiver;
44+
metrics:
45+
- status:
46+
usage: "LABEL"
47+
description: "WAL receiver's status (e.g., streaming, stopped)."
48+
- primary_host:
49+
usage: "LABEL"
50+
description: "Primary instance host."
51+
- written_lsn_bytes:
52+
usage: "GAUGE"
53+
description: "Bytes of WAL data written by the WAL receiver."
54+
- latest_end_lsn_bytes:
55+
usage: "GAUGE"
56+
description: "Bytes of WAL data received by the WAL receiver."
57+
- lag_bytes:
58+
usage: "GAUGE"
59+
description: "WAL replication lag in bytes."
60+
- latest_end_time_seconds:
61+
usage: "GAUGE"
62+
description: "Timestamp of the latest received WAL segment."
63+
- lag_time_seconds:
64+
usage: "GAUGE"
65+
description: "Replication lag time (difference between now and latest_end_time)."
66+
- last_msg_send_time_seconds:
67+
usage: "GAUGE"
68+
description: "Timestamp of the last message sent in seconds since the epoch."
69+
- last_msg_receipt_time_seconds:
70+
usage: "GAUGE"
71+
description: "Timestamp of the last message received in seconds since the epoch."
72+
73+
74+
pg_custom_database_size_custom:
75+
query: "SELECT pg_database.datname, pg_database_size(pg_database.datname) as bytes FROM pg_database"
76+
master: false
77+
cache_seconds: 30
78+
metrics:
79+
- datname:
80+
usage: "LABEL"
81+
description: "Name of the database"
82+
- bytes:
83+
usage: "GAUGE"
84+
description: "Disk space used by the database"
85+
86+
pg_custom_replication_wal:
87+
query: |
88+
SELECT
89+
pg_last_wal_receive_lsn() AS received_lsn,
90+
pg_last_wal_replay_lsn() AS replayed_lsn,
91+
pg_current_wal_lsn() AS current_lsn,
92+
pg_current_wal_lsn() - pg_last_wal_replay_lsn() AS lag_bytes;
93+
metrics:
94+
- received_lsn:
95+
usage: "GAUGE"
96+
description: "Last WAL location received by the standby server."
97+
- replayed_lsn:
98+
usage: "GAUGE"
99+
description: "Last WAL location replayed by the standby server."
100+
- current_lsn:
101+
usage: "GAUGE"
102+
description: "Current WAL location on the primary server."
103+
- lag_bytes:
104+
usage: "GAUGE"
105+
description: "Current WAL replication lag in bytes."
106+
107+
pg_custom_stat_replication:
108+
query: |
109+
SELECT
110+
pid,
111+
usename,
112+
application_name,
113+
client_addr,
114+
state,
115+
flush_lsn::pg_lsn - '0/0'::pg_lsn AS flush_lsn_bytes,
116+
write_lsn::pg_lsn - '0/0'::pg_lsn AS write_lsn_bytes,
117+
replay_lsn::pg_lsn - '0/0'::pg_lsn AS replay_lsn_bytes,
118+
EXTRACT(EPOCH FROM write_lag) AS write_lag_seconds,
119+
EXTRACT(EPOCH FROM flush_lag) AS flush_lag_seconds,
120+
EXTRACT(EPOCH FROM replay_lag) AS replay_lag_seconds,
121+
sync_state
122+
FROM pg_stat_replication;
123+
metrics:
124+
- pid:
125+
usage: "LABEL"
126+
description: "Process ID of the replication connection."
127+
- usename:
128+
usage: "LABEL"
129+
description: "Name of the user connected for replication."
130+
- application_name:
131+
usage: "LABEL"
132+
description: "Application name of the client."
133+
- client_addr:
134+
usage: "LABEL"
135+
description: "Client IP address."
136+
- state:
137+
usage: "LABEL"
138+
description: "State of the replication connection."
139+
- flush_lsn_bytes:
140+
usage: "GAUGE"
141+
description: "Flush LSN in bytes."
142+
- write_lsn_bytes:
143+
usage: "GAUGE"
144+
description: "Write LSN in bytes."
145+
- replay_lsn_bytes:
146+
usage: "GAUGE"
147+
description: "Replay LSN in bytes."
148+
- write_lag_seconds:
149+
usage: "GAUGE"
150+
description: "Write lag in seconds."
151+
- flush_lag_seconds:
152+
usage: "GAUGE"
153+
description: "Flush lag in seconds."
154+
- replay_lag_seconds:
155+
usage: "GAUGE"
156+
description: "Replay lag in seconds."
157+
- sync_state:
158+
usage: "LABEL"
159+
description: "Synchronization state (e.g., async, sync)."
160+
161+
pg_custom_stat_activity_walsender:
162+
query: |
163+
SELECT
164+
pid,
165+
usename,
166+
application_name,
167+
client_addr,
168+
extract(epoch from backend_start) AS backend_start_unix,
169+
state
170+
FROM pg_stat_activity
171+
WHERE backend_type = 'walsender';
172+
metrics:
173+
- pid:
174+
usage: "GAUGE"
175+
description: "Process ID of the WAL sender"
176+
- usename:
177+
usage: "LABEL"
178+
description: "User name associated with the WAL sender"
179+
- application_name:
180+
usage: "LABEL"
181+
description: "Application name of the WAL sender"
182+
- client_addr:
183+
usage: "LABEL"
184+
description: "Client IP address of the WAL sender"
185+
- backend_start_unix:
186+
usage: "GAUGE"
187+
description: "Start time of the backend as a Unix timestamp"
188+
- state:
189+
usage: "LABEL"
190+
description: "Current state of the WAL sender process"
191+
192+
pg_custom_stat_subscription:
193+
query: "SELECT subid, subname, pid, received_lsn, last_msg_send_time, last_msg_receipt_time FROM pg_stat_subscription;"
194+
metrics:
195+
- subid:
196+
usage: "LABEL"
197+
description: "Subscription ID"
198+
- subname:
199+
usage: "LABEL"
200+
description: "Subscription Name"
201+
- pid:
202+
usage: "GAUGE"
203+
description: "Process ID of subscription worker"
204+
- received_lsn:
205+
usage: "GAUGE"
206+
description: "Last received LSN"
207+
- last_msg_send_time:
208+
usage: "GAUGE"
209+
description: "Last message sent time"
210+
- last_msg_receipt_time:
211+
usage: "GAUGE"
212+
description: "Last message receipt time"
213+
214+
pg_custom_publication:
215+
query: "SELECT pubname, puballtables, pubinsert, pubupdate, pubdelete FROM pg_publication;"
216+
metrics:
217+
- pubname:
218+
usage: "LABEL"
219+
description: "Publication Name"
220+
- puballtables:
221+
usage: "GAUGE"
222+
description: "All tables published"
223+
- pubinsert:
224+
usage: "GAUGE"
225+
description: "Insert operations published"
226+
- pubupdate:
227+
usage: "GAUGE"
228+
description: "Update operations published"
229+
- pubdelete:
230+
usage: "GAUGE"
231+
description: "Delete operations published"

0 commit comments

Comments
 (0)