Skip to content

Commit d6311f4

Browse files
mhenrixonmhenrixon
committed
Add Pgbus automated dashboard
Pgbus (https://github.com/mhenrixon/pgbus) is a PostgreSQL-native job processing and event bus for Rails, built on PGMQ. Its AppSignal integration (shipped in the gem, auto-enabled when the appsignal gem is loaded) reports metrics via an ActiveSupport::Notifications subscriber and a minutely probe. This dashboard surfaces job status, throughput, duration, queue depth and latency, dead letter queue depth, worker/process counts, message throughput, event handler status, worker recycling, database health, and stream broadcasts. Triggered by the pgbus_queue_job_count metric. Validated with `rake validate`.
1 parent e518bca commit d6311f4

3 files changed

Lines changed: 482 additions & 0 deletions

File tree

‎README.md‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -21,6 +21,7 @@ Automated Dashboards can be found in the `dashboards/` sub directory. Each direc
2121
- [NGINX](/dashboards/nginx/)
2222
- [Node.js](/dashboards/nodejs/)
2323
- [Oban](/dashboards/oban/)
24+
- [Pgbus](/dashboards/pgbus/) (via AppSignal for Ruby)
2425
- [PostgreSQL](/dashboards/postgres/) (via Vector)
2526
- [Process Memory](/dashboards/process_memory/)
2627
- [Puma](/dashboards/puma/)

‎dashboards/pgbus/README.md‎

Lines changed: 104 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,104 @@
1+
# Pgbus automated dashboard
2+
3+
[Pgbus](https://github.com/mhenrixon/pgbus) is a PostgreSQL-native job processing and event bus for Rails, built on [PGMQ](https://github.com/pgmq/pgmq).
4+
5+
When using the AppSignal integration that ships with the pgbus gem (enabled automatically when the `appsignal` gem is loaded), the Pgbus automated dashboard will appear. Pgbus reports metrics through an `ActiveSupport::Notifications` subscriber and a minutely probe. See the [pgbus AppSignal integration](https://github.com/mhenrixon/pgbus/blob/main/lib/pgbus/integrations/appsignal.rb) for the full list of metrics.
6+
7+
The following graphs are displayed in this automated dashboard:
8+
9+
- [Job status per queue](#job-status-per-queue)
10+
- [Throughput per job class](#throughput-per-job-class)
11+
- [Duration per job class](#duration-per-job-class)
12+
- [Queue depth](#queue-depth)
13+
- [Queue latency](#queue-latency)
14+
- [Dead letter queue depth](#dead-letter-queue-depth)
15+
- [Worker / Process count](#worker--process-count)
16+
- [Messages sent / read](#messages-sent--read)
17+
- [Event handler status](#event-handler-status)
18+
- [Worker recycling](#worker-recycling)
19+
- [Database health](#database-health)
20+
- [Stream broadcasts](#stream-broadcasts)
21+
22+
## Job status per queue
23+
24+
The "Job status per queue" graph shows the number of jobs that were processed, failed, or dead-lettered, broken down per queue.
25+
26+
This graph displays values from the `pgbus_queue_job_count` metric. This graph will show a line for each combination of values of the following metric tags:
27+
28+
- The **queue** in which each job was enqueued.
29+
- The **status** of the job (`processed`, `failed`, or `dead_lettered`).
30+
31+
## Throughput per job class
32+
33+
The "Throughput per job class" graph shows the number of jobs executed per job class.
34+
35+
This graph displays the `count` field of the `transaction_duration` metric, filtered to the `background` namespace, with a line per **action** (job class).
36+
37+
## Duration per job class
38+
39+
The "Duration per job class" graph shows the mean execution time per job class.
40+
41+
This graph displays the `mean` field of the `transaction_duration` metric, filtered to the `background` namespace, with a line per **action** (job class).
42+
43+
## Queue depth
44+
45+
The "Queue depth" graph shows the total number of messages waiting in each queue.
46+
47+
This graph displays values from the `pgbus_queue_depth` metric. This graph will show a line for each combination of values of the following metric tags:
48+
49+
- The **queue** being measured.
50+
- The **hostname** of the process that reported the metric.
51+
52+
## Queue latency
53+
54+
The "Queue latency" graph shows the age of the oldest message in each queue, in milliseconds. High latency means consumers can't keep up with the rate of incoming messages.
55+
56+
This graph displays values from the `pgbus_queue_latency` metric. This graph will show a line for each combination of values of the following metric tags:
57+
58+
- The **queue** being measured.
59+
- The **hostname** of the process that reported the metric.
60+
61+
## Dead letter queue depth
62+
63+
The "Dead letter queue depth" graph shows the number of messages that exceeded their max retries and were routed to a dead letter queue.
64+
65+
This graph displays values from the `pgbus_dlq_depth` metric, with a line per **hostname**.
66+
67+
## Worker / Process count
68+
69+
The "Worker / Process count" graph shows the number of active pgbus supervisor processes.
70+
71+
This graph displays values from the `pgbus_active_processes` metric, with a line per **hostname**.
72+
73+
## Messages sent / read
74+
75+
The "Messages sent / read" graph shows PGMQ message throughput: how many messages were enqueued (`pgbus_messages_sent`) and dequeued (`pgbus_messages_read`).
76+
77+
Both metrics show a line per **queue**.
78+
79+
## Event handler status
80+
81+
The "Event handler status" graph shows the number of processed and failed event-bus handler invocations.
82+
83+
This graph displays values from the `pgbus_event_count` metric. This graph will show a line for each combination of values of the following metric tags:
84+
85+
- The **handler** class that processed each event.
86+
- The **status** of the invocation (`processed` or `failed`).
87+
88+
## Worker recycling
89+
90+
The "Worker recycling" graph shows the number of worker recycle events, broken down by reason.
91+
92+
This graph displays values from the `pgbus_worker_recycled` metric, with a line per **reason** (`max_jobs`, `max_memory`, or `max_lifetime`). Pgbus recycles workers on these thresholds to prevent unbounded memory growth.
93+
94+
## Database health
95+
96+
The "Database health" graph shows the number of dead tuples in queue/archive tables (`pgbus_total_dead_tuples`) and how many tables currently need a vacuum (`pgbus_tables_needing_vacuum`).
97+
98+
Both metrics show a line per **hostname**. These are indicators of MVCC pressure on the PGMQ queue tables.
99+
100+
## Stream broadcasts
101+
102+
The "Stream broadcasts" graph shows the number of Turbo Stream broadcasts over the last 60 minutes (`pgbus_stream_broadcasts_60m`) and the number of active SSE connections (`pgbus_stream_active_connections`).
103+
104+
Both metrics show a line per **hostname**.

0 commit comments

Comments
 (0)