Commit 764bfc38 authored by Jérome Perrin's avatar Jérome Perrin

WIP grafana

parent 35171704
......@@ -15,7 +15,7 @@
[instance-profile]
filename = instance.cfg.in
md5sum = 6ff69aef5a7185f880359a78965e072e
md5sum = b0bd4494451460b9cd2df7d333e5a157
[influxdb-config-file]
filename = influxdb-config-file.cfg.in
......
......@@ -163,7 +163,7 @@ ssl-cert-file = ${grafana-certificate:cert-file}
recipe = slapos.cookbook:wrapper
command-line =
{{ grafana_bin }} -config ${grafana-config-file:output} -homepath {{ grafana_homepath }}
{{ grafana_bin }} server -config ${grafana-config-file:output} -homepath {{ grafana_homepath }}
wrapper-path = ${directory:service}/grafana
[grafana-certificate]
......@@ -267,7 +267,12 @@ init =
"name_override": f"{partition['name']}-mysql",
"servers": [dsn],
"gather_innodb_metrics": True,
"tags": dict(partition.get("static-tags", {}), app=application["name"]),
"tags": dict(
partition.get("static-tags", {}),
app=application["name"],
name=partition["name"],
partition=partition["reference"],
),
}
)
if partition["type"] == "erp5/mariadb":
......@@ -278,39 +283,44 @@ init =
"dsn": dsn,
"query": [
{
"query": "select count(*) as message_count from message",
"query": """
select 'message' as cmf_activity_queue, count(*) as message_count from message
union all select 'message_queue' as cmf_activity_queue, count(*) as message_count from message_queue
""",
"field_columns_include": ["message_count"],
},
{
"query": "select count(*) as message_queue_count from message_queue",
"field_columns_include": ["message_queue_count"],
},
{
"query": "select count(*) as message_failed_count from message where processing_node=-2",
"field_columns_include": ["message_failed_count"],
},
{
"query": "select count(*) as message_queue_failed_count from message_queue where processing_node=-2",
"field_columns_include": ["message_queue_failed_count"],
"tag_columns_include": ["cmf_activity_queue"],
},
{
"query": """
select cast(coalesce(max(UNIX_TIMESTAMP(now()) - UNIX_TIMESTAMP(message.date)), 0) as int)
as message_waiting_time from message
where processing_node in (-1, 0) and message not like '%after_tag%'
select 'message' as cmf_activity_queue, count(*) as failed_message_count
from message where processing_node between -2 and -10
union all select 'message_queue' as cmf_activity_queue, count(*) as failed_message_count
from message_queue where processing_node between -2 and -10
""",
"field_columns_include": ["message_waiting_time"],
"field_columns_include": ["failed_message_count"],
"tag_columns_include": ["cmf_activity_queue"],
},
{
"query": """
select cast(coalesce(max(UNIX_TIMESTAMP(now()) - UNIX_TIMESTAMP(message.date)), 0) as int)
as waiting_time, 'message' as cmf_activity_queue
from message where processing_node in (-1, 0) and message.message not like '%after_tag%'
union all
select cast(coalesce(max(UNIX_TIMESTAMP(now()) - UNIX_TIMESTAMP(message_queue.date)), 0) as int)
as message_queue_waiting_time from message_queue
where processing_node in (-1, 0) and message not like '%after_tag%'
as waiting_time, 'message_queue' as cmf_activity_queue
from message_queue where processing_node in (-1, 0) and message.message_queue not like '%after_tag%'
""",
"field_columns_include": ["message_queue_waiting_time"],
}
"field_columns_include": ["waiting_time"],
"tag_columns_include": ["cmf_activity_queue"],
},
],
"tags": dict(partition.get("static-tags", {}), app=application["name"]),
"tags": dict(
partition.get("static-tags", {}),
app=application["name"],
name=partition["name"],
partition=partition["reference"],
),
}
)
......@@ -326,7 +336,12 @@ init =
],
"grok_timezone": "Local",
"name_override": f"{partition['name']}",
"tags": dict(partition.get("static-tags", {}), app=application["name"]),
"tags": dict(
partition.get("static-tags", {}),
app=application["name"],
name=partition["name"],
partition=partition["reference"],
),
}
)
urls = application.get("urls", [])
......@@ -360,8 +375,6 @@ init =
"slapos": [{
"instance_root": application["instance-root"]}]}})
# TODO: supervisor process finder for
# https://github.com/influxdata/telegraf/tree/master/plugins/inputs/procstat ?
telegraf_slapos_input_command = self.options['telegraf-input-slapos-bin']
inputs["execd"].append({
"name_override": f"{application['name']}-processes",
......@@ -389,9 +402,8 @@ init =
processors["enum"].append({
"namepass": [ f"{application['name']}-processes"],
"mapping": [{
# "tag": "group", # TODO: rename this in input plugin # XXX I don't remember what this means
"tag": "slappart",
"dest": "partition",
"tag": "reference",
"dest": "name",
"value_mappings": partition_mapping,
}]})
......@@ -401,7 +413,6 @@ init =
# - [x] strip hashes from -on-watch
# - [x] activity metrics
# - [ ] alert dashboard
# - [ ] inclu "jerome-dev" partout ???
# - [ ] apdex
# - [ ] "job" is bad name in Explore
......@@ -476,8 +487,6 @@ command-line =
\
-boltdb.shipper.compactor.ring.instance-addr=${loki:ip} \
-boltdb.shipper.compactor.ring.instance-id=${:_buildout_section_name_} \
-common.embedded-cachering.instance-addr=${loki:ip} \
-common.embedded-cachering.instance-id=${:_buildout_section_name_} \
-distributor.ring.instance-addr=${loki:ip} \
-distributor.ring.instance-id=${:_buildout_section_name_} \
-frontend.instance-addr=${loki:ip} \
......@@ -790,8 +799,6 @@ install =
"stages": [
{
"multiline": {
# TODO
#"firstline": "^# Time: \\d{2}\\d{2}\\d{2}\\s\\d{1,2}\\:\\d{2}\\:\\d{2}",
"firstline": r"^# Time: \d{2}.*",
"max_wait_time": "3s"
}
......
......@@ -22,14 +22,6 @@ parts =
loki-config-file
loki-nginx-config-file
; [nodejs]
; <= nodejs-16.19.0
[gowork]
# XXX speed up development cycle by not rebuilding workspace on every software run
# XXX does not work ?
update-command =
[go_github.com_grafana_grafana]
<= go-git-package
......@@ -59,7 +51,7 @@ revision = v1.28.1-0-g3ea9ffbe2
<= go-git-package
go.importpath = github.com/perrinjerome/telegraf-input-slapos
repository = https://github.com/perrinjerome/telegraf-input-slapos
revision = v0.0.1-0-gf8981f3
revision = v0.0.2-0-gd4c5221
[go_github.com_prometheus_prometheus]
<= go-git-package
......@@ -92,7 +84,7 @@ telegraf-bin = ${:bin}/telegraf
telegraf-input-slapos-bin = ${:bin}/telegraf-input-slapos
influx-bin = ${:bin}/influx
influxd-bin = ${:bin}/influxd
grafana-bin = ${:bin}/grafana-server
grafana-bin = ${grafana:binpath}/grafana
grafana-homepath = ${grafana:homepath}
loki-bin = ${:bin}/loki
promtail-bin = ${:bin}/promtail
......@@ -119,6 +111,8 @@ command = bash -ce "
rm -rf ${buildout:directory}/.cache/yarn/
"
homepath = ${go_github.com_grafana_grafana:location}
# XXX "linux-amd64" is not portable here
binpath = ${go_github.com_grafana_grafana:location}/bin/linux-amd64
stop-on-error = true
[download-file-base]
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment