xin/nixos-config
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

modules/prometheus: refactor

Browse source
This commit is contained in:
xinyangli 2024-11-30 11:25:35 +08:00
parent ebf69d94dd
commit ec63c8b6cc
Signed by: xin
SSH key fingerprint: SHA256:UU5pRTl7NiLFJbWJZa+snLylZSXIz5rgHmwjzv8v4oE
26 changed files with 579 additions and 407 deletions
Download patch file Download diff file Expand all files Collapse all files

93
modules/nixos/prometheus/blackbox.nix
View file

@ -1,93 +0,0 @@
{
config,
lib,
pkgs,
...
}:
let
cfg = config.custom.prometheus;
in
{
config = lib.mkIf (cfg.enable && cfg.exporters.blackbox.enable) {
services.prometheus.exporters.blackbox = {
enable = true;
listenAddress = "127.0.0.1";
configFile = pkgs.writeText "blackbox.config.yaml" (
lib.generators.toYAML { } {
modules = {
tcp4_connect = {
prober = "tcp";
tcp = {
ip_protocol_fallback = false;
preferred_ip_protocol = "ip4";
tls = false;
};
timeout = "15s";
};
};
}
);
};
services.prometheus.scrapeConfigs = [
{
job_name = "blackbox";
scrape_interval = "1m";
metrics_path = "/probe";
params = {
module = [ "tcp4_connect" ];
};
static_configs = [
{
targets = [
"tok-00.namely.icu:8080"
"la-00.video.namely.icu:8080"
"auth.xinyang.life:443"
"home.xinyang.life:8000"
];
}
];
relabel_configs = [
{
source_labels = [ "__address__" ];
target_label = "__param_target";
}
{
source_labels = [ "__param_target" ];
target_label = "instance";
}
{
target_label = "__address__";
replacement = "127.0.0.1:${toString config.services.prometheus.exporters.blackbox.port}";
}
];
}
{
job_name = "blackbox_exporter";
static_configs = [
{ targets = [ "127.0.0.1:${toString config.services.prometheus.exporters.blackbox.port}" ]; }
];
}
];
custom.prometheus.ruleModules = [
{
name = "probe_alerts";
rules = [
{
alert = "HighProbeLatency";
expr = "probe_duration_seconds > 0.5";
for = "2m";
labels = {
severity = "warning";
};
annotations = {
summary = "High request latency on {{ $labels.instance }}";
description = "95th percentile of request latency is above 0.5 seconds for the last 2 minutes.";
};
}
];
}
];
};
}

40
modules/nixos/prometheus/caddy.nix
View file

@ -1,40 +0,0 @@
{ config, lib, ... }:
let
cfg = config.custom.prometheus;
in
{
config = lib.mkIf (cfg.enable && cfg.exporters.caddy.enable) {
services.caddy.globalConfig = lib.mkIf cfg.exporters.caddy.enable ''
servers {
metrics
}
'';
services.prometheus.scrapeConfigs = [
{
job_name = "caddy";
static_configs = [ { targets = [ "127.0.0.1:2019" ]; } ];
}
];
custom.prometheus.ruleModules = [
{
name = "caddy_alerts";
rules = [
{
alert = "UpstreamHealthy";
expr = "caddy_reverse_proxy_upstreams_healthy != 1";
for = "5m";
labels = {
severity = "critical";
};
annotations = {
summary = "Upstream {{ $labels.unstream }} not healthy";
};
}
];
}
];
};
}

146
modules/nixos/prometheus/default.nix
View file

@ -1,21 +1,16 @@
{
config,
pkgs,
lib,
...
}:
with lib;
let
inherit (lib)
mkEnableOption
mkOption
mkIf
types
;
cfg = config.custom.prometheus;
mkExporterOption =
enableOption:
(mkOption {
type = types.bool;
default = enableOption;
description = "Enable this exporter";
});
mkRulesOption = mkOption {
type = types.listOf (
@ -30,38 +25,36 @@ let
in
{
imports = [
./blackbox.nix
./caddy.nix
./gotosocial.nix
./immich.nix
./miniflux.nix
./ntfy-sh.nix
./restic.nix
./exporters.nix
./grafana.nix
];
options = {
custom.monitoring = {
grafana = {
enable = mkEnableOption "grafana with oauth only";
};
};
custom.prometheus = {
enable = mkEnableOption "Prometheus instance";
exporters = {
enable = mkOption {
type = types.bool;
default = false;
description = "Enable Prometheus exporter on every supported services";
};
restic.enable = mkExporterOption config.services.restic.server.enable;
blackbox.enable = mkExporterOption false;
caddy.enable = mkExporterOption config.services.caddy.enable;
gotosocial.enable = mkExporterOption config.services.gotosocial.enable;
immich.enable = mkExporterOption config.services.immich.enable;
miniflux.enable = mkExporterOption config.services.miniflux.enable;
ntfy-sh.enable = mkExporterOption config.services.ntfy-sh.enable;
};
grafana = {
enable = mkEnableOption "Grafana Cloud";
password_file = mkOption { type = types.path; };
};
ruleModules = mkRulesOption;
exporters = {
enable = mkEnableOption "prometheus exporter on all supported and enable guarded services";
node = {
enable = mkEnableOption "node exporter";
listenAddress = mkOption {
type = types.str;
default = "${config.networking.hostName}.coho-tet.ts.net";
};
};
blackbox = {
enable = mkEnableOption "blackbox exporter";
listenAddress = mkOption {
type = types.str;
default = "${config.networking.hostName}.coho-tet.ts.net";
};
};
};
};
};
@ -78,46 +71,18 @@ in
reverse_proxy 127.0.0.1:${toString config.services.prometheus.port}
'';
};
services.prometheus = mkIf cfg.enable {
enable = true;
port = 9091;
globalConfig.external_labels = {
hostname = config.networking.hostName;
};
remoteWrite = mkIf cfg.grafana.enable [
{
name = "grafana";
url = "https://prometheus-prod-24-prod-eu-west-2.grafana.net/api/prom/push";
basic_auth = {
username = "1340065";
password_file = cfg.grafana.password_file;
};
}
];
exporters = {
node = {
enable = true;
enabledCollectors = [
"loadavg"
"time"
"systemd"
];
listenAddress = "127.0.0.1";
port = 9100;
};
};
scrapeConfigs = [
{
job_name = "prometheus";
static_configs = [ { targets = [ "localhost:${toString config.services.prometheus.port}" ]; } ];
}
{
job_name = "node";
static_configs = [
{ targets = [ "localhost:${toString config.services.prometheus.exporters.node.port}" ]; }
];
}
];
alertmanager = {
@ -163,55 +128,8 @@ in
};
custom.prometheus.ruleModules = [
{
name = "system_alerts";
name = "prometheus_alerts";
rules = [
{
alert = "SystemdFailedUnits";
expr = "node_systemd_unit_state{state=\"failed\"} > 0";
for = "5m";
labels = {
severity = "critical";
};
annotations = {
summary = "Systemd has failed units on {{ $labels.instance }}";
description = "There are {{ $value }} failed units on {{ $labels.instance }}. Immediate attention required!";
};
}
{
alert = "HighLoadAverage";
expr = "node_load1 > 0.8 * count without (cpu) (node_cpu_seconds_total{mode=\"idle\"})";
for = "1m";
labels = {
severity = "warning";
};
annotations = {
summary = "High load average detected on {{ $labels.instance }}";
description = "The 1-minute load average ({{ $value }}) exceeds 80% the number of CPUs.";
};
}
{
alert = "HighTransmitTraffic";
expr = "rate(node_network_transmit_bytes_total{device!=\"lo\"}[5m]) > 100000000";
for = "1m";
labels = {
severity = "warning";
};
annotations = {
summary = "High network transmit traffic on {{ $labels.instance }} ({{ $labels.device }})";
description = "The network interface {{ $labels.device }} on {{ $labels.instance }} is transmitting data at a rate exceeding 100 MB/s for the last 1 minute.";
};
}
{
alert = "NetworkTrafficExceedLimit";
expr = ''increase(node_network_transmit_bytes_total{device!="lo",device!~"tailscale.*",device!~"wg.*",device!~"br.*"}[30d]) > 322122547200'';
for = "0m";
labels = {
severity = "critical";
};
annotations = {
summary = "Outbound network traffic exceed 300GB for last 30 day";
};
}
{
alert = "JobDown";
expr = "up == 0";

65
modules/nixos/prometheus/exporters.nix Normal file
View file

@ -0,0 +1,65 @@
{
config,
pkgs,
lib,
...
}:
let
inherit (lib) mkIf;
cfg = config.custom.prometheus.exporters;
in
{
config = {
services.prometheus.exporters.node = mkIf cfg.node.enable {
enable = true;
enabledCollectors = [
"loadavg"
"time"
"systemd"
];
listenAddress = cfg.node.listenAddress;
port = 9100;
};
services.prometheus.exporters.blackbox = mkIf cfg.blackbox.enable {
enable = true;
listenAddress = cfg.blackbox.listenAddress;
configFile = pkgs.writeText "blackbox.config.yaml" (
lib.generators.toYAML { } {
modules = {
tcp4_connect = {
prober = "tcp";
tcp = {
ip_protocol_fallback = false;
preferred_ip_protocol = "ip4";
tls = false;
};
timeout = "15s";
};
};
}
);
};
services.gotosocial.settings = {
metrics-enabled = true;
};
services.immich.environment = {
IMMICH_TELEMETRY_INCLUDE = "all";
};
services.restic.server.prometheus = true;
systemd.services.miniflux.environment.METRICS_COLLECTOR = "1";
services.ntfy-sh.settings.enable-metrics = true;
services.caddy.globalConfig = ''
servers {
metrics
}
admin ${config.networking.hostName}.coho-tet.ts.net:2019 {
}
'';
};
}

17
modules/nixos/prometheus/gotosocial.nix
View file

@ -1,17 +0,0 @@
{ config, lib, ... }:
let
cfg = config.custom.prometheus;
in
{
config = lib.mkIf (cfg.enable && cfg.exporters.gotosocial.enable) {
services.gotosocial.settings = {
metrics-enabled = true;
};
services.prometheus.scrapeConfigs = [
{
job_name = "gotosocial";
static_configs = [ { targets = [ "localhost:8080" ]; } ];
}
];
};
}

43
modules/nixos/prometheus/grafana.nix Normal file
View file

@ -0,0 +1,43 @@
{ config, lib, ... }:
let
cfg = config.custom.monitoring.grafana;
in
{
config = lib.mkIf cfg.enable {
sops.templates."grafana.env".content = ''
GF_AUTH_GENERIC_OAUTH_CLIENT_SECRET=${config.sops.placeholder."grafana/oauth_secret"}
'';
services.grafana = {
enable = true;
settings = {
server = {
http_addr = "127.0.0.1";
http_port = 3003;
root_url = "https://grafana.xinyang.life";
domain = "grafana.xinyang.life";
};
"auth.generic_oauth" = {
enabled = true;
name = "Kanidm";
client_id = "grafana";
scopes = "openid,profile,email,groups";
auth_url = "https://auth.xinyang.life/ui/oauth2";
token_url = "https://auth.xinyang.life/oauth2/token";
api_url = "https://auth.xinyang.life/oauth2/openid/grafana/userinfo";
use_pkce = true;
use_refresh_token = true;
allow_sign_up = true;
login_attribute_path = "preferred_username";
groups_attribute_path = "groups";
role_attribute_path = "contains(grafana_role[*], 'GrafanaAdmin') && 'GrafanaAdmin' || contains(grafana_role[*], 'Admin') && 'Admin' || contains(grafana_role[*], 'Editor') && 'Editor' || 'Viewer'";
allow_assign_grafana_admin = true;
auto_login = true;
};
"auth" = {
disable_login_form = true;
};
};
};
systemd.services.grafana.serviceConfig.EnvironmentFile = config.sops.templates."grafana.env".path;
};
}

25
modules/nixos/prometheus/immich.nix
View file

@ -1,25 +0,0 @@
{ config, lib, ... }:
let
cfg = config.custom.prometheus;
immichEnv = config.services.immich.environment;
metricPort =
if builtins.hasAttr "IMMICH_API_METRICS_PORT" immichEnv then
immichEnv.IMMICH_API_METRICS_PORT
else
8081;
in
{
config = lib.mkIf (cfg.enable && cfg.exporters.immich.enable) {
services.immich.environment = {
IMMICH_METRICS = "true";
};
services.prometheus.scrapeConfigs = [
{
job_name = "immich";
static_configs = [ { targets = [ "127.0.0.1:${toString metricPort}" ]; } ];
}
];
};
}

15
modules/nixos/prometheus/miniflux.nix
View file

@ -1,15 +0,0 @@
{ config, lib, ... }:
let
cfg = config.custom.prometheus;
in
{
config = lib.mkIf (cfg.enable && cfg.exporters.miniflux.enable) {
systemd.services.miniflux.environment.METRICS_COLLECTOR = "1";
services.prometheus.scrapeConfigs = [
{
job_name = "miniflux";
static_configs = [ { targets = [ config.systemd.services.miniflux.environment.LISTEN_ADDR ]; } ];
}
];
};
}

15
modules/nixos/prometheus/ntfy-sh.nix
View file

@ -1,15 +0,0 @@
{ config, lib, ... }:
let
cfg = config.custom.prometheus;
in
{
config = lib.mkIf (cfg.enable && cfg.exporters.ntfy-sh.enable) {
services.ntfy-sh.settings.enable-metrics = true;
services.prometheus.scrapeConfigs = [
{
job_name = "ntfy-sh";
static_configs = [ { targets = [ "ntfy.xinyang.life" ]; } ];
}
];
};
}

25
modules/nixos/prometheus/restic.nix
View file

@ -3,17 +3,26 @@ let
cfg = config.custom.prometheus;
in
{
config = lib.mkIf (cfg.enable && cfg.exporters.restic.enable) {
config = {
services.restic.server.prometheus = true;
services.prometheus.scrapeConfigs = [
(lib.mkIf cfg.exporters.restic.enable {
job_name = "restic";
static_configs = [ { targets = [ config.services.restic.server.listenAddress ]; } ];
})
];
custom.prometheus.templates.scrape.mkResticScrapes =
{
address,
port ? null,
...
}:
let
portStr = if port then ":${toString port}" else "";
in
[
(lib.mkIf cfg.exporters.restic.enable {
job_name = "restic";
static_configs = [ { targets = [ "${address}${portStr}" ]; } ];
})
];
custom.prometheus.ruleModules = [
custom.prometheus.templates.rules.mkResticRules = [
{
name = "restic_alerts";
rules = [