CI, forges, registries and business apps: Jenkins, GitHub, Zot, Odoo
Failing builds, growing queues, red workflows, the registry and the ERP, read-only by the agent.
Principle
These integrations follow the rules of the others (Applications and services): in auto mode, a probe whose software answers at its usual address turns itself on; an address or an account set in the configuration states the intent, and failures are reported from the first run. Once the software has been seen, its <name>.can_connect check turns critical if it stops answering. Several targets are declared with instances (Configuration). Nothing is ever written: no build started, no workflow re-run, no image pushed, no Odoo database opened.
Each integration has its monitor templates (families “CI et forges”, “Dépôts et registres”, “Applications métier”) and its ready-made dashboard (Dashboards › Catalogue).
Jenkins
The agent reads the Jenkins JSON API every 60 seconds: /api/json (jobs and their latest builds, folders and multibranch projects down to three levels), /queue/api/json (build queue), /computer/api/json (executors and nodes), and, when the Metrics plugin is installed, its health checks. In auto mode, the probe turns on when a Jenkins answers on 127.0.0.1:8080 (X-Jenkins header).
integrations:
jenkins:
url: https://ci.example.com # default http://127.0.0.1:8080
username: hypha # read-only account
password: 11aa22bb… # its API token (Profile › Security › API Token)
jobs: ["prod/*", "deploy-*"] # optional: full-name patterns; empty = all
max_jobs: 500 # jobs published one by one, at most
Account permissions (matrix strategy): Overall/Read, Job/Read, Agent/ExtendedRead. Optional: Metrics/HealthCheck (Metrics plugin) for its health checks; Agent/Connect for node free space and response time, which Jenkins only shows with that permission. Agent/Connect also allows connecting a build agent: only grant it if node disk space matters more to you than that risk.
| Metric | Meaning |
|---|---|
jenkins.jobs.total, .failing, .unstable, .disabled, .building | tracked jobs, last build failed, unstable, disabled, running |
jenkins.job.health (job) | job health (0 to 100, the Jenkins weather) |
jenkins.job.last_build.duration (job) | duration of the last completed build, in seconds |
jenkins.job.last_success.age (job) | time since the last successful build |
jenkins.queue.size, .blocked, .buildable, .stuck, .oldest_age | build queue: waiting, blocked, ready, marked “stuck”, wait of the oldest |
jenkins.executors.total, .busy, .idle, .usage | executors, busy, idle, busy share (%) |
jenkins.nodes.total, .offline; jenkins.node.offline, .executors (node) | nodes and offline nodes |
jenkins.node.disk_free, .disk_free_percent, .temp_free, .response_time (node) | node free space and response time (Agent/Connect permission) |
jenkins.health_check (check) | Metrics plugin checks (1 healthy, 0 failing) |
jenkins.quieting_down | 1 while a shutdown is prepared |
Checks: jenkins.can_connect; jenkins.job.status per job (CRIT when the last build failed, WARN when unstable or aborted; a job never built or disabled is not judged); jenkins.node.status per node (CRIT offline, WARN taken offline by an administrator); jenkins.queue (WARN when Jenkins marks builds as stuck, with the reason: “there are no nodes with the label…”); jenkins.health (CRIT when a Metrics plugin check fails, WARN during a prepared shutdown).
Templates: Jenkins injoignable, Tâche Jenkins en échec, Nœud Jenkins hors ligne, Builds bloqués dans la file Jenkins, Attente dans la file Jenkins, Exécuteurs Jenkins saturés, Place libre d'un nœud Jenkins, Santé de Jenkins. Ready-made dashboard: Jenkins.
Tested on Jenkins 2.580.2 LTS (jenkins/jenkins:lts-jdk21 image, Folders and Metrics plugins), with an account holding the permissions above: successful, failed, unstable and nested jobs, a build stuck for lack of a node with its label, a never-connected node (CRIT), Jenkins stopped (CRIT, then back to green).
GitHub
Read-only connector to github.com or GitHub Enterprise Cloud (*.ghe.com), built like the cloud connectors: a single agent polls it, and its metrics carry that agent's name. Every interval (5 minutes by default), for each named repository:
- the default branch (read again every 6 hours);
- the latest 100 GitHub Actions runs: the latest completed run of each workflow on the default branch (pull requests and tags are left out), queued and running runs;
- with a token that allows it: open Dependabot, code scanning and secret scanning alerts (100 at most per kind);
- the API quota (
/rate_limit, which does not count against the quota).
integrations:
github:
repositories: [my-org/site, my-org/api]
token: github_pat_… # optional for a public repository
# endpoint: https://api.my-company.ghe.com # GitHub Enterprise Cloud (data residency)
interval: 5m
security: auto # alerts read when the token allows it; false to skip them
tags: {forge: github}
Without a token, only public repositories can be read and GitHub grants 60 requests per hour to the machine's public address (two per repository and per poll): the interval is then at least 5 minutes, and the agent warns when the named repositories would exceed that quota. With a token (5,000 requests per hour), one minute at least. Recommended token: a fine-grained token limited to the tracked repositories, read-only: Actions, Metadata, and for alerts Dependabot alerts, Code scanning alerts, Secret scanning alerts. An alert kind the token cannot read is skipped and named in the github.can_connect message.
| Metric | Meaning |
|---|---|
github.workflows.failing (repo) | workflows whose latest run on the default branch failed |
github.workflow.run.duration (repo, workflow) | duration of the latest completed run |
github.runs.queued, .in_progress, .queued.oldest_age (repo) | queued runs, running runs, wait of the oldest queued run (seconds) |
github.dependabot.alerts, github.code_scanning.alerts (repo, severity) | open alerts by severity (critical, high, medium, low) |
github.secret_scanning.alerts (repo) | exposed secrets, open alerts |
github.rate_limit.limit, .remaining, .used_percent | API quota |
Checks: github.can_connect; github.workflow.status per workflow (CRIT failed or timed out, WARN cancelled or waiting for an action); github.security per repository (CRIT when a critical alert or an exposed secret is open, WARN for a high-severity alert); github.rate_limit (WARN below 10 % of the quota, CRIT when exhausted). GitHub's own workflows (Dependabot, dependency graph) are named by their path (dependabot/update-graph).
Templates: Connecteur GitHub en échec, Workflow GitHub Actions en échec, Exécutions GitHub Actions en file, Alertes de sécurité GitHub, Quota de l'API GitHub. Ready-made dashboard: GitHub.
Tested on the public github.com API, without a token, with a public repository (successful workflows, Dependabot dynamic workflow). The token, security alerts and GitHub Enterprise Cloud follow the GitHub REST API documentation (version 2022-11-28) and its response formats, reproduced in the agent tests; they have not been read on a real account yet.
Zot
The Zot OCI registry publishes its metrics in Prometheus format through its metrics extension (full image; the “minimal” image does not have it). In Zot's config.json:
"extensions": { "metrics": { "enable": true, "prometheus": { "path": "/metrics" } } }
Then the zot profile of the OpenMetrics reader:
integrations:
openmetrics:
- profile: zot
url: http://127.0.0.1:5000/metrics # the http.port
# headers: {Authorization: "Basic …"} # when Zot requires an account
| Metric | Meaning |
|---|---|
zot.http.requests (code, method) | requests per second |
zot.http.latency (method) | latency (histogram: .p50, .p95, .p99, .avg) |
zot.repo.downloads, zot.repo.uploads (repo) | image downloads and uploads per second |
zot.repo.storage (repo) | space used per repository, in bytes |
zot.storage.lock_latency (locktype) | storage lock wait (histogram) |
zot.gc.duration | storage garbage collection (summary) |
zot.scheduler.queue (priority), zot.scheduler.workers (state), .workers_total, zot.scheduler.generators | background tasks: queue, busy or idle workers |
Templates: Zot injoignable, Tâches en retard dans Zot, Verrous du stockage Zot lents. Ready-made dashboard: Zot.
Tested on Zot v2.1.23 (zot-linux-amd64 image, in a container): an image pushed then pulled, the registry stopped (openmetrics.can_connect CRIT, then back to green).
Odoo
The agent reads Odoo (17 and later) every 30 seconds, without an account:
/web/health; since Odoo 18,?db_server_status=1also checks that the PostgreSQL server answers: its failure returns “fail”;- the version through XML-RPC (
/xmlrpc/2/common,versionmethod), falling back to/web/webclient/version_infowhen XML-RPC is disabled or removed; - the response time of each;
- when the url points to the machine itself: Odoo processes in
/proc(master or threaded server, workers, gevent process for live notifications) and their resident memory. Withsetproctitleinstalled, HTTP and cron workers are told apart; without it (official image), they are “worker”.
integrations:
odoo:
url: http://127.0.0.1:8069 # default; or the address served by the proxy
processes: true # default: yes when the url points to the machine
| Metric | Meaning |
|---|---|
odoo.response_time (endpoint: health, xmlrpc or version_info) | response time, in milliseconds |
odoo.database.reachable | 1 when Odoo's PostgreSQL server answers (Odoo 18 and later) |
odoo.processes, odoo.processes.memory (role: main, worker, http, cron, gevent) | processes and resident memory per role |
odoo.workers | running workers (multi-process mode, workers > 0) |
Checks: odoo.can_connect (with the version) and odoo.health (CRIT when /web/health returns “fail”). PostgreSQL itself is followed by the PostgreSQL integration.
Templates: Odoo injoignable, Odoo en défaut, Odoo lent. Ready-made dashboard: Odoo.
Tested on Odoo 19.0 (official odoo:19 image, multi-process mode with two HTTP workers and one cron, PostgreSQL 17): version read through XML-RPC, five processes seen (master, three workers, gevent), connections to the database forbidden (odoo.health CRIT, then back to green), Odoo stopped (CRIT).
Updated: 21 November 2026