config.example.yml

Rama main · réplica de lectura del repositorio original.

# IsItDown — Light edition configuration.
#
# Copy this file to config.yml and edit it. In the container it is mounted at
# /app/config/config.yml, so it can be changed without rebuilding the image.
#
# Secrets are NEVER written here. Use ${ENV_VAR} references; the values come
# from the environment (see .env.example).

pollIntervalMinutes: 3      # how often to poll every provider
requestTimeoutSeconds: 8    # per-request timeout
maxRetries: 3               # attempts per provider per cycle, with backoff
failureThreshold: 5         # consecutive failures before a "monitoring degraded" warning
adaptivePolling: true       # while a provider has an open incident, poll it on the cadence below
adaptiveIntervalMinutes: 1  # that cadence; never slower than the provider's own interval
confirmSamples: 1           # consecutive polls that must agree before a change notifies (1 = off)
# Correlated-outage detection (roadmap 2.7): when this many providers go bad
# inside the window below, one "this looks like a shared failure" alert replaces
# the individual ones. 0 and 1 are off, which is the default — it is the only
# setting here that replaces alerts rather than adding them.
correlationThreshold: 0
correlationWindowMinutes: 10
locale: en                  # language for notification messages: en | it (falls back to en)

services:
  - name: GitHub
    id: github
    adapter: statuspage
    baseUrl: https://www.githubstatus.com
    # Optional: poll this provider on its own cadence instead of the global
    # pollIntervalMinutes above. A page that publishes every few minutes and one
    # that changes twice a year do not deserve the same one. Omit to follow the
    # global interval.
    intervalMinutes: 2
    # Optional: the group this provider belongs to — "my stack". Providers in
    # the same group get a combined status (worst member wins), and one routing
    # rule below can cover all of them with `provider: group:deploy-path`. A
    # lowercase slug; omit it to leave the provider ungrouped.
    group: deploy-path
    # Optional: the monthly uptime this provider is supposed to deliver, as a
    # percentage. The UI edition turns it into an error budget, reports what the
    # month has spent of it, and sends one alert when the rate says the month
    # will miss. Omit it when nobody promised anything — most providers.
    slaTarget: 99.9
    # Optional: what an hour of this provider being down costs you, in your own
    # currency. Shown beside each incident, in the reliability table and in the
    # monthly report. Omit it and no figure is shown — an unpriced outage is not
    # a free one.
    outageCostPerHour: 1200
    # Optional: what breaks here when this provider does, and who answers for
    # it. Both are added under every message about it, so an alert at 3 a.m.
    # reads as a handover note.
    impact: deploys blocked, CI stopped
    owner: "@alice"
    # Optional: what to do when it is down, in Markdown. Shown on the provider's
    # page; every alert ends with its first link, or its first line.
    runbook: |
      Switch CI to the GitLab mirror: https://wiki.example.com/runbooks/github
    # Optional: free labels beside the group. A routing rule can name one as
    # `provider: tag:critical`.
    tags: [paid, critical]
    # Optional: planned mutes — a weekly window (may wrap midnight) or one span.
    # While one runs the provider is muted exactly like the dashboard's Mute.
    muteSchedules:
      - { days: [sun], start: "03:00", end: "04:00", timeZone: Europe/Rome }
    # Optional: maintenance you declare because the provider does not. It
    # silences, is announced and is drawn on the timeline like the provider's
    # own; it always needs an end.
    # maintenanceWindows:
    #   - { title: Our deploy, startsAt: "2026-12-01T22:00:00Z", endsAt: "2026-12-01T23:00:00Z" }

  - name: Cloudflare
    id: cloudflare
    adapter: statuspage
    baseUrl: https://www.cloudflarestatus.com
    group: deploy-path
    # Optional: watch only part of a provider. Component ids come from
    # <baseUrl>/api/v2/summary.json, where Cloudflare groups its data centres by
    # region. With scopeToComponents an incident attributed only to components
    # outside this list is ignored, and the reported status is the worst of these
    # components rather than the page-wide indicator. A notice the provider
    # attaches to no component at all is page-wide and always reported.
    scopeToComponents: true
    components:
      - id: 57ctn3f2qsyj
        name: Amsterdam, Netherlands - (AMS)
      - id: k8q1kmz4gfms
        name: Milan, Italy - (MXP)

  - name: Anthropic
    id: anthropic
    adapter: statuspage
    # status.anthropic.com 301-redirects here; the canonical host avoids the hop
    baseUrl: https://status.claude.com

  # A page that publishes neither JSON nor a feed: read one element's text and
  # map the words onto a severity. Fragile by nature — a selector that stops
  # matching is reported as a failing provider, never as "operational" — so use
  # it only where there is nothing better.
  #
  # - name: Example
  #   id: example
  #   adapter: html
  #   baseUrl: https://status.example.com/
  #   options:
  #     selector: ".status-banner"
  #     # Optional. Each severity keeps its built-in words unless you give it
  #     # your own; on the page, the longest match wins.
  #     operational: "all systems operational"
  #     major_outage: "outage, down"

  # A probe of the service itself rather than of a page about it. Here it is
  # also the second opinion on GitHub's own status page: when this probe cannot
  # reach the API and githubstatus.com still reports operational with no
  # incident open, the disagreement is itself an alert — the silent-outage
  # cross-check.
  #
  # - name: GitHub API
  #   id: github-api
  #   adapter: http
  #   baseUrl: https://api.github.com
  #   crossChecks: github
  #   options:
  #     path: /meta
  #     expectStatus: "200-299"
  #
  # Which source is the record for a provider — `declared` (its own status page)
  # or `observed` (our own reading), roadmap 9.1. Left out, which is the normal
  # case, the adapter decides: a probe is `observed` because it *is* the
  # reading, and anything that parses somebody's page is `declared`. Set it only
  # to disagree with that default.
  #
  # - name: Acme
  #   id: acme
  #   adapter: statuspage
  #   baseUrl: https://status.acme.example
  #   authority: observed

notifications:
  telegram:
    enabled: true
    botToken: "${TELEGRAM_BOT_TOKEN}"
    chatId: "${TELEGRAM_CHAT_ID}"
    # Optional, per channel (roadmap 3.20 and 3.15). `locale` overrides the
    # notification language above for this channel alone — so this chat can read
    # Italian while the webhook payload below stays English. `template` replaces
    # the default message: substitution only, no conditions and no loops, and an
    # unknown token stops the container with the token named. See
    # docs/configuration.md#311-per-channel-locale-and-template for the tokens.
    # locale: it
    # template: "[{{severity}}] {{provider}} — {{title}}\n{{url}}"
  webhook:
    enabled: false
    url: "${WEBHOOK_URL}"
    # Optional. With a secret set, every request carries X-IsItDown-Timestamp
    # and X-IsItDown-Signature (HMAC-SHA256 over "<timestamp>.<body>"), so a
    # receiver can verify the payload came from here. Unset means unsigned.
    secret: "${WEBHOOK_SECRET}"
  discord:
    enabled: false
    # Server settings → Integrations → Webhooks → Copy Webhook URL
    webhookUrl: "${DISCORD_WEBHOOK_URL}"
  slack:
    enabled: false
    # An incoming webhook of a Slack app installed in the target channel
    webhookUrl: "${SLACK_WEBHOOK_URL}"
  ntfy:
    enabled: false
    # The topic URL, server included — ntfy.sh or an instance of your own
    topicUrl: "${NTFY_TOPIC_URL}"
    # Optional: only a server with access control needs a token
    token: "${NTFY_TOKEN}"
  gotify:
    enabled: false
    # The server itself, not a message path
    serverUrl: "${GOTIFY_URL}"
    # An application token, from the Gotify app you created for IsItDown
    token: "${GOTIFY_TOKEN}"
  pushover:
    enabled: false
    # The API token of the Pushover application you created for IsItDown
    token: "${PUSHOVER_TOKEN}"
    # Your user key, or a group key, from the Pushover dashboard
    userKey: "${PUSHOVER_USER_KEY}"
    # Optional: one registered device; empty delivers to every device
    device: "${PUSHOVER_DEVICE}"
  matrix:
    enabled: false
    # The homeserver's root, not a room path
    homeserverUrl: "${MATRIX_HOMESERVER_URL}"
    # The internal room id, not an alias: !abc:example.org
    roomId: "${MATRIX_ROOM_ID}"
    # An access token for the user that posts
    accessToken: "${MATRIX_ACCESS_TOKEN}"
  pagerduty:
    enabled: false
    # The integration key of an Events API v2 integration on a service. An
    # incident that resolves closes the alert it opened, keyed on the change.
    routingKey: "${PAGERDUTY_ROUTING_KEY}"
    # Optional: "eu" for an account in PagerDuty's EU service region
    region: "${PAGERDUTY_REGION}"
  opsgenie:
    enabled: false
    # An API key from an Opsgenie API integration
    apiKey: "${OPSGENIE_API_KEY}"
    # Optional: "eu" for an account on Opsgenie's EU instance
    region: "${OPSGENIE_REGION}"
  apprise:
    enabled: false
    # An Apprise API server, which reaches the ~80 services Apprise knows
    serverUrl: "${APPRISE_SERVER_URL}"
    # Either the key of a configuration the server stores (preferred — the
    # service URLs behind it are credentials) ...
    configKey: "${APPRISE_CONFIG_KEY}"
    # ... or the Apprise URLs themselves, comma-separated, for a stateless
    # server. One of the two is required.
    urls: "${APPRISE_URLS}"
  teams:
    enabled: false
    # A Teams channel webhook: Workflows → "Post to a channel when a webhook
    # request is received", or an Office 365 connector on an older tenant
    webhookUrl: "${TEAMS_WEBHOOK_URL}"
  email:
    enabled: false
    # An SMTP submission server. STARTTLS is used whenever the server offers it.
    host: "${SMTP_HOST}"
    # Optional, 587 by default. 465 is implicit TLS and is treated as such.
    port: "${SMTP_PORT}"
    # Optional: "true" forces TLS from the first byte, as on port 465
    secure: "${SMTP_SECURE}"
    # Optional. Credentials are never sent over an unencrypted connection
    # unless this says "true" — meant for a relay on this machine.
    allowInsecureAuth: "${SMTP_ALLOW_INSECURE_AUTH}"
    # Optional: "true" accepts a certificate no public CA signed
    allowSelfSigned: "${SMTP_ALLOW_SELF_SIGNED}"
    # Optional: a relay that trusts this network needs no credentials
    username: "${SMTP_USERNAME}"
    password: "${SMTP_PASSWORD}"
    from: "${SMTP_FROM}"
    # One address, or several separated by commas
    to: "${SMTP_TO}"
  # On-call chains (roadmap 3.9). Like PagerDuty and Opsgenie, each keys its
  # alert on the change, so a provider's own "resolved" closes what it opened.
  splunk-oncall:
    enabled: false
    # The REST endpoint integration's API key, and the routing key that picks
    # the escalation policy
    apiKey: "${SPLUNK_ONCALL_API_KEY}"
    routingKey: "${SPLUNK_ONCALL_ROUTING_KEY}"
  squadcast:
    enabled: false
    # An Incident Webhook integration's URL
    webhookUrl: "${SQUADCAST_WEBHOOK_URL}"
  zenduty:
    enabled: false
    # An API integration's key
    integrationKey: "${ZENDUTY_INTEGRATION_KEY}"
  incident-io:
    enabled: false
    # An HTTP alert source: its config id and its bearer token
    alertSourceId: "${INCIDENT_IO_ALERT_SOURCE_ID}"
    token: "${INCIDENT_IO_TOKEN}"
  grafana-oncall:
    enabled: false
    # A Formatted Webhook integration's URL
    webhookUrl: "${GRAFANA_ONCALL_WEBHOOK_URL}"
  # Team chat through each server's API, so an incident's updates edit the
  # message they belong to instead of adding one each.
  mattermost:
    enabled: false
    serverUrl: "${MATTERMOST_URL}"
    # A bot account's access token, and the channel's id (not its name)
    token: "${MATTERMOST_TOKEN}"
    channelId: "${MATTERMOST_CHANNEL_ID}"
  rocketchat:
    enabled: false
    serverUrl: "${ROCKETCHAT_URL}"
    # The bot user's id and a personal access token of it, and the room's id
    userId: "${ROCKETCHAT_USER_ID}"
    token: "${ROCKETCHAT_TOKEN}"
    roomId: "${ROCKETCHAT_ROOM_ID}"
  zulip:
    enabled: false
    siteUrl: "${ZULIP_SITE_URL}"
    botEmail: "${ZULIP_BOT_EMAIL}"
    apiKey: "${ZULIP_API_KEY}"
    # The channel ("stream") to post in
    stream: "${ZULIP_STREAM}"
    # Optional: one fixed topic. Unset, each provider gets a topic of its own.
    topic: "${ZULIP_TOPIC}"
  google-chat:
    enabled: false
    # A space's incoming webhook. An incident's messages share one thread.
    webhookUrl: "${GOOGLE_CHAT_WEBHOOK_URL}"
  # Phones. Numbers are E.164 (+39...). Route only what is worth waking up
  # for to the call — a rule with minSeverity: major_outage.
  twilio-sms:
    enabled: false
    accountSid: "${TWILIO_SMS_ACCOUNT_SID}"
    authToken: "${TWILIO_SMS_AUTH_TOKEN}"
    from: "${TWILIO_SMS_FROM}"
    to: "${TWILIO_SMS_TO}"
  twilio-call:
    enabled: false
    accountSid: "${TWILIO_CALL_ACCOUNT_SID}"
    authToken: "${TWILIO_CALL_AUTH_TOKEN}"
    from: "${TWILIO_CALL_FROM}"
    to: "${TWILIO_CALL_TO}"
  bark:
    enabled: false
    # The device key the Bark iOS app shows
    deviceKey: "${BARK_DEVICE_KEY}"
    # Optional: a self-hosted bark-server. Unset means https://api.day.app
    serverUrl: "${BARK_SERVER_URL}"
  signal:
    enabled: false
    # A signal-cli-rest-api server, which holds the account's keys
    serverUrl: "${SIGNAL_API_URL}"
    # The registered number to send from, and who receives: numbers or
    # group.… ids, comma-separated
    number: "${SIGNAL_NUMBER}"
    recipients: "${SIGNAL_RECIPIENTS}"
  # Machines.
  syslog:
    enabled: false
    # udp://host:514, tcp://host:601 or tls://host:6514 — RFC 5424 messages
    url: "${SYSLOG_URL}"
    # Optional: local0 (the default) to local7
    facility: "${SYSLOG_FACILITY}"
  mqtt:
    enabled: false
    # mqtt://broker:1883, or mqtts:// for TLS
    brokerUrl: "${MQTT_BROKER_URL}"
    # Optional: the topic prefix, "isitdown" by default. Messages go to
    # <prefix>/<provider>; a provider's state is published retained.
    topic: "${MQTT_TOPIC}"
    # Optional: only for a broker that wants credentials
    username: "${MQTT_USERNAME}"
    password: "${MQTT_PASSWORD}"
  # Public posts. The operator's impact and owner notes are never posted.
  mastodon:
    enabled: false
    instanceUrl: "${MASTODON_INSTANCE_URL}"
    # An application's access token with write:statuses
    accessToken: "${MASTODON_ACCESS_TOKEN}"
    # Optional: public (the default), unlisted or private
    visibility: "${MASTODON_VISIBILITY}"
  bluesky:
    enabled: false
    handle: "${BLUESKY_HANDLE}"
    # An app password (Settings → App passwords), never the account's own
    appPassword: "${BLUESKY_APP_PASSWORD}"
    # Optional: the account's PDS, when it is not bsky.social
    serviceUrl: "${BLUESKY_SERVICE_URL}"

# Which channels hear about what. The first rule that matches a change decides,
# rules below it are not consulted — put specific ones on top. Leave this
# section out entirely and every change goes to every enabled channel, which is
# what happens without it.
#
# classes: status | incident | maintenance | monitoring (default: all)
# minSeverity: any | degraded | partial_outage | major_outage (default: any)
# channels: channel names from `notifications` above, ["*"] for all of
#   them, or [] to stay silent
#
# A rule's severity is the worse of the status a change came from and the one
# it went to, so a recovery from a major outage reaches the same channels the
# outage did.
#
# routing:
#   # A provider watched out of curiosity: on the dashboard, never in a message.
#   - provider: anthropic
#     channels: []
#   # The whole deploy path, in one rule: it stays right when the stack gains a
#   # fifth provider, which four provider ids never would.
#   - provider: group:deploy-path
#     classes: [status, incident]
#     minSeverity: degraded
#     channels: [telegram]
#   # Real trouble, anywhere: every channel gets attention.
#   - provider: "*"
#     classes: [status, incident]
#     minSeverity: major_outage
#     channels: [telegram]
#   # Everything else, including scheduled maintenance and IsItDown's own
#   # polling trouble: somewhere quieter.
#   - provider: "*"
#     channels: [slack]

# How much of what the rules above admit actually goes out, and in how many
# messages. Leave this section out and all four controls are off, which is what
# every installation before them was doing.
#
# delivery:
#   # Quiet hours: inside the window, only changes at or above the floor are
#   # sent. Anything under it is dropped, not deferred — the digest below is the
#   # control that defers. The window may wrap midnight, and is read in
#   # `timeZone` ("auto" means the zone the container runs in).
#   quietHours:
#     enabled: true
#     start: "23:00"
#     end: "07:00"
#     timeZone: "Europe/Rome"
#     minSeverity: major_outage
#   # Digest mode: collect changes under `immediateFloor` and send them as one
#   # message per window. Anything at or above the floor still goes out the
#   # moment it happens, so the window's length only ever delays the news you
#   # said was not urgent. A batch collected when the container stops is lost
#   # rather than sent an hour late.
#   digest:
#     enabled: true
#     windowMinutes: 15
#     immediateFloor: major_outage
#   # A ceiling per provider, per rolling hour. What the cap holds back is
#   # counted, and the next message that gets through says how much it hid.
#   cap:
#     enabled: true
#     maxPerHour: 6
#   # One incident becomes one message that is rewritten as the incident moves,
#   # on channels that can edit one (telegram, discord). The others still get a
#   # message per update.
#   updateInPlace: true
#
#   # A heads-up before a maintenance window the provider has declared opens
#   # (not while it runs): "GitHub maintenance in 2 h, expected 30 min".
#   maintenanceNotice:
#     enabled: true
#     leadMinutes: 120
#
#   # A major change nobody took in charge is said again on another channel
#   # after this long. UI edition only: it needs the "I'm on it" button to know
#   # whether anybody answered.
#   escalation:
#     enabled: true
#     afterMinutes: 15
#     channel: pushover
#     minSeverity: major_outage
#
#   # When a channel refuses this many messages in a row, say so on the
#   # others instead of leaving it to a badge in the delivery log.
#   channelHealth:
#     enabled: true
#     afterFailures: 3
#
#   # When several providers fail together and then recover, send one message
#   # ("6 providers, 47 minutes, probable cause: cloudflare") instead of one
#   # "resolved" per provider. The members' own recoveries are held back until
#   # the last one is back.
#   incidentSummary:
#     enabled: true
#
#   # A provider that posts several updates of the same incident in a short
#   # time produces one message: the last state, sent once nothing newer has
#   # arrived for this long.
#   coalesce:
#     enabled: true
#     windowMinutes: 2
#
#   # While a provider is already in major outage, further updates of its
#   # incident go to the digest window instead of interrupting. A change of
#   # severity or the resolution still goes out at once.
#   knownOutage:
#     enabled: true
#
#   # UI edition only. Attach the provider's last 24 hours, as an SVG, to status
#   # and incident messages. Only the email and webhook channels can carry it.
#   attachChart: true
#
#   # UI edition only. The fleet's week or month on one channel: fleet uptime,
#   # the weakest providers, incidents, error budget spent. It is not an alert,
#   # so routing rules do not apply, but it waits out quiet hours.
#   report:
#     enabled: true
#     frequency: weekly   # weekly | monthly
#     channel: slack
#
#   # Who is on call this week: a rotation of the recipients defined below,
#   # one per week, Monday to Sunday UTC. Whoever's week it is also hears every
#   # change at or above minSeverity, on top of what the routing rules chose.
#   onCall:
#     enabled: true
#     minSeverity: major_outage
#     startsOn: 2026-10-05   # any day of the first week
#     rotation: [alice, bob]
#
#   # UI edition only. Once a week every enabled channel gets a message that
#   # says it is a routine test; a failure raises the failing-channel alert.
#   deliveryTest:
#     enabled: true

# People reached through more than one channel, preferred first. A change a
# rule sends to several of one person's channels goes to the first only, and to
# the next only if that one fails — so nobody hears it twice, and nobody has to
# mute their second channel to stop it.
#
# recipients:
#   - id: alice
#     name: Alice
#     channels: [telegram, discord]