zulip_ops: Configure stats to be pushed to status.zulip.com.

puppet/zulip_ops/files/statuspage-pusher

@@ -0,0 +1,80 @@
+#!/usr/bin/env python3
+
+import configparser
+import http.client
+import json
+import logging
+import time
+import urllib.parse
+
+
+def fetch_metric(metric_id: str, query: str) -> float:
+    logging.info("Fetching %s", metric_id)
+    params = urllib.parse.urlencode({"query": query})
+    conn = http.client.HTTPConnection("localhost:9090")
+    conn.request("GET", "/api/v1/query?" + params)
+    response = conn.getresponse()
+    if response.status != 200:
+        raise Exception(
+            f"Failed to request {metric_id} from Prometheus: {response.status} {response.reason}"
+        )
+    raw_bytes = response.read()
+    conn.close()
+    data = json.loads(raw_bytes)
+    if data["status"] != "success":
+        raise Exception("Failed to request {metric_id}: {raw_bytes.decode()}")
+    return float(data["data"]["result"][0]["value"][1])
+
+
+def push_metric(metric_id: str, metric_value: float, page_id: str, oauth_token: str) -> None:
+    conn = http.client.HTTPSConnection("api.statuspage.io")
+    params = json.dumps({"data": {"timestamp": time.time(), "value": metric_value}})
+    headers = {"Content-Type": "application/json", "Authorization": "OAuth " + oauth_token}
+    conn.request(
+        "POST", "/v1/pages/" + page_id + "/metrics/" + metric_id + "/data.json", params, headers
+    )
+    response = conn.getresponse()
+    if not (response.status >= 200 and response.status < 300):
+        raise Exception(
+            f"Failed to push {metric_id} to statuspage.io: {response.status} {response.reason}"
+        )
+    logging.info("Wrote %s: %s", metric_id, response.read().decode())
+    conn.close()
+
+
+def update_metric(metric_id: str, query: str, page_id: str, oauth_token: str) -> None:
+    metric_value = fetch_metric(metric_id, query)
+    push_metric(metric_id, metric_value, page_id, oauth_token)
+
+
+def main() -> None:
+    logging.basicConfig(format="%(asctime)s statuspage-pusher: %(message)s", level=logging.INFO)
+
+    secrets_file = configparser.RawConfigParser()
+    secrets_file.read("/etc/zulip/zulip-secrets.conf")
+
+    oauth_token = secrets_file.get("secrets", "statuspage_token")
+    if oauth_token is None:
+        raise RuntimeError("statuspage_token secret is required")
+
+    config_file = configparser.RawConfigParser()
+    config_file.read("/etc/zulip/zulip.conf")
+    page_id = config_file.get("statuspage", "page_id")
+    if page_id is None:
+        raise RuntimeError("statuspage_page_id secret is required")
+
+    metrics_file = configparser.RawConfigParser()
+    metrics_file.read("/etc/zulip/statuspage.conf")
+    metrics = metrics_file["metrics"]
+
+    while True:
+        for metric_id in metrics:
+            try:
+                update_metric(metric_id, metrics.get(metric_id), page_id, oauth_token)
+            except Exception as e:
+                logging.exception(e)
+        time.sleep(30)
+
+
+if __name__ == "__main__":
+    main()

puppet/zulip_ops/manifests/profile/prometheus_server.pp

@@ -12,6 +12,9 @@ class zulip_ops::profile::prometheus_server {
   # Ditto the Akamai logs
   include zulip_ops::prometheus::akamai
 
+  # Export prometheus stats to status.zulip.com
+  include zulip_ops::statuspage
+
   $version = $zulip::common::versions['prometheus']['version']
   $dir = "/srv/zulip-prometheus-${version}"
   $bin = "${dir}/prometheus"

puppet/zulip_ops/manifests/statuspage.pp

@@ -0,0 +1,29 @@
+# @summary Plumb Prometheus stats into status.zulip.com
+#
+# Requires a /etc/zulip/statuspage.conf which maps statuspage.io
+# metric_ids to Prometheus queries.
+class zulip_ops::statuspage {
+  $bin = '/usr/local/bin/statuspage-pusher'
+
+  file { $bin:
+    ensure => file,
+    owner  => 'root',
+    group  => 'root',
+    mode   => '0755',
+    source => 'puppet:///modules/zulip_ops/statuspage-pusher',
+  }
+
+  $page_id = zulipsecret('secrets','statuspage_page_id','')
+  file { "${zulip::common::supervisor_conf_dir}/statuspage-pusher.conf":
+    ensure  => file,
+    require => [
+      Package[supervisor],
+      File[$bin],
+    ],
+    owner   => 'root',
+    group   => 'root',
+    mode    => '0644',
+    content => template('zulip_ops/supervisor/conf.d/statuspage-pusher.conf.template.erb'),
+    notify  => Service[supervisor],
+  }
+}

puppet/zulip_ops/templates/supervisor/conf.d/statuspage-pusher.conf.template.erb

@@ -0,0 +1,8 @@
+[program:statuspage-pusher]
+command=<%= @bin %>
+priority=10
+autostart=true
+autorestart=true
+user=zulip
+redirect_stderr=true
+stdout_logfile=/var/log/zulip/statuspage-pusher.log