From 3aba2789d314434f5c58ca568be6384ecfb27534 Mon Sep 17 00:00:00 2001
From: Alex Vandiver <alexmv@zulip.com>
Date: Wed, 26 Apr 2023 17:56:11 +0000
Subject: [PATCH] prometheus: Add an exporter for wal-g backup properties.

Since backups may now taken on arbitrary hosts, we need a blackbox
monitor that _some_ backup was produced.

Add a Prometheus exporter which calls `wal-g backup-list` and reports
statistics about the backups.

This could be extended to include `wal-g wal-verify`, but that
requires a connection to the PostgreSQL server.
---
 puppet/zulip/files/postgresql/wal-g-exporter  | 153 ++++++++++++++++++
 .../manifests/profile/prometheus_server.pp    |   3 +
 .../zulip_ops/manifests/prometheus/wal_g.pp   |  34 ++++
 ...rometheus_wal_g_exporter.conf.template.erb |  12 ++
 4 files changed, 202 insertions(+)
 create mode 100755 puppet/zulip/files/postgresql/wal-g-exporter
 create mode 100644 puppet/zulip_ops/manifests/prometheus/wal_g.pp
 create mode 100644 puppet/zulip_ops/templates/supervisor/conf.d/prometheus_wal_g_exporter.conf.template.erb

diff --git a/puppet/zulip/files/postgresql/wal-g-exporter b/puppet/zulip/files/postgresql/wal-g-exporter
new file mode 100755
index 0000000000..d92c3c982b
--- /dev/null
+++ b/puppet/zulip/files/postgresql/wal-g-exporter
@@ -0,0 +1,153 @@
+#!/usr/bin/env python3
+import configparser
+import contextlib
+import json
+import logging
+import subprocess
+import sys
+from collections import defaultdict
+from datetime import datetime, timedelta, timezone
+from http.server import BaseHTTPRequestHandler, HTTPServer
+from typing import Dict, List, Mapping, Optional, Protocol
+
+
+class GaugeMetric(Protocol):
+    def __call__(self, value: float, labels: Optional[Mapping[str, str]] = None, /) -> None:
+        pass
+
+
+class WalGPrometheusServer(BaseHTTPRequestHandler):
+    METRIC_PREFIX = "wal_g_backup_"
+
+    metrics: Dict[str, List[str]] = {}
+    metric_values: Dict[str, Dict[str, str]] = defaultdict(dict)
+
+    server_version = "wal-g-prometheus-server/1.0"
+
+    def gauge(
+        self, name: str, description: Optional[str] = None, default_value: Optional[float] = None
+    ) -> GaugeMetric:
+        if name in self.metrics:
+            raise ValueError(f"Redefinition of {name} metric")
+        self.metrics[name] = [f"# TYPE {self.METRIC_PREFIX}{name} gauge"]
+        if description is not None:
+            self.metrics[name].append(f"# HELP {self.METRIC_PREFIX}{name} {description}")
+
+        def inner(value: float, labels: Optional[Mapping[str, str]] = None) -> None:
+            label_str = ""
+            if labels:
+                label_str = "{" + ",".join([f'{k}="{v}"' for k, v in labels.items()]) + "}"
+            self.metric_values[name][label_str] = f"{self.METRIC_PREFIX}{name}{label_str} {value}"
+
+        if default_value is not None:
+            inner(default_value)
+        return inner
+
+    def print_metrics(self) -> None:
+        lines = []
+        for metric_name in self.metrics:
+            if self.metric_values[metric_name]:
+                # Print preamble
+                lines += self.metrics[metric_name]
+                for metric_value in self.metric_values[metric_name].values():
+                    lines.append(metric_value)
+                lines.append("")
+        self.wfile.write("\n".join(lines).encode())
+
+    def do_GET(self) -> None:
+        if self.path != "/metrics":
+            self.send_response(404)
+            self.end_headers()
+            sys.stderr.flush()
+            return
+
+        self.send_response(200)
+        self.send_header("Content-type", "text/plain; version=0.0.4")
+        self.end_headers()
+
+        self.metrics = {}
+        self.metric_values = defaultdict(dict)
+
+        backup_ok = self.gauge("ok", "If wal-g backup-list was OK", 0)
+        backup_count = self.gauge("count", "Number of backups")
+        backup_earliest_age_seconds = self.gauge("earliest_age_seconds", "Age of the oldest backup")
+        backup_latest_age_seconds = self.gauge(
+            "latest_age_seconds", "Age of the most recent backup"
+        )
+        backup_latest_duration_seconds = self.gauge(
+            "latest_duration_seconds", "Duration the most recent backup took, in seconds"
+        )
+        backup_latest_compressed_size_bytes = self.gauge(
+            "latest_compressed_size_bytes", "Size of the most recent backup, in bytes"
+        )
+        backup_latest_uncompressed_size_bytes = self.gauge(
+            "latest_uncompressed_size_bytes",
+            "Uncompressed size of the most recent backup, in bytes",
+        )
+        backup_total_compressed_size_bytes = self.gauge(
+            "total_compressed_size_bytes", "Total compressed size of all backups, in bytes"
+        )
+
+        now = datetime.now(tz=timezone.utc)
+        try:
+            config_file = configparser.RawConfigParser()
+            config_file.read("/etc/zulip/zulip-secrets.conf")
+            bucket = config_file["secrets"]["s3_backups_bucket"]
+
+            backup_list_output = subprocess.check_output(
+                ["env-wal-g", "backup-list", "--detail", "--json"],
+                text=True,
+            )
+            data = json.loads(backup_list_output)
+            backup_count(len(data), {"bucket": bucket})
+
+            backup_total_compressed_size_bytes(
+                sum(e["compressed_size"] for e in data), {"bucket": bucket}
+            )
+
+            if len(data) > 0:
+                data.sort(key=lambda e: e["time"], reverse=True)
+                latest = data[0]
+                labels = {
+                    "host": latest["hostname"],
+                    "pg_version": str(latest["pg_version"]),
+                    "bucket": bucket,
+                }
+                backup_latest_compressed_size_bytes(latest["compressed_size"], labels)
+                backup_latest_uncompressed_size_bytes(latest["uncompressed_size"], labels)
+
+                def t(key: str, e: Dict[str, str] = latest) -> datetime:
+                    return datetime.strptime(e[key], e["date_fmt"]).replace(tzinfo=timezone.utc)
+
+                backup_earliest_age_seconds(
+                    (now - t("start_time", data[-1])) / timedelta(seconds=1),
+                    {
+                        "host": data[-1]["hostname"],
+                        "pg_version": data[-1]["pg_version"],
+                        "bucket": bucket,
+                    },
+                )
+                backup_latest_age_seconds((now - t("start_time")) / timedelta(seconds=1), labels)
+                backup_latest_duration_seconds(
+                    (t("finish_time") - t("start_time")) / timedelta(seconds=1), labels
+                )
+            backup_ok(1)
+        except Exception as e:
+            logging.exception(e)
+        finally:
+            self.print_metrics()
+            self.log_message(
+                "Served in %.2f seconds",
+                (datetime.now(tz=timezone.utc) - now) / timedelta(seconds=1),
+            )
+            sys.stderr.flush()
+
+
+logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s: %(message)s")
+server = HTTPServer(("127.0.0.1", 9188), WalGPrometheusServer)
+logging.info("Starting server...")
+with contextlib.suppress(KeyboardInterrupt):
+    server.serve_forever()
+
+server.server_close()
+logging.info("Stopping server...")
diff --git a/puppet/zulip_ops/manifests/profile/prometheus_server.pp b/puppet/zulip_ops/manifests/profile/prometheus_server.pp
index 16cf3ba0cc..5997e8d585 100644
--- a/puppet/zulip_ops/manifests/profile/prometheus_server.pp
+++ b/puppet/zulip_ops/manifests/profile/prometheus_server.pp
@@ -6,6 +6,9 @@ class zulip_ops::profile::prometheus_server {
   include zulip_ops::profile::base
   include zulip_ops::prometheus::base
 
+  # This blackbox monitoring of the backup system runs locally
+  include zulip_ops::prometheus::wal_g
+
   $version = $zulip::common::versions['prometheus']['version']
   $dir = "/srv/zulip-prometheus-${version}"
   $bin = "${dir}/prometheus"
diff --git a/puppet/zulip_ops/manifests/prometheus/wal_g.pp b/puppet/zulip_ops/manifests/prometheus/wal_g.pp
new file mode 100644
index 0000000000..0851c8752d
--- /dev/null
+++ b/puppet/zulip_ops/manifests/prometheus/wal_g.pp
@@ -0,0 +1,34 @@
+# @summary Prometheus monitoring of wal-g backups
+#
+class zulip_ops::prometheus::wal_g {
+  include zulip_ops::prometheus::base
+  include zulip::supervisor
+  include zulip::wal_g
+
+  file { '/usr/local/bin/wal-g-exporter':
+    ensure  => file,
+    require => User[zulip],
+    owner   => 'zulip',
+    group   => 'zulip',
+    mode    => '0755',
+    source  => 'puppet:///modules/zulip/postgresql/wal-g-exporter',
+  }
+
+  # We embed the hash of the contents into the name of the process, so
+  # that `supervisorctl reread` knows that it has updated.
+  $full_exporter_hash = sha256(file('zulip/postgresql/wal-g-exporter'))
+  $exporter_hash = $full_exporter_hash[0,8]
+  file { "${zulip::common::supervisor_conf_dir}/prometheus_wal_g_exporter.conf":
+    ensure  => file,
+    require => [
+      User[zulip],
+      Package[supervisor],
+      File['/usr/local/bin/wal-g-exporter'],
+    ],
+    owner   => 'root',
+    group   => 'root',
+    mode    => '0644',
+    content => template('zulip_ops/supervisor/conf.d/prometheus_wal_g_exporter.conf.template.erb'),
+    notify  => Service[supervisor],
+  }
+}
diff --git a/puppet/zulip_ops/templates/supervisor/conf.d/prometheus_wal_g_exporter.conf.template.erb b/puppet/zulip_ops/templates/supervisor/conf.d/prometheus_wal_g_exporter.conf.template.erb
new file mode 100644
index 0000000000..e8cf6c533f
--- /dev/null
+++ b/puppet/zulip_ops/templates/supervisor/conf.d/prometheus_wal_g_exporter.conf.template.erb
@@ -0,0 +1,12 @@
+[program:prometheus_wal_g_exporter]
+# We record the hash of the script so that we can update this file
+# with it, which will make `supervisorctl reread && supervisorctl
+# update` restart this job.
+command=/usr/local/bin/wal-g-exporter
+process_name=wal-g-exporter_<%= @exporter_hash %>
+priority=10
+autostart=true
+autorestart=true
+user=zulip
+redirect_stderr=true
+stdout_logfile=/var/log/zulip/wal_g_exporter.log