Add zpool monitor

docs/monitors/zpool.rst

@@ -0,0 +1,12 @@
+zpool - zpool status
+^^^^^^^^^^^^^^^^^^^^
+
+Check zpools' statuses to make sure they're ``ONLINE``.
+
+.. confval:: pools
+
+    :type: comma-separated list of string
+    :required: false
+    :default: none (all pools)
+
+    the pools to have their status checked. If empty, all pools are checked.

pyproject.toml

@@ -54,6 +54,7 @@ ring-doorbell = ">=0.6.0"
 paramiko = ">=2.7.2,<4.0.0"
 pyaarlo = ">=0.7.1.3,<0.9.0.0"
 icmplib = "^3.0.3"
+packaging = "^25.0"
 
 [tool.poetry.group.dev.dependencies]
 boto3-stubs = {extras = ["sns"], version = "^1.40.47"}

simplemonitor/Monitors/__init__.py

@@ -19,6 +19,7 @@
     MonitorPortAudit,
     MonitorSwap,
     MonitorZap,
+    MonitorZpool,
 )
 from .network import (
     MonitorDNS,
@@ -75,5 +76,6 @@
     "MonitorUnixService",
     "MonitorWindowsDHCPScope",
     "MonitorZap",
+    "MonitorZpool",
     "RemoteHostsMonitor",
 ]

simplemonitor/Monitors/host.py

@@ -8,9 +8,10 @@
 import shlex
 import subprocess  # nosec
 import time
-from typing import Tuple, cast
+from typing import Optional, Tuple, cast
 
 from markupsafe import escape
+from packaging.version import parse
 
 from ..util import bytes_to_size_string, size_string_to_bytes
 from .monitor import Monitor, register
@@ -599,3 +600,106 @@ def get_params(self) -> Tuple:
             self.result_max,
             self.show_output,
         )
+
+
+@register
+class MonitorZpool(Monitor):
+    """Check zpool status is healthy"""
+
+    monitor_type = "zpool"
+
+    def __init__(self, name: str, config_options: dict) -> None:
+        super().__init__(name, config_options)
+        self.pools = cast(
+            list[str],
+            map(
+                str.strip,
+                self.get_config_option("pools", required_type="str").split(","),
+            ),
+        )
+        self.use_json = False
+        try:
+            zpool_output = subprocess.run(
+                ["zpool", "--version"], capture_output=True
+            ).stdout.decode()  # nosec
+            zpool_version = parse(zpool_output.splitlines()[0].split("-")[1])
+            if zpool_version >= parse("2.4.0"):
+                self.use_json = True
+        except Exception:
+            self.monitor_logger.warning(
+                "Failed to divine zpool version; using text parsing"
+            )
+
+    def describe(self) -> str:
+        if not self.pools:
+            pools = "all zpools"
+        else:
+            pools = "zpools: " + ", ".join(self.pools)
+        return f"Checking zpool status for {pools}"
+
+    def get_params(self) -> Tuple:
+        return (self.pools,)
+
+    def _run_test_json(self) -> Optional[str]:
+        """Return error information, or None if all ok."""
+        messages: list[str] = []
+        cmd = ["zpool", "status", "-j"]
+        if self.pools:
+            cmd.extend(self.pools)
+        try:
+            _output = subprocess.run(cmd, capture_output=True)
+        except subprocess.SubprocessError:
+            return "Failed to run zpool status"
+        try:
+            zpool_info = json.loads(_output.stdout.decode())
+        except Exception:
+            return "Failed to parse zpool JSON output"
+        for pool in zpool_info["pools"].values():
+            if pool["state"] != "ONLINE":
+                messages.append(f"pool {pool['name']} is {pool['state']}")
+        if len(messages) == 0:
+            return None
+        return ", ".join(messages)
+
+    def _run_test_text(self) -> Optional[str]:
+        """Return error information, or None if all ok."""
+        messages: list[str] = []
+        cmd = ["zpool", "status"]
+        if self.pools:
+            cmd.extend(self.pools)
+        try:
+            _output = subprocess.run(cmd, capture_output=True)
+        except subprocess.SubprocessError:
+            return "Failed to run zpool status"
+        zpool_info = _output.stdout.decode()
+        current_pool = None
+        for line in zpool_info.splitlines():
+            matches = re.match(r" *pool: (.+)", line)
+            if matches:
+                if current_pool:
+                    return "Failed to parse zpool status output"
+                current_pool = matches.group(1)
+                continue
+            matches = re.match(r" *state: ([A-Z_]+)", line)
+            if matches:
+                if not current_pool:
+                    return "Failed to parse zpool status output"
+                status = matches.group(1)
+                if status != "ONLINE":
+                    messages.append(f"pool {current_pool} is {matches.group(1)}")
+                current_pool = None
+        if current_pool:
+            messages.append(f"Failed to find status for pool {current_pool}")
+        if len(messages) == 0:
+            return None
+        return ", ".join(messages)
+
+    def run_test(self) -> bool:
+        if self.use_json:
+            message = self._run_test_json()
+        else:
+            message = self._run_test_text()
+        if message:
+            return self.record_fail(message)
+        else:
+            return self.record_success("all pools ONLINE")