add docker storage, docker driver checks

4 files changed, 484 insertions, 0 deletions

diff --git a/roles/openshift_health_checker/openshift_checks/docker_storage.py b/roles/openshift_health_checker/openshift_checks/docker_storage.py
new file mode 100644
index 000000000..2dfe10a02
--- /dev/null
+++ b/roles/openshift_health_checker/openshift_checks/docker_storage.py
@@ -0,0 +1,110 @@
+# pylint: disable=missing-docstring
+import json
+
+from openshift_checks import OpenShiftCheck, OpenShiftCheckException, get_var
+
+
+class DockerStorage(OpenShiftCheck):
+    """Check Docker storage sanity.
+
+    Check for thinpool usage during a containerized installation
+    """
+
+    name = "docker_storage"
+    tags = ["preflight"]
+
+    max_thinpool_data_usage_percent = 90.0
+    max_thinpool_meta_usage_percent = 90.0
+
+    @classmethod
+    def is_active(cls, task_vars):
+        """Only run on hosts that depend on Docker."""
+        is_containerized = get_var(task_vars, "openshift", "common", "is_containerized")
+        is_node = "nodes" in get_var(task_vars, "group_names", default=[])
+        return (super(DockerStorage, cls).is_active(task_vars) and is_containerized) or is_node
+
+    def run(self, tmp, task_vars):
+        try:
+            self.max_thinpool_data_usage_percent = float(get_var(task_vars, "max_thinpool_data_usage_percent",
+                                                                 default=self.max_thinpool_data_usage_percent))
+            self.max_thinpool_meta_usage_percent = float(get_var(task_vars, "max_thinpool_metadata_usage_percent",
+                                                                 default=self.max_thinpool_meta_usage_percent))
+        except ValueError as err:
+            return {
+                "failed": True,
+                "msg": "Unable to convert thinpool data usage limit to float: {}".format(str(err))
+            }
+
+        err_msg = self.check_thinpool_usage(task_vars)
+        if err_msg:
+            return {"failed": True, "msg": err_msg}
+
+        return {}
+
+    def check_thinpool_usage(self, task_vars):
+        lvs = self.get_lvs_data(task_vars)
+        lv_data = self.extract_thinpool_obj(lvs)
+
+        data_percent = self.get_thinpool_data_usage(lv_data)
+        metadata_percent = self.get_thinpool_metadata_usage(lv_data)
+
+        if data_percent > self.max_thinpool_data_usage_percent:
+            msg = "thinpool data usage above maximum threshold of {threshold}%"
+            return msg.format(threshold=self.max_thinpool_data_usage_percent)
+
+        if metadata_percent > self.max_thinpool_meta_usage_percent:
+            msg = "thinpool metadata usage above maximum threshold of {threshold}%"
+            return msg.format(threshold=self.max_thinpool_meta_usage_percent)
+
+        return ""
+
+    def get_lvs_data(self, task_vars):
+        lvs_cmd = "/sbin/lvs --select vg_name=docker --select lv_name=docker-pool --report-format json"
+        result = self.exec_cmd(lvs_cmd, task_vars)
+
+        if result.get("failed", False):
+            msg = "no thinpool usage data returned by the host: {}"
+            raise OpenShiftCheckException(msg.format(result.get("msg", "")))
+
+        try:
+            data_json = json.loads(result.get("stdout", ""))
+        except ValueError as err:
+            raise OpenShiftCheckException("Invalid JSON value returned by lvs command: {}".format(str(err)))
+
+        data = data_json.get("report")
+        if not data:
+            raise OpenShiftCheckException("no thinpool usage data returned by the host.")
+
+        return data
+
+    @staticmethod
+    def get_thinpool_data_usage(thinpool_lv_data):
+        data = thinpool_lv_data.get("data_percent")
+        if not data:
+            raise OpenShiftCheckException("no thinpool usage data returned by the host.")
+
+        return float(data)
+
+    @staticmethod
+    def get_thinpool_metadata_usage(thinpool_lv_data):
+        data = thinpool_lv_data.get("metadata_percent")
+        if not data:
+            raise OpenShiftCheckException("no thinpool usage data returned by the host.")
+
+        return float(data)
+
+    @staticmethod
+    def extract_thinpool_obj(thinpool_data):
+        if not thinpool_data or not thinpool_data[0]:
+            raise OpenShiftCheckException("no thinpool usage data returned by the host.")
+
+        lv_data = thinpool_data[0].get("lv")
+        if not lv_data or not lv_data[0]:
+            raise OpenShiftCheckException("no thinpool usage data returned by the host.")
+
+        return lv_data[0]
+
+    def exec_cmd(self, cmd_str, task_vars):
+        return self.execute_module("command", {
+            "_raw_params": cmd_str,
+        }, task_vars)
diff --git a/roles/openshift_health_checker/openshift_checks/docker_storage_driver.py b/roles/openshift_health_checker/openshift_checks/docker_storage_driver.py
new file mode 100644
index 000000000..94ea7ba9c
--- /dev/null
+++ b/roles/openshift_health_checker/openshift_checks/docker_storage_driver.py
@@ -0,0 +1,50 @@
+# pylint: disable=missing-docstring
+from openshift_checks import OpenShiftCheck, get_var
+
+
+class DockerStorageDriver(OpenShiftCheck):
+    """Check Docker storage driver compatibility.
+
+    This check ensures that Docker is using a supported storage driver,
+    and that Loopback is not being used (if using devicemapper).
+    """
+
+    name = "docker_storage_driver"
+    tags = ["preflight"]
+
+    storage_drivers = ["devicemapper", "overlay2"]
+
+    @classmethod
+    def is_active(cls, task_vars):
+        """Skip non-containerized installations."""
+        is_containerized = get_var(task_vars, "openshift", "common", "is_containerized")
+        return super(DockerStorageDriver, cls).is_active(task_vars) and is_containerized
+
+    def run(self, tmp, task_vars):
+        info = self.execute_module("docker_info", {}, task_vars).get("info", {})
+
+        if not self.is_supported_storage_driver(info):
+            msg = "Unsupported Docker storage driver detected. Supported storage drivers: {drivers}"
+            return {"failed": True, "msg": msg.format(drivers=', '.join(self.storage_drivers))}
+
+        if self.is_using_loopback_device(info):
+            msg = "Use of loopback devices is discouraged. Try running Docker with `--storage-opt dm.thinpooldev`"
+            return {"failed": True, "msg": msg}
+
+        return {}
+
+    def is_supported_storage_driver(self, docker_info):
+        return docker_info.get("Driver", "") in self.storage_drivers
+
+    @staticmethod
+    def is_using_loopback_device(docker_info):
+        # Loopback device usage is only an issue if using devicemapper.
+        # Skip this check if using any other storage driver.
+        if docker_info.get("Driver", "") != "devicemapper":
+            return False
+
+        for status in docker_info.get("DriverStatus", []):
+            if status[0] == "Data loop file":
+                return bool(status[1])
+
+        return False
diff --git a/roles/openshift_health_checker/test/docker_storage_driver_test.py b/roles/openshift_health_checker/test/docker_storage_driver_test.py
new file mode 100644
index 000000000..34a8f827a
--- /dev/null
+++ b/roles/openshift_health_checker/test/docker_storage_driver_test.py
@@ -0,0 +1,81 @@
+import pytest
+
+
+from openshift_checks.docker_storage_driver import DockerStorageDriver
+
+
+@pytest.mark.parametrize('is_containerized,is_active', [
+    (False, False),
+    (True, True),
+])
+def test_is_active(is_containerized, is_active):
+    task_vars = dict(
+        openshift=dict(common=dict(is_containerized=is_containerized)),
+    )
+    assert DockerStorageDriver.is_active(task_vars=task_vars) == is_active
+
+
+@pytest.mark.parametrize('info,failed,extra_words', [
+    (
+        {
+            "Driver": "devicemapper",
+            "DriverStatus": [("Pool Name", "docker-docker--pool")],
+        },
+        False,
+        [],
+    ),
+    (
+        {
+            "Driver": "devicemapper",
+            "DriverStatus": [("Data loop file", "true")],
+        },
+        True,
+        ["Use of loopback devices is discouraged"],
+    ),
+    (
+        {
+            "Driver": "overlay2",
+            "DriverStatus": []
+        },
+        False,
+        [],
+    ),
+    (
+        {
+            "Driver": "overlay",
+        },
+        True,
+        ["Unsupported Docker storage driver"],
+    ),
+    (
+        {
+            "Driver": "unsupported",
+        },
+        True,
+        ["Unsupported Docker storage driver"],
+    ),
+])
+def test_check_storage_driver(info, failed, extra_words):
+    def execute_module(module_name, args, tmp=None, task_vars=None):
+        if module_name != "docker_info":
+            return {
+                "changed": False,
+            }
+
+        return {
+            "info": info
+        }
+
+    task_vars = dict(
+        openshift=dict(common=dict(is_containerized=True))
+    )
+
+    check = DockerStorageDriver(execute_module=execute_module).run(tmp=None, task_vars=task_vars)
+
+    if failed:
+        assert check["failed"]
+    else:
+        assert not check.get("failed", False)
+
+    for word in extra_words:
+        assert word in check["msg"]
diff --git a/roles/openshift_health_checker/test/docker_storage_test.py b/roles/openshift_health_checker/test/docker_storage_test.py
new file mode 100644
index 000000000..73c433383
--- /dev/null
+++ b/roles/openshift_health_checker/test/docker_storage_test.py
@@ -0,0 +1,243 @@
+import pytest
+import json
+
+
+from openshift_checks.docker_storage import DockerStorage, OpenShiftCheckException
+
+
+@pytest.mark.parametrize('is_containerized,is_active', [
+    (False, False),
+    (True, True),
+])
+def test_is_active(is_containerized, is_active):
+    task_vars = dict(
+        openshift=dict(common=dict(is_containerized=is_containerized)),
+    )
+    assert DockerStorage.is_active(task_vars=task_vars) == is_active
+
+
+@pytest.mark.parametrize('stdout,message,failed,extra_words', [
+    (None, "", True, ["no thinpool usage data"]),
+    ("", "", False, ["Invalid JSON value returned by lvs command"]),
+    (None, "invalid response", True, ["invalid response"]),
+    ("invalid", "invalid response", False, ["Invalid JSON value"]),
+])
+def test_get_lvs_data_with_failed_response(stdout, message, failed, extra_words):
+    def execute_module(module_name, args, tmp=None, task_vars=None):
+        if module_name != "command":
+            return {
+                "changed": False,
+            }
+
+        response = {
+            "stdout": stdout,
+            "msg": message,
+            "failed": failed,
+        }
+
+        if stdout is None:
+            response.pop("stdout")
+
+        return response
+
+    task_vars = dict(
+        max_thinpool_data_usage_percent=90.0
+    )
+
+    check = DockerStorage(execute_module=execute_module)
+    with pytest.raises(OpenShiftCheckException) as excinfo:
+        check.run(tmp=None, task_vars=task_vars)
+
+    for word in extra_words:
+        assert word in str(excinfo.value)
+
+
+@pytest.mark.parametrize('limit_percent,failed,extra_words', [
+    ("90.0", False, []),
+    (80.0, False, []),
+    ("invalid percent", True, ["Unable to convert", "to float", "invalid percent"]),
+    ("90%", True, ["Unable to convert", "to float", "90%"]),
+])
+def test_invalid_value_for_thinpool_usage_limit(limit_percent, failed, extra_words):
+    def execute_module(module_name, args, tmp=None, task_vars=None):
+        if module_name != "command":
+            return {
+                "changed": False,
+            }
+
+        return {
+            "stdout": json.dumps({
+                "report": [
+                    {
+                        "lv": [
+                            {"lv_name": "docker-pool", "vg_name": "docker", "lv_attr": "twi-aot---", "lv_size": "6.95g",
+                             "pool_lv": "", "origin": "", "data_percent": "58.96", "metadata_percent": "4.77",
+                             "move_pv": "", "mirror_log": "", "copy_percent": "", "convert_lv": ""},
+                        ]
+                    }
+                ]
+            }),
+            "failed": False,
+        }
+
+    task_vars = dict(
+        max_thinpool_data_usage_percent=limit_percent
+    )
+
+    check = DockerStorage(execute_module=execute_module).run(tmp=None, task_vars=task_vars)
+
+    if failed:
+        assert check["failed"]
+
+        for word in extra_words:
+            assert word in check["msg"]
+    else:
+        assert not check.get("failed", False)
+
+
+def test_get_lvs_data_with_valid_response():
+    def execute_module(module_name, args, tmp=None, task_vars=None):
+        if module_name != "command":
+            return {
+                "changed": False,
+            }
+
+        return {
+            "stdout": json.dumps({
+                "report": [
+                    {
+                        "lv": [
+                            {"lv_name": "docker-pool", "vg_name": "docker", "lv_attr": "twi-aot---", "lv_size": "6.95g",
+                             "pool_lv": "", "origin": "", "data_percent": "58.96", "metadata_percent": "4.77",
+                             "move_pv": "", "mirror_log": "", "copy_percent": "", "convert_lv": ""}
+                        ]
+                    }
+                ]
+            })
+        }
+
+    task_vars = dict(
+        max_thinpool_data_usage_percent="90"
+    )
+
+    check = DockerStorage(execute_module=execute_module).run(tmp=None, task_vars=task_vars)
+    assert not check.get("failed", False)
+
+
+@pytest.mark.parametrize('response,extra_words', [
+    (
+        {
+            "report": [{}],
+        },
+        ["no thinpool usage data"],
+    ),
+    (
+        {
+            "report": [
+                {
+                    "lv": [
+                        {"vg_name": "docker", "lv_attr": "twi-aot---", "lv_size": "6.95g",
+                         "move_pv": "", "mirror_log": "", "copy_percent": "", "convert_lv": ""}
+                    ]
+                }
+            ],
+        },
+        ["no thinpool usage data"],
+    ),
+    (
+        {
+            "report": [
+                {
+                    "lv": [],
+                }
+            ],
+        },
+        ["no thinpool usage data"],
+    ),
+    (
+        {
+            "report": [
+                {
+                    "lv": [
+                        {"lv_name": "docker-pool", "vg_name": "docker", "lv_attr": "twi-aot---", "lv_size": "6.95g",
+                         "pool_lv": "", "origin": "", "data_percent": "58.96",
+                         "move_pv": "", "mirror_log": "", "copy_percent": "", "convert_lv": ""}
+                    ]
+                }
+            ],
+        },
+        ["no thinpool usage data"],
+    ),
+])
+def test_get_lvs_data_with_incomplete_response(response, extra_words):
+    def execute_module(module_name, args, tmp=None, task_vars=None):
+        if module_name != "command":
+            return {
+                "changed": False,
+            }
+
+        return {
+            "stdout": json.dumps(response)
+        }
+
+    task_vars = dict(
+        max_thinpool_data_usage_percent=90.0
+    )
+
+    check = DockerStorage(execute_module=execute_module)
+    with pytest.raises(OpenShiftCheckException) as excinfo:
+        check.run(tmp=None, task_vars=task_vars)
+
+    assert "no thinpool usage data" in str(excinfo.value)
+
+
+@pytest.mark.parametrize('response,extra_words', [
+    (
+        {
+            "report": [
+                {
+                    "lv": [
+                        {"lv_name": "docker-pool", "vg_name": "docker", "lv_attr": "twi-aot---", "lv_size": "6.95g",
+                         "pool_lv": "", "origin": "", "data_percent": "100.0", "metadata_percent": "90.0",
+                         "move_pv": "", "mirror_log": "", "copy_percent": "", "convert_lv": ""}
+                    ]
+                }
+            ],
+        },
+        ["thinpool data usage above maximum threshold"],
+    ),
+    (
+        {
+            "report": [
+                {
+                    "lv": [
+                        {"lv_name": "docker-pool", "vg_name": "docker", "lv_attr": "twi-aot---", "lv_size": "6.95g",
+                         "pool_lv": "", "origin": "", "data_percent": "10.0", "metadata_percent": "91.0",
+                         "move_pv": "", "mirror_log": "", "copy_percent": "", "convert_lv": ""}
+                    ]
+                }
+            ],
+        },
+        ["thinpool metadata usage above maximum threshold"],
+    ),
+])
+def test_get_lvs_data_with_high_thinpool_usage(response, extra_words):
+    def execute_module(module_name, args, tmp=None, task_vars=None):
+        if module_name != "command":
+            return {
+                "changed": False,
+            }
+
+        return {
+            "stdout": json.dumps(response),
+        }
+
+    task_vars = dict(
+        max_thinpool_data_usage_percent="90"
+    )
+
+    check = DockerStorage(execute_module=execute_module).run(tmp=None, task_vars=task_vars)
+
+    assert check["failed"]
+    for word in extra_words:
+        assert word in check["msg"]