15876 FIX sap_hana_db_status: Be ok for passive nodes

Change-Id: Iaf3ba01dc71386081ba2f1d3b1ee843ca90784d4

.werks/15876

@@ -0,0 +1,14 @@
+Title: sap_hana_db_status: Be ok for passive nodes
+Class: fix
+Compatible: compat
+Component: checks
+Date: 1698058635
+Edition: cre
+Knowledge: undoc
+Level: 1
+State: unknown
+Version: 2.1.0p36
+
+The service <i>"SAP HANA Database Status"</i> went critical on a passive node.
+In now remains ok and reports <i>"System is in passive mode"</i>.
+

cmk/base/plugins/agent_based/sap_hana_db_status.py

@@ -4,16 +4,18 @@
 # This file is part of Checkmk (https://checkmk.com). It is subject to the terms and
 # conditions defined in the file COPYING, which is part of this source code package.
 
-from typing import Dict
+from typing import Mapping, Optional
 
 from .agent_based_api.v1 import IgnoreResultsError, register, Result, Service, State
 from .agent_based_api.v1.type_defs import CheckResult, DiscoveryResult, StringTable
 from .utils import sap_hana
 
+SectionDBStatus = Mapping[str, str]
+
 MAP_DB_STATUS = {"OK": State.OK, "WARNING": State.WARN}
 
 
-def parse_sap_hana_db_status(string_table: StringTable) -> Dict[str, str]:
+def parse_sap_hana_db_status(string_table: StringTable) -> SectionDBStatus:
     return {
         sid_instance: lines[0][0] if lines else ""
         for sid_instance, lines in sap_hana.parse_sap_hana(string_table).items()
@@ -26,23 +28,51 @@ def parse_sap_hana_db_status(string_table: StringTable) -> Dict[str, str]:
 )
 
 
-def discovery_sap_hana_db_status(section: sap_hana.ParsedSection) -> DiscoveryResult:
-    for item in section:
+def discover_sap_hana_db_status(
+    section_sap_hana_db_status: Optional[SectionDBStatus],
+    section_sap_hana_replication_status: Optional[sap_hana.ParsedSection],
+) -> DiscoveryResult:
+    if not section_sap_hana_db_status:
+        return
+    for item in section_sap_hana_db_status:
         yield Service(item=item)
 
 
-def check_sap_hana_db_status(item: str, section: Dict[str, str]) -> CheckResult:
-    db_status = section.get(item)
+def check_sap_hana_db_status(
+    item: str,
+    section_sap_hana_db_status: Optional[SectionDBStatus],
+    section_sap_hana_replication_status: Optional[sap_hana.ParsedSection],
+) -> CheckResult:
+    if section_sap_hana_db_status is None:
+        return
+
+    db_status = section_sap_hana_db_status.get(item)
 
     if not db_status:
         raise IgnoreResultsError("Login into database failed.")
 
+    db_state = MAP_DB_STATUS.get(db_status, State.CRIT)
+    repl_state = (
+        None
+        if section_sap_hana_replication_status is None
+        else section_sap_hana_replication_status.get(item, {}).get("sys_repl_status")
+    )
+
+    if (
+        db_state is State.CRIT
+        and repl_state is not None
+        and sap_hana.get_replication_state(repl_state)[1] == "passive"
+    ):
+        yield Result(state=State.OK, summary="System is in passive mode")
+        return
+
     yield Result(state=MAP_DB_STATUS.get(db_status, State.CRIT), summary=db_status)
 
 
 register.check_plugin(
     name="sap_hana_db_status",
     service_name="SAP HANA Database Status %s",
-    discovery_function=discovery_sap_hana_db_status,
+    sections=["sap_hana_db_status", "sap_hana_replication_status"],
+    discovery_function=discover_sap_hana_db_status,
     check_function=check_sap_hana_db_status,
 )

cmk/base/plugins/agent_based/sap_hana_replication_status.py

@@ -4,24 +4,13 @@
 # This file is part of Checkmk (https://checkmk.com). It is subject to the terms and
 # conditions defined in the file COPYING, which is part of this source code package.
 
-from typing import Any, Final, Mapping
+from collections.abc import Mapping
+from typing import Any
 
-from .agent_based_api.v1 import IgnoreResultsError, register, Result, Service, State
+from .agent_based_api.v1 import IgnoreResultsError, register, Result, Service
 from .agent_based_api.v1.type_defs import CheckResult, DiscoveryResult, StringTable
 from .utils import sap_hana
 
-SAP_HANA_REPL_STATUS_MAP: Final = {
-    "0": (State.UNKNOWN, "unknown status from replication script", "state_unknown"),
-    "10": (State.CRIT, "no system replication", "state_no_replication"),
-    "11": (State.CRIT, "error", "state_error"),
-    # "12" accuatly stands for "unknown replication status", but as per customer"s information
-    # (see SUP-1436), this should be indicated as "passive" replication aka secondary SAP HANA node.
-    "12": (State.OK, "passive", "state_replication_unknown"),
-    "13": (State.WARN, "initializing", "state_initializing"),
-    "14": (State.OK, "syncing", "state_syncing"),
-    "15": (State.OK, "active", "state_active"),
-}
-
 
 def parse_sap_hana_replication_status(string_table: StringTable) -> sap_hana.ParsedSection:
     section: sap_hana.ParsedSection = {}
@@ -62,10 +51,7 @@ def check_sap_hana_replication_status(
     if not data:
         raise IgnoreResultsError("Login into database failed.")
 
-    sys_repl_status = data["sys_repl_status"]
-    state, state_readable, param_key = SAP_HANA_REPL_STATUS_MAP.get(
-        sys_repl_status, (State.UNKNOWN, "unknown[%s]" % sys_repl_status, "state_unknown")
-    )
+    state, state_readable, param_key = sap_hana.get_replication_state(data["sys_repl_status"])
 
     yield Result(
         state=params.get(param_key, state), summary="System replication: %s" % state_readable

cmk/base/plugins/agent_based/utils/sap_hana.py

@@ -41,3 +41,24 @@ def parse_sap_hana_cluster_aware(info):
         elif instance is not None:
             instance.append([e.strip('"') for e in line])
     return parsed
+
+
+def get_replication_state(raw: str) -> tuple[State, str, str]:
+    if raw == "0":
+        return State.UNKNOWN, "unknown status from replication script", "state_unknown"
+    if raw == "10":
+        return State.CRIT, "no system replication", "state_no_replication"
+    if raw == "11":
+        return State.CRIT, "error", "state_error"
+    if raw == "12":
+        # "12" actually stands for "unknown replication status", but as per customer's information
+        # (see SUP-1436), this should be indicated as "passive" replication aka secondary SAP HANA node.
+        return State.OK, "passive", "state_replication_unknown"
+    if raw == "13":
+        return State.WARN, "initializing", "state_initializing"
+    if raw == "14":
+        return State.OK, "syncing", "state_syncing"
+    if raw == "15":
+        return State.OK, "active", "state_active"
+
+    return State.UNKNOWN, "unknown[%s]" % raw, "state_unknown"

tests/unit/cmk/base/plugins/agent_based/test_sap_hana_db_status.py

@@ -4,16 +4,18 @@
 # This file is part of Checkmk (https://checkmk.com). It is subject to the terms and
 # conditions defined in the file COPYING, which is part of this source code package.
 
-import pytest
+from typing import Mapping, Sequence
 
-from cmk.utils.type_defs import CheckPluginName, SectionName
+import pytest
 
+from cmk.base.plugins.agent_based import sap_hana_db_status as shds
 from cmk.base.plugins.agent_based.agent_based_api.v1 import (
     IgnoreResultsError,
     Result,
     Service,
     State,
 )
+from cmk.base.plugins.agent_based.agent_based_api.v1.type_defs import CheckResult, StringTable
 
 
 @pytest.mark.parametrize(
@@ -28,9 +30,8 @@
         )
     ],
 )
-def test_parse_sap_hana_db_status(fix_register, info, expected_result):
-    section_plugin = fix_register.agent_sections[SectionName("sap_hana_db_status")]
-    assert section_plugin.parse_function(info) == expected_result
+def test_parse_sap_hana_db_status(info: StringTable, expected_result: Mapping[str, str]) -> None:
+    assert shds.parse_sap_hana_db_status(info) == expected_result
 
 
 @pytest.mark.parametrize(
@@ -45,10 +46,13 @@ def test_parse_sap_hana_db_status(fix_register, info, expected_result):
         ),
     ],
 )
-def test_inventory_sap_hana_db_status(fix_register, info, expected_result):
-    section = fix_register.agent_sections[SectionName("sap_hana_db_status")].parse_function(info)
-    plugin = fix_register.check_plugins[CheckPluginName("sap_hana_db_status")]
-    assert list(plugin.discovery_function(section)) == expected_result
+def test_inventory_sap_hana_db_status(
+    info: StringTable, expected_result: Sequence[Service]
+) -> None:
+    assert (
+        list(shds.discover_sap_hana_db_status(shds.parse_sap_hana_db_status(info), None))
+        == expected_result
+    )
 
 
 @pytest.mark.parametrize(
@@ -83,10 +87,11 @@ def test_inventory_sap_hana_db_status(fix_register, info, expected_result):
         ),
     ],
 )
-def test_check_sap_hana_db_status(fix_register, item, info, expected_result):
-    section = fix_register.agent_sections[SectionName("sap_hana_db_status")].parse_function(info)
-    plugin = fix_register.check_plugins[CheckPluginName("sap_hana_db_status")]
-    assert list(plugin.check_function(item, section)) == expected_result
+def test_check_sap_hana_db_status(
+    item: str, info: StringTable, expected_result: CheckResult
+) -> None:
+    section = shds.parse_sap_hana_db_status(info)
+    assert list(shds.check_sap_hana_db_status(item, section, None)) == expected_result
 
 
 @pytest.mark.parametrize(
@@ -100,8 +105,25 @@ def test_check_sap_hana_db_status(fix_register, item, info, expected_result):
         ),
     ],
 )
-def test_check_sap_hana_db_status_stale(fix_register, item, info):
-    section = fix_register.agent_sections[SectionName("sap_hana_db_status")].parse_function(info)
-    plugin = fix_register.check_plugins[CheckPluginName("sap_hana_db_status")]
+def test_check_sap_hana_db_status_stale(item: str, info: StringTable) -> None:
+    section = shds.parse_sap_hana_db_status(info)
     with pytest.raises(IgnoreResultsError):
-        list(plugin.check_function(item, section))
+        list(shds.check_sap_hana_db_status(item, section, None))
+
+
+def test_check_sap_hana_ddb_status_passive_ok() -> None:
+    section = shds.parse_sap_hana_db_status(
+        [
+            ["[[HXE 98]]"],
+            [
+                "Status: error, Details: hdbsql ERROR: * -10709: Connection failed (RTE:[89006] System call 'connect' failed, rc=113:No route to host"
+            ],
+        ]
+    )
+    assert all(
+        r.state is State.OK
+        for r in shds.check_sap_hana_db_status(
+            "HXE 98", section, {"HXE 98": {"sys_repl_status": "12"}}
+        )
+        if isinstance(r, Result)
+    )