Add test_smoke test re-open (#454)

BON4 · web-flow · commit 4310dc9980ed · 2024-05-03T09:11:24.000-03:00
* Smoke testing. Garbage ignorance. No resources conflicts.

* Smoke testing. Garbage ignorance. No resources conflicts.

* Smoke testing. Garbage ignorance. No resources conflicts.

* Smoke testing. Garbage ignorance. No resources conflicts.

* Smoke testing. Garbage ignorance. No resources conflicts.

* Smoke testing. Garbage ignorance. No resources conflicts.
diff --git a/tests/integration/ha_tests/helpers.py b/tests/integration/ha_tests/helpers.py
@@ -1,5 +1,6 @@
 # Copyright 2022 Canonical Ltd.
 # See LICENSE file for licensing details.
+import json
 import logging
 import os
 import random
@@ -24,6 +25,7 @@
 from ..helpers import (
     APPLICATION_NAME,
     db_connect,
+    execute_query_on_unit,
     get_patroni_cluster,
     get_unit_address,
     run_command_on_unit,
@@ -862,3 +864,102 @@ async def reused_full_cluster_recovery_storage(ops_test: OpsTest, unit_name) ->
         "/var/snap/charmed-postgresql/common/var/log/patroni/patroni.log*",
     )
     return True
+
+
+async def is_storage_exists(ops_test: OpsTest, storage_id: str) -> bool:
+    """Returns True if storage exists by provided storage ID."""
+    complete_command = [
+        "show-storage",
+        "-m",
+        f"{ops_test.controller_name}:{ops_test.model.info.name}",
+        storage_id,
+        "--format=json",
+    ]
+    return_code, stdout, _ = await ops_test.juju(*complete_command)
+    if return_code != 0:
+        if return_code == 1:
+            return storage_id in stdout
+        raise Exception(
+            "Expected command %s to succeed instead it failed: %s with code: ",
+            complete_command,
+            stdout,
+            return_code,
+        )
+    return storage_id in str(stdout)
+
+
+async def create_db(ops_test: OpsTest, app: str, db: str) -> None:
+    """Creates database with specified name."""
+    unit = ops_test.model.applications[app].units[0]
+    unit_address = await unit.get_public_address()
+    password = await get_password(ops_test, app)
+
+    conn = db_connect(unit_address, password)
+    conn.autocommit = True
+    cursor = conn.cursor()
+    cursor.execute(f"CREATE DATABASE {db};")
+    cursor.close()
+    conn.close()
+
+
+async def check_db(ops_test: OpsTest, app: str, db: str) -> bool:
+    """Returns True if database with specified name already exists."""
+    unit = ops_test.model.applications[app].units[0]
+    unit_address = await unit.get_public_address()
+    password = await get_password(ops_test, app)
+
+    assert password is not None
+
+    query = await execute_query_on_unit(
+        unit_address,
+        password,
+        f"select datname from pg_catalog.pg_database where datname = '{db}';",
+    )
+
+    if "ERROR" in query:
+        raise Exception(f"Database check is failed with postgresql err: {query}")
+
+    return db in query
+
+
+async def get_any_deatached_storage(ops_test: OpsTest) -> str:
+    """Returns any of the current available deatached storage."""
+    return_code, storages_list, stderr = await ops_test.juju(
+        "storage", "-m", f"{ops_test.controller_name}:{ops_test.model.info.name}", "--format=json"
+    )
+    if return_code != 0:
+        raise Exception(f"failed to get storages info with error: {stderr}")
+
+    parsed_storages_list = json.loads(storages_list)
+    for storage_name, storage in parsed_storages_list["storage"].items():
+        if (str(storage["status"]["current"]) == "detached") and (str(storage["life"] == "alive")):
+            return storage_name
+
+    raise Exception("failed to get deatached storage")
+
+
+async def check_password_auth(ops_test: OpsTest, unit_name: str) -> bool:
+    """Checks if "operator" password is valid for current postgresql db."""
+    stdout = await run_command_on_unit(
+        ops_test,
+        unit_name,
+        """grep -E 'password authentication failed for user' /var/snap/charmed-postgresql/common/var/log/postgresql/postgresql*""",
+    )
+    return 'password authentication failed for user "operator"' not in stdout
+
+
+async def remove_unit_force(ops_test: OpsTest, unit_name: str):
+    """Removes unit with --force --no-wait."""
+    app_name = unit_name.split("/")[0]
+    complete_command = ["remove-unit", f"{unit_name}", "--force", "--no-wait", "--no-prompt"]
+    return_code, stdout, _ = await ops_test.juju(*complete_command)
+    if return_code != 0:
+        raise Exception(
+            "Expected command %s to succeed instead it failed: %s with code: ",
+            complete_command,
+            stdout,
+            return_code,
+        )
+
+    for unit in ops_test.model.applications[app_name].units:
+        assert unit != unit_name
diff --git a/tests/integration/ha_tests/test_smoke.py b/tests/integration/ha_tests/test_smoke.py
@@ -0,0 +1,221 @@
+#!/usr/bin/env python3
+# Copyright 2021 Canonical Ltd.
+# See LICENSE file for licensing details.
+
+import logging
+from asyncio import TimeoutError
+
+import pytest
+from juju import tag
+from pytest_operator.plugin import OpsTest
+from tenacity import Retrying, stop_after_delay, wait_fixed
+
+from ..helpers import (
+    APPLICATION_NAME,
+    CHARM_SERIES,
+)
+from ..juju_ import juju_major_version
+from .helpers import (
+    add_unit_with_storage,
+    check_db,
+    check_password_auth,
+    create_db,
+    get_any_deatached_storage,
+    is_postgresql_ready,
+    is_storage_exists,
+    remove_unit_force,
+    storage_id,
+)
+
+TEST_DATABASE_NAME = "test_database"
+DUP_APPLICATION_NAME = "postgres-test-dup"
+
+logger = logging.getLogger(__name__)
+
+
+@pytest.mark.group(1)
+@pytest.mark.abort_on_fail
+async def test_app_force_removal(ops_test: OpsTest, charm: str):
+    """Remove unit with force while storage is alive."""
+    async with ops_test.fast_forward():
+        # Deploy the charm.
+        logger.info("deploying charm")
+        await ops_test.model.deploy(
+            charm,
+            application_name=APPLICATION_NAME,
+            num_units=1,
+            series=CHARM_SERIES,
+            storage={"pgdata": {"pool": "lxd-btrfs", "size": 8046}},
+            config={"profile": "testing"},
+        )
+
+        logger.info("waiting for idle")
+        await ops_test.model.wait_for_idle(apps=[APPLICATION_NAME], status="active", timeout=1500)
+        assert ops_test.model.applications[APPLICATION_NAME].units[0].workload_status == "active"
+
+        primary_name = ops_test.model.applications[APPLICATION_NAME].units[0].name
+
+        logger.info("waiting for postgresql")
+        for attempt in Retrying(stop=stop_after_delay(15 * 3), wait=wait_fixed(3), reraise=True):
+            with attempt:
+                assert await is_postgresql_ready(ops_test, primary_name)
+
+        logger.info("getting storage id")
+        storage_id_str = storage_id(ops_test, primary_name)
+
+        # Check if storage exists after application deployed
+        logger.info("werifing is storage exists")
+        for attempt in Retrying(stop=stop_after_delay(15 * 3), wait=wait_fixed(3), reraise=True):
+            with attempt:
+                assert await is_storage_exists(ops_test, storage_id_str)
+
+        # Create test database to check there is no resources conflicts
+        logger.info("creating db")
+        await create_db(ops_test, APPLICATION_NAME, TEST_DATABASE_NAME)
+
+        # Check that test database is not exists for new unit
+        logger.info("checking db")
+        assert await check_db(ops_test, APPLICATION_NAME, TEST_DATABASE_NAME)
+
+        # Destroy charm
+        logger.info("force removing charm")
+        if juju_major_version == 2:
+            await remove_unit_force(ops_test, primary_name)
+        else:
+            await ops_test.model.destroy_unit(
+                primary_name, force=True, destroy_storage=False, max_wait=1500
+            )
+
+        # Storage should remain
+        logger.info("werifing is storage exists")
+        for attempt in Retrying(stop=stop_after_delay(15 * 3), wait=wait_fixed(3), reraise=True):
+            with attempt:
+                assert await is_storage_exists(ops_test, storage_id_str)
+
+
+@pytest.mark.group(1)
+@pytest.mark.abort_on_fail
+async def test_charm_garbage_ignorance(ops_test: OpsTest, charm: str):
+    """Test charm deploy in dirty environment with garbage storage."""
+    async with ops_test.fast_forward():
+        logger.info("checking garbage storage")
+        garbage_storage = None
+        for attempt in Retrying(stop=stop_after_delay(30 * 3), wait=wait_fixed(3), reraise=True):
+            with attempt:
+                garbage_storage = await get_any_deatached_storage(ops_test)
+
+        logger.info("add unit with attached storage")
+        await add_unit_with_storage(ops_test, APPLICATION_NAME, garbage_storage)
+
+        primary_name = ops_test.model.applications[APPLICATION_NAME].units[0].name
+
+        logger.info("waiting for postgresql")
+        for attempt in Retrying(stop=stop_after_delay(15 * 3), wait=wait_fixed(3), reraise=True):
+            with attempt:
+                assert await is_postgresql_ready(ops_test, primary_name)
+
+        logger.info("getting storage id")
+        storage_id_str = storage_id(ops_test, primary_name)
+
+        assert storage_id_str == garbage_storage
+
+        # Check if storage exists after application deployed
+        logger.info("werifing is storage exists")
+        for attempt in Retrying(stop=stop_after_delay(15 * 3), wait=wait_fixed(3), reraise=True):
+            with attempt:
+                assert await is_storage_exists(ops_test, storage_id_str)
+
+        # Check that test database exists for new unit
+        logger.info("checking db")
+        assert await check_db(ops_test, APPLICATION_NAME, TEST_DATABASE_NAME)
+
+        logger.info("removing charm")
+        await ops_test.model.destroy_unit(primary_name)
+
+
+@pytest.mark.group(1)
+@pytest.mark.abort_on_fail
+@pytest.mark.skipif(juju_major_version < 3, reason="Requires juju 3 or higher")
+async def test_app_resources_conflicts_v3(ops_test: OpsTest, charm: str):
+    """Test application deploy in dirty environment with garbage storage from another application."""
+    async with ops_test.fast_forward():
+        logger.info("checking garbage storage")
+        garbage_storage = None
+        for attempt in Retrying(stop=stop_after_delay(30 * 3), wait=wait_fixed(3), reraise=True):
+            with attempt:
+                garbage_storage = await get_any_deatached_storage(ops_test)
+
+        logger.info("deploying duplicate application with attached storage")
+        await ops_test.model.deploy(
+            charm,
+            application_name=DUP_APPLICATION_NAME,
+            num_units=1,
+            series=CHARM_SERIES,
+            attach_storage=[tag.storage(garbage_storage)],
+            config={"profile": "testing"},
+        )
+
+        # Reducing the update status frequency to speed up the triggering of deferred events.
+        await ops_test.model.set_config({"update-status-hook-interval": "10s"})
+
+        logger.info("waiting for duplicate application to be blocked")
+        try:
+            await ops_test.model.wait_for_idle(
+                apps=[DUP_APPLICATION_NAME], timeout=1000, status="blocked"
+            )
+        except TimeoutError:
+            logger.info("Application is not in blocked state. Checking logs...")
+
+        # Since application have postgresql db in storage from external application it should not be able to connect due to new password
+        logger.info("checking operator password auth")
+        assert not await check_password_auth(
+            ops_test, ops_test.model.applications[DUP_APPLICATION_NAME].units[0].name
+        )
+
+
+@pytest.mark.group(1)
+@pytest.mark.abort_on_fail
+@pytest.mark.skipif(juju_major_version != 2, reason="Requires juju 2")
+async def test_app_resources_conflicts_v2(ops_test: OpsTest, charm: str):
+    """Test application deploy in dirty environment with garbage storage from another application."""
+    async with ops_test.fast_forward():
+        logger.info("checking garbage storage")
+        garbage_storage = None
+        for attempt in Retrying(stop=stop_after_delay(30 * 3), wait=wait_fixed(3), reraise=True):
+            with attempt:
+                garbage_storage = await get_any_deatached_storage(ops_test)
+
+        # Deploy duplicaate charm
+        logger.info("deploying duplicate application")
+        await ops_test.model.deploy(
+            charm,
+            application_name=DUP_APPLICATION_NAME,
+            num_units=1,
+            series=CHARM_SERIES,
+            config={"profile": "testing"},
+        )
+
+        logger.info("force removing charm")
+        await remove_unit_force(
+            ops_test, ops_test.model.applications[DUP_APPLICATION_NAME].units[0].name
+        )
+
+        # Add unit with garbage storage
+        logger.info("adding charm with attached storage")
+        add_unit_cmd = f"add-unit {DUP_APPLICATION_NAME} --model={ops_test.model.info.name} --attach-storage={garbage_storage}".split()
+        return_code, _, _ = await ops_test.juju(*add_unit_cmd)
+        assert return_code == 0, "Failed to add unit with storage"
+
+        logger.info("waiting for duplicate application to be blocked")
+        try:
+            await ops_test.model.wait_for_idle(
+                apps=[DUP_APPLICATION_NAME], timeout=1000, status="blocked"
+            )
+        except TimeoutError:
+            logger.info("Application is not in blocked state. Checking logs...")
+
+        # Since application have postgresql db in storage from external application it should not be able to connect due to new password
+        logger.info("checking operator password auth")
+        assert not await check_password_auth(
+            ops_test, ops_test.model.applications[DUP_APPLICATION_NAME].units[0].name
+        )
