frostfs-testlib/src/frostfs_testlib/healthcheck/basic_healthcheck.py

from frostfs_testlib.cli.frostfs_cli.cli import FrostfsCli
from frostfs_testlib.healthcheck.interfaces import Healthcheck
from frostfs_testlib.reporter import get_reporter
from frostfs_testlib.resources.cli import FROSTFS_CLI_EXEC
from frostfs_testlib.steps.node_management import storage_node_healthcheck
from frostfs_testlib.storage.cluster import ClusterNode

reporter = get_reporter()


class BasicHealthcheck(Healthcheck):
    @reporter.step_deco("Perform healthcheck for {cluster_node}")
    def perform(self, cluster_node: ClusterNode):
        result = self.storage_healthcheck(cluster_node)
        if result:
            raise AssertionError(result)

    @reporter.step_deco("Tree healthcheck on {cluster_node}")
    def tree_healthcheck(self, cluster_node: ClusterNode) -> str | None:
        host = cluster_node.host
        service_config = host.get_service_config(cluster_node.storage_node.name)
        wallet_path = service_config.attributes["wallet_path"]
        wallet_password = service_config.attributes["wallet_password"]

        shell = host.get_shell()
        wallet_config_path = f"/tmp/{cluster_node.storage_node.name}-config.yaml"
        wallet_config = f'wallet: {wallet_path}\npassword: "{wallet_password}"'
        shell.exec(f"echo '{wallet_config}' > {wallet_config_path}")

        remote_cli = FrostfsCli(
            shell,
            host.get_cli_config(FROSTFS_CLI_EXEC).exec_path,
            config_file=wallet_config_path,
        )
        result = remote_cli.tree.healthcheck(rpc_endpoint="127.0.0.1:8080")
        if result.return_code != 0:
            return f"Error during tree healthcheck (rc={result.return_code}): {result.stdout}. \n Stderr: {result.stderr}"

    @reporter.step_deco("Storage healthcheck on {cluster_node}")
    def storage_healthcheck(self, cluster_node: ClusterNode) -> str | None:
        result = storage_node_healthcheck(cluster_node.storage_node)
        if result.health_status != "READY" or result.network_status != "ONLINE":
            return f"Node {cluster_node} is not healthy. Health={result.health_status}. Network={result.network_status}"