split up diagnose from init-config

Author: obs-gh-mattcotter

integration/scripts/test_configure.py

@@ -1,121 +1,45 @@
 #!/usr/bin/env python3
-from fabric import Result
-
-import re
 import utils as u
-import test_start as start
-
-
-def _diagnose_result_is_successful(result: Result) -> bool:
-    return re.search(r"All \d+ checks passed", result.stdout) is not None
 
 
 @u.print_test_decorator
 def run_test_windows(remote_host: u.Host, env_vars: dict) -> None:
-    """
-    Test to validate connection of observe-agent to Observe
-
-    Args:
-        remote_host (Host): instance to ssh into
-        env_vars (dict): environment variables passed into for testing
-
-    Raises:
-        ValueError: Something failed with initial config or observe-agent -> observe connection
-    """
-
     init_command = r'Set-Location "C:\Program Files\Observe\observe-agent"; ./observe-agent init-config --token {} --observe_url {}'.format(
         env_vars["observe_token"], env_vars["observe_url"]
     )
-    diagnose_command = r'Set-Location "C:\Program Files\Observe\observe-agent"; ./observe-agent diagnose'
 
     # Set up correct config with observe url and token
     result = remote_host.run_command(init_command)
-
-    start.run_test_windows(remote_host, env_vars)
-
-    # Check diagnose command
-    result = remote_host.run_command(diagnose_command)
-    if _diagnose_result_is_successful(result):
-        print(" ✅ observe-agent -> observe validation passed! ")
-    else:
+    if result.exited != 0 or result.stderr:
         u.print_remote_result(result)
-        raise ValueError(
-            f"❌ Failed: observe-agent -> observe validation (regex on diagnose output did not match)"
-        )
+        raise ValueError("❌ Error in init-config")
 
 
 @u.print_test_decorator
 def run_test_docker(remote_host: u.Host, env_vars: dict) -> None:
-    docker_prefix = u.get_docker_prefix(remote_host, u.get_random_test_name(), False)
-    result = remote_host.run_command(docker_prefix + " config")
-    u.print_remote_result(result)
-    u.die("❌ skipping docker test")
-    container_name = u.get_random_test_name()
-    env_vars["container_name"] = container_name
-    start.run_test_docker(remote_host, env_vars)
-
-    container_id = u.get_docker_container(remote_host, container_name)
-    exec_prefix = f"sudo docker exec {container_id} ./observe-agent"
+    docker_prefix = u.get_docker_prefix(remote_host, False)
     init_command = "{} init-config --token {} --observe_url {}".format(
-        exec_prefix, env_vars["observe_token"], env_vars["observe_url"]
+        docker_prefix, env_vars["observe_token"], env_vars["observe_url"]
     )
-    diagnose_command = exec_prefix + " diagnose"
 
     # Set up correct config with observe url and token
     result = remote_host.run_command(init_command)
     if result.exited != 0 or result.stderr:
         u.print_remote_result(result)
         raise ValueError("❌ Error in init-config")
 
-    # Restart the process
-    result = remote_host.run_command(f"sudo docker restart {container_id}")
-    if result.exited != 0 or result.stderr:
-        u.print_remote_result(result)
-        raise ValueError("❌ Error in restart")
-
-    # Check diagnose command
-    result = remote_host.run_command(diagnose_command)
-    if _diagnose_result_is_successful(result):
-        print(" ✅ observe-agent -> observe validation passed! ")
-    else:
-        u.print_remote_result(result)
-        raise ValueError(
-            f"❌ Failed: observe-agent -> observe validation (regex on diagnose output did not match)"
-        )
-
 
 @u.print_test_decorator
 def run_test_linux(remote_host: u.Host, env_vars: dict) -> None:
-    """
-    Test to validate connection of observe-agent to Observe
-
-    Args:
-        remote_host (Host): instance to ssh into
-        env_vars (dict): environment variables passed into for testing
-
-    Raises:
-        ValueError: Something failed with initial config or observe-agent -> observe connection
-    """
-
     init_command = "sudo observe-agent init-config --token {} --observe_url {}".format(
         env_vars["observe_token"], env_vars["observe_url"]
     )
-    diagnose_command = "observe-agent diagnose"
 
     # Set up correct config with observe url and token
     result = remote_host.run_command(init_command)
-
-    start.run_test_linux(remote_host, env_vars)
-
-    # Check diagnose command
-    result = remote_host.run_command(diagnose_command)
-    if _diagnose_result_is_successful(result):
-        print(" ✅ observe-agent -> observe validation passed! ")
-    else:
+    if result.exited != 0 or result.stderr:
         u.print_remote_result(result)
-        raise ValueError(
-            f"❌ Failed: observe-agent -> observe validation (regex on diagnose output did not match)"
-        )
+        raise ValueError("❌ Error in init-config")
 
 
 if __name__ == "__main__":

integration/scripts/test_diagnose.py

@@ -0,0 +1,69 @@
+#!/usr/bin/env python3
+from fabric import Result
+
+import re
+import utils as u
+
+
+def _check_diagnose_result(result: Result) -> bool:
+    passed = re.search(r"All \d+ checks passed", result.stdout) is not None
+    if passed:
+        print(" ✅ observe-agent -> observe validation passed! ")
+    else:
+        u.print_remote_result(result)
+        raise ValueError(
+            f"❌ Failed: observe-agent -> observe validation (regex on diagnose output did not match)"
+        )
+
+
+@u.print_test_decorator
+def run_test_windows(remote_host: u.Host, env_vars: dict) -> None:
+    diagnose_command = r'Set-Location "C:\Program Files\Observe\observe-agent"; ./observe-agent diagnose'
+
+    # Check diagnose command
+    result = remote_host.run_command(diagnose_command)
+    _check_diagnose_result(result)
+
+
+@u.print_test_decorator
+def run_test_docker(remote_host: u.Host, env_vars: dict) -> None:
+    container_id = u.get_docker_container(remote_host)
+    exec_prefix = f"sudo docker exec {container_id} ./observe-agent"
+    diagnose_command = exec_prefix + " diagnose"
+
+    # Check diagnose command
+    result = remote_host.run_command(diagnose_command)
+    _check_diagnose_result(result)
+
+
+@u.print_test_decorator
+def run_test_linux(remote_host: u.Host, env_vars: dict) -> None:
+    diagnose_command = "observe-agent diagnose"
+
+    # Check diagnose command
+    result = remote_host.run_command(diagnose_command)
+    _check_diagnose_result(result)
+
+
+if __name__ == "__main__":
+
+    env_vars = u.get_env_vars(need_observe=True)
+    remote_host = u.Host(
+        host_ip=env_vars["host"],
+        username=env_vars["user"],
+        key_file_path=env_vars["key_filename"],
+        password=env_vars["password"],
+    )
+
+    # Test SSH Connection before starting test of interest
+    remote_host.test_conection(int(env_vars["machine_config"]["sleep"]))
+
+    if (
+        "redhat" in env_vars["machine_config"]["distribution"]
+        or "debian" in env_vars["machine_config"]["distribution"]
+    ):
+        run_test_linux(remote_host, env_vars)
+    elif "windows" in env_vars["machine_config"]["distribution"]:
+        run_test_windows(remote_host, env_vars)
+    elif "docker" in env_vars["machine_config"]["distribution"]:
+        run_test_docker(remote_host, env_vars)

integration/scripts/test_start.py

@@ -25,7 +25,12 @@ def _check_status_loop(
     agent_status = False
     for _ in range(start_timeout):
         metrics_dict = defaultdict(list)
-        result = remote_host.run_command(status_command)
+        try:
+            result = remote_host.run_command(status_command)
+        except Exception as e:
+            print("Ignoring exception: ", e)
+            time.sleep(1)
+            continue
         for line in result.stdout.splitlines():
             if ":" in line:
                 metric, value = line.split(":", 1)
@@ -37,13 +42,12 @@ def _check_status_loop(
             print("✅ Observe Agent is active and running without errors!")
             agent_status = True
             break
-        else:
-            print(
-                "❌ Observe Agent is not running. Retry Count is {}/{}...".format(
-                    _ + 1, start_timeout
-                )
+        print(
+            "❌ Observe Agent is not running. Retry Count is {}/{}...".format(
+                _ + 1, start_timeout
             )
-            time.sleep(1)
+        )
+        time.sleep(1)
     return agent_status
 
 
@@ -93,9 +97,7 @@ def run_test_windows(remote_host: u.Host, env_vars: dict) -> None:
 
 @u.print_test_decorator
 def run_test_docker(remote_host: u.Host, env_vars: dict) -> None:
-    u.upload_default_docker_config(env_vars, remote_host)
-    container_name = env_vars.get("container_name", u.get_random_test_name())
-    docker_prefix = u.get_docker_prefix(remote_host, container_name, True)
+    docker_prefix = u.get_docker_prefix(remote_host, True)
     start_command = "start"
     start_timeout = 30  # how long to wait for observe-agent to start
 
@@ -107,7 +109,7 @@ def run_test_docker(remote_host: u.Host, env_vars: dict) -> None:
         print("✅ Observe Agent started successfully: " + result.stdout)
 
     # Get Observe Agent Container ID
-    container_id = u.get_docker_container(remote_host, container_name)
+    container_id = u.get_docker_container(remote_host)
     status_command = f"sudo docker exec {container_id} ./observe-agent status"
 
     # Check Agent Status

integration/scripts/test_version.py

@@ -71,7 +71,7 @@ def run_test_windows(remote_host: u.Host, env_vars: dict) -> None:
 @u.print_test_decorator
 def run_test_docker(remote_host: u.Host, env_vars: dict) -> None:
     u.upload_default_docker_config(env_vars, remote_host)
-    docker_prefix = u.get_docker_prefix(remote_host, u.get_random_test_name(), False)
+    docker_prefix = u.get_docker_prefix(remote_host, False)
     config_file_linux = "/etc/observe-agent/observe-agent.yaml"
     version_pattern = re.compile(r"^\d+\.\d+\.\d+(-[A-Za-z0-9-]+)?$")
 

integration/scripts/utils.py

@@ -5,7 +5,6 @@
 from paramiko.ssh_exception import AuthenticationException, NoValidConnectionsError
 
 import os
-import random
 import sys
 import time
 
@@ -248,12 +247,6 @@ def test_conection(self, timeout=60) -> None:
         raise RuntimeError(" ❌ The SSH connection failed")
 
 
-def get_random_test_name() -> str:
-    random.seed(time.time())
-    num_digits = 6
-    return "test-" + str(random.randint(10 ** (num_digits - 1), (10**num_digits) - 1))
-
-
 def get_docker_image(remote_host: Host) -> str:
     result = remote_host.run_command(
         'sudo docker images --format "{{.Repository}}:{{.Tag}}"'
@@ -265,17 +258,16 @@ def get_docker_image(remote_host: Host) -> str:
     return images[0]
 
 
-def get_docker_prefix(remote_host: Host, name: str, detach: bool) -> str:
+def get_docker_prefix(remote_host: Host, detach: bool) -> str:
     image = get_docker_image(remote_host)
-    return f'sudo docker run {"-d --restart always" if detach else ""} \
+    return f'sudo docker run {"-d --restart on-failure" if detach else ""} \
         --mount type=bind,source=/proc,target=/hostfs/proc,readonly \
         --mount type=bind,source=/snap,target=/hostfs/snap,readonly \
         --mount type=bind,source=/boot,target=/hostfs/boot,readonly \
         --mount type=bind,source=/var/lib,target=/hostfs/var/lib,readonly \
         --mount type=bind,source=/var/log,target=/hostfs/var/log,readonly \
         --mount type=bind,source=/var/lib/docker/containers,target=/var/lib/docker/containers,readonly \
         --mount type=bind,source=$(pwd)/observe-agent.yaml,target=/etc/observe-agent/observe-agent.yaml \
-        --name {name} \
         --pid host {image}'
 
 
@@ -290,12 +282,8 @@ def upload_default_docker_config(env_vars: dict, remote_host: Host) -> None:
     remote_host.put_file(local_path=observe_agent_file_path, remote_path=home_dir)
 
 
-def get_docker_container(remote_host: Host, name: str) -> str:
-    get_container_command = (
-        'sudo docker ps --filter "status=running" --filter "name='
-        + name
-        + '" --format "{{.ID}} {{.Image}} {{.CreatedAt}}"'
-    )
+def get_docker_container(remote_host: Host) -> str:
+    get_container_command = 'sudo docker ps --filter "status=running" --format "{{.ID}} {{.Image}} {{.CreatedAt}}"'
     result = remote_host.run_command(get_container_command)
     running = [
         line.strip() for line in result.stdout.splitlines() if "SNAPSHOT" in line

integration/tests/integration.tftest.hcl

@@ -15,7 +15,6 @@ run "setup_observe_variables" {
 }
 
 
-
 run "test_ec2_connection" {
   module {
     source  = "observeinc/collection/aws//modules/testing/exec"
@@ -41,8 +40,6 @@ run "test_ec2_connection" {
 }
 
 
-
-
 run "test_install" {
   module {
     source  = "observeinc/collection/aws//modules/testing/exec"
@@ -68,8 +65,6 @@ run "test_install" {
 }
 
 
-
-
 run "test_version" {
   module {
     source  = "observeinc/collection/aws//modules/testing/exec"
@@ -95,8 +90,6 @@ run "test_version" {
 }
 
 
-
-
 run "test_configure" {
   module {
     source  = "observeinc/collection/aws//modules/testing/exec"
@@ -123,6 +116,7 @@ run "test_configure" {
   }
 }
 
+
 run "test_start" {
   module {
     source  = "observeinc/collection/aws//modules/testing/exec"
@@ -147,3 +141,27 @@ run "test_start" {
   }
 }
 
+
+run "test_diagnose" {
+  module {
+    source  = "observeinc/collection/aws//modules/testing/exec"
+    version = "2.9.0"
+  }
+
+  variables {
+    command = "python3 ./scripts/test_diagnose.py"
+    env_vars = {
+      HOST           = run.setup_ec2.public_ip
+      USER           = run.setup_ec2.user_name
+      KEY_FILENAME   = run.setup_ec2.private_key_path
+      PASSWORD       = run.setup_ec2.password
+      MACHINE_NAME   = run.setup_ec2.machine_name
+      MACHINE_CONFIG = run.setup_ec2.machine_config
+    }
+  }
+
+  assert {
+    condition     = output.error == ""
+    error_message = "Error in Diagnose Test"
+  }
+}