fix(core): DinD issues #141, #329 (#368)

Fix #141 - find IP from custom network if the container is not using the
default network
Close #329 - This seems fixed in the underlying docker libraries.
Improve support for Docker in Docker running on a custom network, by
attempting to find the right custom network and use it for new
containers. This adds support for using testcontainers-python running
the GitHub Actions Runner Controller to run self-hosted actions runners
on prem, when you run your workflows in containers.

---------

Co-authored-by: Dee Moore <dee42moore@gmail.com>
Co-authored-by: David Ankin <daveankin@gmail.com>
Co-authored-by: Balint Bartha <39852431+totallyzen@users.noreply.github.com>

core/testcontainers/core/docker_client.py

@@ -11,8 +11,10 @@
 #    License for the specific language governing permissions and limitations
 #    under the License.
 import functools as ft
+import ipaddress
 import os
 import urllib
+import urllib.parse
 from os.path import exists
 from pathlib import Path
 from typing import Optional, Union
@@ -34,7 +36,7 @@ class DockerClient:
     """
 
     def __init__(self, **kwargs) -> None:
-        docker_host = read_tc_properties().get("tc.host")
+        docker_host = get_docker_host()
 
         if docker_host:
             LOGGER.info(f"using host {docker_host}")
@@ -57,6 +59,12 @@ def run(
         remove: bool = False,
         **kwargs,
     ) -> Container:
+        # If the user has specified a network, we'll assume the user knows best
+        if "network" not in kwargs and not get_docker_host():
+            # Otherwise we'll try to find the docker host for dind usage.
+            host_network = self.find_host_network()
+            if host_network:
+                kwargs["network"] = host_network
         container = self.client.containers.run(
             image,
             command=command,
@@ -71,6 +79,30 @@ def run(
         )
         return container
 
+    def find_host_network(self) -> Optional[str]:
+        """
+        Try to find the docker host network.
+
+        :return: The network name if found, None if not set.
+        """
+        # If we're docker in docker running on a custom network, we need to inherit the
+        # network settings, so we can access the resulting container.
+        try:
+            docker_host = ipaddress.IPv4Address(self.host())
+            # See if we can find the host on our networks
+            for network in self.client.networks.list(filters={"type": "custom"}):
+                if "IPAM" in network.attrs:
+                    for config in network.attrs["IPAM"]["Config"]:
+                        try:
+                            subnet = ipaddress.IPv4Network(config["Subnet"])
+                        except ipaddress.AddressValueError:
+                            continue
+                        if docker_host in subnet:
+                            return network.name
+        except ipaddress.AddressValueError:
+            pass
+        return None
+
     def port(self, container_id: str, port: int) -> int:
         """
         Lookup the public-facing port that is NAT-ed to :code:`port`.
@@ -94,14 +126,26 @@ def bridge_ip(self, container_id: str) -> str:
         Get the bridge ip address for a container.
         """
         container = self.get_container(container_id)
-        return container["NetworkSettings"]["Networks"]["bridge"]["IPAddress"]
+        network_name = self.network_name(container_id)
+        return container["NetworkSettings"]["Networks"][network_name]["IPAddress"]
+
+    def network_name(self, container_id: str) -> str:
+        """
+        Get the name of the network this container runs on
+        """
+        container = self.get_container(container_id)
+        name = container["HostConfig"]["NetworkMode"]
+        if name == "default":
+            return "bridge"
+        return name
 
     def gateway_ip(self, container_id: str) -> str:
         """
         Get the gateway ip address for a container.
         """
         container = self.get_container(container_id)
-        return container["NetworkSettings"]["Networks"]["bridge"]["Gateway"]
+        network_name = self.network_name(container_id)
+        return container["NetworkSettings"]["Networks"][network_name]["Gateway"]
 
     def host(self) -> str:
         """
@@ -145,3 +189,7 @@ def read_tc_properties() -> dict[str, str]:
             tuples = [line.split("=") for line in contents.readlines() if "=" in line]
             settings = {**settings, **{item[0].strip(): item[1].strip() for item in tuples}}
     return settings
+
+
+def get_docker_host() -> Optional[str]:
+    return read_tc_properties().get("tc.host") or os.getenv("DOCKER_HOST")

core/tests/test_docker_in_docker.py

@@ -1,11 +1,28 @@
-import pytest
-
+import time
+import socket
 from testcontainers.core.container import DockerContainer
 from testcontainers.core.docker_client import DockerClient
 from testcontainers.core.waiting_utils import wait_for_logs
 
 
-@pytest.mark.xfail(reason="https://github.com/docker/docker-py/issues/2717")
+def _wait_for_dind_return_ip(client, dind):
+    # get ip address for DOCKER_HOST
+    # avoiding DockerContainer class here to prevent code changes affecting the test
+    docker_host_ip = client.bridge_ip(dind.id)
+    # Wait for startup
+    timeout = 10
+    start_wait = time.perf_counter()
+    while True:
+        try:
+            with socket.create_connection((docker_host_ip, 2375), timeout=timeout):
+                break
+        except ConnectionRefusedError:
+            if time.perf_counter() - start_wait > timeout:
+                raise RuntimeError("Docker in docker took longer than 10 seconds to start")
+            time.sleep(0.01)
+    return docker_host_ip
+
+
 def test_wait_for_logs_docker_in_docker():
     # real dind isn't possible (AFAIK) in CI
     # forwarding the socket to a container port is at least somewhat the same
@@ -18,21 +35,53 @@ def test_wait_for_logs_docker_in_docker():
     )
 
     not_really_dind.start()
+    docker_host_ip = _wait_for_dind_return_ip(client, not_really_dind)
+    docker_host = f"tcp://{docker_host_ip}:2375"
 
-    # get ip address for DOCKER_HOST
-    # avoiding DockerContainer class here to prevent code changes affecting the test
-    specs = client.get_container(not_really_dind.id)
-    docker_host_ip = specs["NetworkSettings"]["Networks"]["bridge"]["IPAddress"]
+    with DockerContainer(
+        image="hello-world",
+        docker_client_kw={"environment": {"DOCKER_HOST": docker_host, "DOCKER_CERT_PATH": "", "DOCKER_TLS_VERIFY": ""}},
+    ) as container:
+        assert container.get_container_host_ip() == docker_host_ip
+        wait_for_logs(container, "Hello from Docker!")
+        stdout, stderr = container.get_logs()
+        assert stdout, "There should be something on stdout"
+
+    not_really_dind.stop()
+    not_really_dind.remove()
+
+
+def test_dind_inherits_network():
+    client = DockerClient()
+    try:
+        custom_network = client.client.networks.create("custom_network", driver="bridge", check_duplicate=True)
+    except Exception:
+        custom_network = client.client.networks.list(names=["custom_network"])[0]
+    not_really_dind = client.run(
+        image="alpine/socat",
+        command="tcp-listen:2375,fork,reuseaddr unix-connect:/var/run/docker.sock",
+        volumes={"/var/run/docker.sock": {"bind": "/var/run/docker.sock"}},
+        detach=True,
+    )
+
+    not_really_dind.start()
+
+    docker_host_ip = _wait_for_dind_return_ip(client, not_really_dind)
     docker_host = f"tcp://{docker_host_ip}:2375"
 
     with DockerContainer(
         image="hello-world",
         docker_client_kw={"environment": {"DOCKER_HOST": docker_host, "DOCKER_CERT_PATH": "", "DOCKER_TLS_VERIFY": ""}},
     ) as container:
         assert container.get_container_host_ip() == docker_host_ip
+        # Check the gateways are the same, so they can talk to each other
+        assert container.get_docker_client().gateway_ip(container.get_wrapped_container().id) == client.gateway_ip(
+            not_really_dind.id
+        )
         wait_for_logs(container, "Hello from Docker!")
         stdout, stderr = container.get_logs()
         assert stdout, "There should be something on stdout"
 
     not_really_dind.stop()
     not_really_dind.remove()
+    custom_network.remove()