Unix Sockets for HTTP Replication (matrix-org#15708)

Unix socket support for `federation` and `client` Listeners has existed now for a little while(since [1.81.0](matrix-org#15353)), but there was one last hold out before it could be complete: HTTP Replication communication. This should finish it up. The Listeners would have always worked, but would have had no way to be talked to/at.

---------

Co-authored-by: Eric Eastwood <[email protected]>
Co-authored-by: Olivier Wilkinson (reivilibre) <[email protected]>
Co-authored-by: Eric Eastwood <[email protected]>

Author: 3 people

changelog.d/15708.feature

@@ -0,0 +1 @@
+Add Unix Socket support for HTTP Replication Listeners. Document and provide usage instructions for utilizing Unix sockets in Synapse. Contributed by Jason Little.

docker/conf-workers/nginx.conf.j2

@@ -35,7 +35,11 @@ server {
 
     # Send all other traffic to the main process
     location ~* ^(\\/_matrix|\\/_synapse) {
+{% if using_unix_sockets %}
+        proxy_pass http://unix:/run/main_public.sock;
+{% else %}
         proxy_pass http://localhost:8080;
+{% endif %}
         proxy_set_header X-Forwarded-For $remote_addr;
         proxy_set_header X-Forwarded-Proto $scheme;
         proxy_set_header Host $host;

docker/conf-workers/shared.yaml.j2

@@ -6,6 +6,9 @@
 {% if enable_redis %}
 redis:
     enabled: true
+    {% if using_unix_sockets %}
+    path: /tmp/redis.sock
+    {% endif %}
 {% endif %}
 
 {% if appservice_registrations is not none %}

docker/conf-workers/supervisord.conf.j2

@@ -19,7 +19,11 @@ username=www-data
 autorestart=true
 
 [program:redis]
+{% if using_unix_sockets %}
+command=/usr/local/bin/prefix-log /usr/local/bin/redis-server --unixsocket /tmp/redis.sock
+{% else %}
 command=/usr/local/bin/prefix-log /usr/local/bin/redis-server
+{% endif %}
 priority=1
 stdout_logfile=/dev/stdout
 stdout_logfile_maxbytes=0

docker/conf-workers/worker.yaml.j2

@@ -8,7 +8,11 @@ worker_name: "{{ name }}"
 
 worker_listeners:
   - type: http
+{% if using_unix_sockets %}
+    path: "/run/worker.{{ port }}"
+{% else %}
     port: {{ port }}
+{% endif %}
 {% if listener_resources %}
     resources:
       - names:

docker/conf/homeserver.yaml

@@ -36,12 +36,17 @@ listeners:
 
   # Allow configuring in case we want to reverse proxy 8008
   # using another process in the same container
+{% if SYNAPSE_USE_UNIX_SOCKET %}
+  # Unix sockets don't care about TLS or IP addresses or ports
+  - path: '/run/main_public.sock'
+    type: http
+{% else %}
   - port: {{ SYNAPSE_HTTP_PORT or 8008 }}
     tls: false
     bind_addresses: ['::']
     type: http
     x_forwarded: false
-
+{% endif %}
     resources:
       - names: [client]
         compress: true
@@ -57,8 +62,11 @@ database:
     user: "{{ POSTGRES_USER or "synapse" }}"
     password: "{{ POSTGRES_PASSWORD }}"
     database: "{{ POSTGRES_DB or "synapse" }}"
+{% if not SYNAPSE_USE_UNIX_SOCKET %}
+{# Synapse will use a default unix socket for Postgres when host/port is not specified (behavior from `psycopg2`). #}
     host: "{{ POSTGRES_HOST or "db" }}"
     port: "{{ POSTGRES_PORT or "5432" }}"
+{% endif %}
     cp_min: 5
     cp_max: 10
 {% else %}

docker/configure_workers_and_start.py

@@ -74,6 +74,9 @@
 MAIN_PROCESS_INSTANCE_NAME = "main"
 MAIN_PROCESS_LOCALHOST_ADDRESS = "127.0.0.1"
 MAIN_PROCESS_REPLICATION_PORT = 9093
+# Obviously, these would only be used with the UNIX socket option
+MAIN_PROCESS_UNIX_SOCKET_PUBLIC_PATH = "/run/main_public.sock"
+MAIN_PROCESS_UNIX_SOCKET_PRIVATE_PATH = "/run/main_private.sock"
 
 # A simple name used as a placeholder in the WORKERS_CONFIG below. This will be replaced
 # during processing with the name of the worker.
@@ -407,11 +410,15 @@ def add_worker_roles_to_shared_config(
         )
 
         # Map of stream writer instance names to host/ports combos
-        instance_map[worker_name] = {
-            "host": "localhost",
-            "port": worker_port,
-        }
-
+        if os.environ.get("SYNAPSE_USE_UNIX_SOCKET", False):
+            instance_map[worker_name] = {
+                "path": f"/run/worker.{worker_port}",
+            }
+        else:
+            instance_map[worker_name] = {
+                "host": "localhost",
+                "port": worker_port,
+            }
     # Update the list of stream writers. It's convenient that the name of the worker
     # type is the same as the stream to write. Iterate over the whole list in case there
     # is more than one.
@@ -423,10 +430,15 @@ def add_worker_roles_to_shared_config(
 
             # Map of stream writer instance names to host/ports combos
             # For now, all stream writers need http replication ports
-            instance_map[worker_name] = {
-                "host": "localhost",
-                "port": worker_port,
-            }
+            if os.environ.get("SYNAPSE_USE_UNIX_SOCKET", False):
+                instance_map[worker_name] = {
+                    "path": f"/run/worker.{worker_port}",
+                }
+            else:
+                instance_map[worker_name] = {
+                    "host": "localhost",
+                    "port": worker_port,
+                }
 
 
 def merge_worker_template_configs(
@@ -718,17 +730,29 @@ def generate_worker_files(
     # Note that yaml cares about indentation, so care should be taken to insert lines
     # into files at the correct indentation below.
 
+    # Convenience helper for if using unix sockets instead of host:port
+    using_unix_sockets = environ.get("SYNAPSE_USE_UNIX_SOCKET", False)
     # First read the original config file and extract the listeners block. Then we'll
     # add another listener for replication. Later we'll write out the result to the
     # shared config file.
-    listeners = [
-        {
-            "port": MAIN_PROCESS_REPLICATION_PORT,
-            "bind_address": MAIN_PROCESS_LOCALHOST_ADDRESS,
-            "type": "http",
-            "resources": [{"names": ["replication"]}],
-        }
-    ]
+    listeners: List[Any]
+    if using_unix_sockets:
+        listeners = [
+            {
+                "path": MAIN_PROCESS_UNIX_SOCKET_PRIVATE_PATH,
+                "type": "http",
+                "resources": [{"names": ["replication"]}],
+            }
+        ]
+    else:
+        listeners = [
+            {
+                "port": MAIN_PROCESS_REPLICATION_PORT,
+                "bind_address": MAIN_PROCESS_LOCALHOST_ADDRESS,
+                "type": "http",
+                "resources": [{"names": ["replication"]}],
+            }
+        ]
     with open(config_path) as file_stream:
         original_config = yaml.safe_load(file_stream)
         original_listeners = original_config.get("listeners")
@@ -769,7 +793,17 @@ def generate_worker_files(
 
     # A list of internal endpoints to healthcheck, starting with the main process
     # which exists even if no workers do.
-    healthcheck_urls = ["http://localhost:8080/health"]
+    # This list ends up being part of the command line to curl, (curl added support for
+    # Unix sockets in version 7.40).
+    if using_unix_sockets:
+        healthcheck_urls = [
+            f"--unix-socket {MAIN_PROCESS_UNIX_SOCKET_PUBLIC_PATH} "
+            # The scheme and hostname from the following URL are ignored.
+            # The only thing that matters is the path `/health`
+            "http://localhost/health"
+        ]
+    else:
+        healthcheck_urls = ["http://localhost:8080/health"]
 
     # Get the set of all worker types that we have configured
     all_worker_types_in_use = set(chain(*requested_worker_types.values()))
@@ -806,8 +840,12 @@ def generate_worker_files(
         # given worker_type needs to stay assigned and not be replaced.
         worker_config["shared_extra_conf"].update(shared_config)
         shared_config = worker_config["shared_extra_conf"]
-
-        healthcheck_urls.append("http://localhost:%d/health" % (worker_port,))
+        if using_unix_sockets:
+            healthcheck_urls.append(
+                f"--unix-socket /run/worker.{worker_port} http://localhost/health"
+            )
+        else:
+            healthcheck_urls.append("http://localhost:%d/health" % (worker_port,))
 
         # Update the shared config with sharding-related options if necessary
         add_worker_roles_to_shared_config(
@@ -826,6 +864,7 @@ def generate_worker_files(
             "/conf/workers/{name}.yaml".format(name=worker_name),
             **worker_config,
             worker_log_config_filepath=log_config_filepath,
+            using_unix_sockets=using_unix_sockets,
         )
 
         # Save this worker's port number to the correct nginx upstreams
@@ -846,8 +885,13 @@ def generate_worker_files(
     nginx_upstream_config = ""
     for upstream_worker_base_name, upstream_worker_ports in nginx_upstreams.items():
         body = ""
-        for port in upstream_worker_ports:
-            body += f"    server localhost:{port};\n"
+        if using_unix_sockets:
+            for port in upstream_worker_ports:
+                body += f"    server unix:/run/worker.{port};\n"
+
+        else:
+            for port in upstream_worker_ports:
+                body += f"    server localhost:{port};\n"
 
         # Add to the list of configured upstreams
         nginx_upstream_config += NGINX_UPSTREAM_CONFIG_BLOCK.format(
@@ -877,10 +921,15 @@ def generate_worker_files(
     # If there are workers, add the main process to the instance_map too.
     if workers_in_use:
         instance_map = shared_config.setdefault("instance_map", {})
-        instance_map[MAIN_PROCESS_INSTANCE_NAME] = {
-            "host": MAIN_PROCESS_LOCALHOST_ADDRESS,
-            "port": MAIN_PROCESS_REPLICATION_PORT,
-        }
+        if using_unix_sockets:
+            instance_map[MAIN_PROCESS_INSTANCE_NAME] = {
+                "path": MAIN_PROCESS_UNIX_SOCKET_PRIVATE_PATH,
+            }
+        else:
+            instance_map[MAIN_PROCESS_INSTANCE_NAME] = {
+                "host": MAIN_PROCESS_LOCALHOST_ADDRESS,
+                "port": MAIN_PROCESS_REPLICATION_PORT,
+            }
 
     # Shared homeserver config
     convert(
@@ -890,6 +939,7 @@ def generate_worker_files(
         appservice_registrations=appservice_registrations,
         enable_redis=workers_in_use,
         workers_in_use=workers_in_use,
+        using_unix_sockets=using_unix_sockets,
     )
 
     # Nginx config
@@ -900,6 +950,7 @@ def generate_worker_files(
         upstream_directives=nginx_upstream_config,
         tls_cert_path=os.environ.get("SYNAPSE_TLS_CERT"),
         tls_key_path=os.environ.get("SYNAPSE_TLS_KEY"),
+        using_unix_sockets=using_unix_sockets,
     )
 
     # Supervisord config
@@ -909,6 +960,7 @@ def generate_worker_files(
         "/etc/supervisor/supervisord.conf",
         main_config_path=config_path,
         enable_redis=workers_in_use,
+        using_unix_sockets=using_unix_sockets,
     )
 
     convert(

docs/development/contributing_guide.md

@@ -370,6 +370,7 @@ The above will run a monolithic (single-process) Synapse with SQLite as the data
     See the [worker documentation](../workers.md) for additional information on workers.
 - Passing `ASYNCIO_REACTOR=1` as an environment variable to use the Twisted asyncio reactor instead of the default one.
 - Passing `PODMAN=1` will use the [podman](https://podman.io/) container runtime, instead of docker.
+- Passing `UNIX_SOCKETS=1` will utilise Unix socket functionality for Synapse, Redis, and Postgres(when applicable).
 
 To increase the log level for the tests, set `SYNAPSE_TEST_LOG_LEVEL`, e.g:
 ```sh

docs/usage/configuration/config_documentation.md

@@ -462,6 +462,20 @@ See the docs [request log format](../administration/request_log.md).
 * `additional_resources`: Only valid for an 'http' listener. A map of
    additional endpoints which should be loaded via dynamic modules.
 
+Unix socket support (_Added in Synapse 1.88.0_):
+* `path`: A path and filename for a Unix socket. Make sure it is located in a
+  directory with read and write permissions, and that it already exists (the directory
+  will not be created). Defaults to `None`.
+  * **Note**: The use of both `path` and `port` options for the same `listener` is not
+    compatible.
+  * The `x_forwarded` option defaults to true  when using Unix sockets and can be omitted.
+  * Other options that would not make sense to use with a UNIX socket, such as 
+    `bind_addresses` and `tls` will be ignored and can be removed.
+* `mode`: The file permissions to set on the UNIX socket. Defaults to `666`
+* **Note:** Must be set as `type: http` (does not support `metrics` and `manhole`). 
+  Also make sure that `metrics` is not included in `resources` -> `names`
+
+
 Valid resource names are:
 
 * `client`: the client-server API (/_matrix/client), and the synapse admin API (/_synapse/admin). Also implies `media` and `static`.
@@ -474,7 +488,7 @@ Valid resource names are:
 
 * `media`: the media API (/_matrix/media).
 
-* `metrics`: the metrics interface. See [here](../../metrics-howto.md).
+* `metrics`: the metrics interface. See [here](../../metrics-howto.md). (Not compatible with Unix sockets)
 
 * `openid`: OpenID authentication. See [here](../../openid.md).
 
@@ -533,6 +547,22 @@ listeners:
     bind_addresses: ['::1', '127.0.0.1']
     type: manhole
 ```
+Example configuration #3:
+```yaml
+listeners:
+  # Unix socket listener: Ideal for Synapse deployments behind a reverse proxy, offering
+  # lightweight interprocess communication without TCP/IP overhead, avoid port
+  # conflicts, and providing enhanced security through system file permissions.
+  #
+  # Note that x_forwarded will default to true, when using a UNIX socket. Please see
+  # https://matrix-org.github.io/synapse/latest/reverse_proxy.html.
+  #
+  - path: /var/run/synapse/main_public.sock
+    type: http
+    resources:
+      - names: [client, federation]
+```
+
 ---
 ### `manhole_settings`
 
@@ -3949,6 +3979,14 @@ instance_map:
     host: localhost
     port: 8034
 ```
+Example configuration(#2, for UNIX sockets):
+```yaml
+instance_map:
+  main:
+    path: /var/run/synapse/main_replication.sock
+  worker1:
+    path: /var/run/synapse/worker1_replication.sock
+```
 ---
 ### `stream_writers`
 
@@ -4108,6 +4146,18 @@ worker_listeners:
     resources:
       - names: [client, federation]
 ```
+Example configuration(#2, using UNIX sockets with a `replication` listener):
+```yaml
+worker_listeners:
+  - type: http
+    path: /var/run/synapse/worker_public.sock
+    resources:
+      - names: [client, federation]
+  - type: http
+    path: /var/run/synapse/worker_replication.sock
+    resources:
+      - names: [replication]
+```
 ---
 ### `worker_manhole`
 

docs/workers.md

@@ -95,9 +95,12 @@ for the main process
 * Secondly, you need to enable
 [redis-based replication](usage/configuration/config_documentation.md#redis)
 * You will need to add an [`instance_map`](usage/configuration/config_documentation.md#instance_map) 
-with the `main` process defined, as well as the relevant connection information from 
-it's HTTP `replication` listener (defined in step 1 above). Note that the `host` defined 
-is the address the worker needs to look for the `main` process at, not necessarily the same address that is bound to.
+with the `main` process defined, as well as the relevant connection information from
+it's HTTP `replication` listener (defined in step 1 above).
+  * Note that the `host` defined is the address the worker needs to look for the `main`
+  process at, not necessarily the same address that is bound to.
+  * If you are using Unix sockets for the `replication` resource, make sure to
+  use a `path` to the socket file instead of a `port`.
 * Optionally, a [shared secret](usage/configuration/config_documentation.md#worker_replication_secret)
 can be used to authenticate HTTP traffic between workers. For example:
 

scripts-dev/complement.sh

@@ -253,6 +253,10 @@ if [[ -n "$ASYNCIO_REACTOR" ]]; then
   export PASS_SYNAPSE_COMPLEMENT_USE_ASYNCIO_REACTOR=true
 fi
 
+if [[ -n "$UNIX_SOCKETS" ]]; then
+  # Enable full on Unix socket mode for Synapse, Redis and Postgresql
+  export PASS_SYNAPSE_USE_UNIX_SOCKET=1
+fi
 
 if [[ -n "$SYNAPSE_TEST_LOG_LEVEL" ]]; then
   # Set the log level to what is desired