Add a dockerfile for running a set of Synapse worker processes (#9162)

This PR adds a Dockerfile and some supporting files to the `docker/` directory. The Dockerfile's intention is to spin up a container with:

* A Synapse main process.
* Any desired worker processes, defined by a `SYNAPSE_WORKERS` environment variable supplied at runtime.
* A redis for worker communication.
* A nginx for routing traffic.
* A supervisord to start all worker processes and monitor them if any go down.

Note that **this is not currently intended to be used in production**. If you'd like to use Synapse workers with Docker, instead make use of the official image, with one worker per container. The purpose of this dockerfile is currently to allow testing Synapse in worker mode with the [Complement](https://github.com/matrix-org/complement/) test suite.

`configure_workers_and_start.py` is where most of the magic happens in this PR. It reads from environment variables (documented in the file) and creates all necessary config files for the processes. It is the entrypoint of the Dockerfile, and thus is run any time the docker container is spun up, recreating all config files in case you want to use a different set of workers. One can specify which workers they'd like to use by setting the `SYNAPSE_WORKERS` environment variable (as a comma-separated list of arbitrary worker names) or by setting it to `*` for all worker processes. We will be using the latter in CI.

Huge thanks to @MatMaul for helping get this all working :tada: This PR is paired with its equivalent on the Complement side: https://github.com/matrix-org/complement/pull/62.

Note, for the purpose of testing this PR before it's merged: You'll need to (re)build the base Synapse docker image for everything to work (`matrixdotorg/synapse:latest`). Then build the worker-based docker image on top (`matrixdotorg/synapse:workers`).

changelog.d/9162.misc

@@ -0,0 +1 @@
+Add a dockerfile for running Synapse in worker-mode under Complement.

docker/Dockerfile-workers

@@ -0,0 +1,23 @@
+# Inherit from the official Synapse docker image
+FROM matrixdotorg/synapse
+
+# Install deps
+RUN apt-get update
+RUN apt-get install -y supervisor redis nginx
+
+# Remove the default nginx sites
+RUN rm /etc/nginx/sites-enabled/default
+
+# Copy Synapse worker, nginx and supervisord configuration template files
+COPY ./docker/conf-workers/* /conf/
+
+# Expose nginx listener port
+EXPOSE 8080/tcp
+
+# Volume for user-editable config files, logs etc.
+VOLUME ["/data"]
+
+# A script to read environment variables and create the necessary
+# files to run the desired worker configuration. Will start supervisord.
+COPY ./docker/configure_workers_and_start.py /configure_workers_and_start.py
+ENTRYPOINT ["/configure_workers_and_start.py"]

docker/README-testing.md

@@ -0,0 +1,140 @@
+# Running tests against a dockerised Synapse
+
+It's possible to run integration tests against Synapse
+using [Complement](https://github.com/matrix-org/complement). Complement is a Matrix Spec
+compliance test suite for homeservers, and supports any homeserver docker image configured
+to listen on ports 8008/8448. This document contains instructions for building Synapse
+docker images that can be run inside Complement for testing purposes.
+
+Note that running Synapse's unit tests from within the docker image is not supported.
+
+## Testing with SQLite and single-process Synapse
+
+> Note that `scripts-dev/complement.sh` is a script that will automatically build 
+> and run an SQLite-based, single-process of Synapse against Complement.
+
+The instructions below will set up Complement testing for a single-process, 
+SQLite-based Synapse deployment.
+
+Start by building the base Synapse docker image. If you wish to run tests with the latest
+release of Synapse, instead of your current checkout, you can skip this step. From the
+root of the repository:
+
+```sh
+docker build -t matrixdotorg/synapse -f docker/Dockerfile .
+```
+
+This will build an image with the tag `matrixdotorg/synapse`.
+
+Next, build the Synapse image for Complement. You will need a local checkout 
+of Complement. Change to the root of your Complement checkout and run:
+
+```sh
+docker build -t complement-synapse -f "dockerfiles/Synapse.Dockerfile" dockerfiles
+```
+
+This will build an image with the tag `complement-synapse`, which can be handed to 
+Complement for testing via the `COMPLEMENT_BASE_IMAGE` environment variable. Refer to 
+[Complement's documentation](https://github.com/matrix-org/complement/#running) for 
+how to run the tests, as well as the various available command line flags.
+
+## Testing with PostgreSQL and single or multi-process Synapse
+
+The above docker image only supports running Synapse with SQLite and in a 
+single-process topology. The following instructions are used to build a Synapse image for 
+Complement that supports either single or multi-process topology with a PostgreSQL 
+database backend.
+
+As with the single-process image, build the base Synapse docker image. If you wish to run
+tests with the latest release of Synapse, instead of your current checkout, you can skip
+this step. From the root of the repository:
+
+```sh
+docker build -t matrixdotorg/synapse -f docker/Dockerfile .
+```
+
+This will build an image with the tag `matrixdotorg/synapse`.
+
+Next, we build a new image with worker support based on `matrixdotorg/synapse:latest`. 
+Again, from the root of the repository:
+
+```sh
+docker build -t matrixdotorg/synapse-workers -f docker/Dockerfile-workers .
+```
+
+This will build an image with the tag` matrixdotorg/synapse-workers`.
+
+It's worth noting at this point that this image is fully functional, and 
+can be used for testing against locally. See instructions for using the container 
+under
+[Running the Dockerfile-worker image standalone](#running-the-dockerfile-worker-image-standalone)
+below.
+
+Finally, build the Synapse image for Complement, which is based on
+`matrixdotorg/synapse-workers`. You will need a local checkout of Complement. Change to
+the root of your Complement checkout and run:
+
+```sh
+docker build -t matrixdotorg/complement-synapse-workers -f dockerfiles/SynapseWorkers.Dockerfile dockerfiles
+```
+
+This will build an image with the tag `complement-synapse`, which can be handed to
+Complement for testing via the `COMPLEMENT_BASE_IMAGE` environment variable. Refer to
+[Complement's documentation](https://github.com/matrix-org/complement/#running) for
+how to run the tests, as well as the various available command line flags.
+
+## Running the Dockerfile-worker image standalone
+
+For manual testing of a multi-process Synapse instance in Docker,
+[Dockerfile-workers](Dockerfile-workers) is a Dockerfile that will produce an image
+bundling all necessary components together for a workerised homeserver instance.
+
+This includes any desired Synapse worker processes, a nginx to route traffic accordingly,
+a redis for worker communication and a supervisord instance to start up and monitor all
+processes. You will need to provide your own postgres container to connect to, and TLS 
+is not handled by the container.
+
+Once you've built the image using the above instructions, you can run it. Be sure 
+you've set up a volume according to the [usual Synapse docker instructions](README.md).
+Then run something along the lines of:
+
+```
+docker run -d --name synapse \
+    --mount type=volume,src=synapse-data,dst=/data \
+    -p 8008:8008 \
+    -e SYNAPSE_SERVER_NAME=my.matrix.host \
+    -e SYNAPSE_REPORT_STATS=no \
+    -e POSTGRES_HOST=postgres \
+    -e POSTGRES_USER=postgres \
+    -e POSTGRES_PASSWORD=somesecret \
+    -e SYNAPSE_WORKER_TYPES=synchrotron,media_repository,user_dir \
+    -e SYNAPSE_WORKERS_WRITE_LOGS_TO_DISK=1 \
+    matrixdotorg/synapse-workers
+```
+
+...substituting `POSTGRES*` variables for those that match a postgres host you have 
+available (usually a running postgres docker container).
+
+The `SYNAPSE_WORKER_TYPES` environment variable is a comma-separated list of workers to
+use when running the container. All possible worker names are defined by the keys of the
+`WORKERS_CONFIG` variable in [this script](configure_workers_and_start.py), which the
+Dockerfile makes use of to generate appropriate worker, nginx and supervisord config
+files.
+
+Sharding is supported for a subset of workers, in line with the
+[worker documentation](../docs/workers.md). To run multiple instances of a given worker
+type, simply specify the type multiple times in `SYNAPSE_WORKER_TYPES`
+(e.g `SYNAPSE_WORKER_TYPES=event_creator,event_creator...`).
+
+Otherwise, `SYNAPSE_WORKER_TYPES` can either be left empty or unset to spawn no workers
+(leaving only the main process). The container is configured to use redis-based worker
+mode.
+
+Logs for workers and the main process are logged to stdout and can be viewed with 
+standard `docker logs` tooling. Worker logs contain their worker name 
+after the timestamp.
+
+Setting `SYNAPSE_WORKERS_WRITE_LOGS_TO_DISK=1` will cause worker logs to be written to
+`<data_dir>/logs/<worker_name>.log`. Logs are kept for 1 week and rotate every day at 00:
+00, according to the container's clock. Logging for the main process must still be 
+configured by modifying the homeserver's log config in your Synapse data volume.

docker/README.md

@@ -11,7 +11,7 @@ is not supported by this image.
 
 ## Volumes
 
-By default, the image expects a single volume, located at ``/data``, that will hold:
+By default, the image expects a single volume, located at `/data`, that will hold:
 
 * configuration files;
 * uploaded media and thumbnails;
@@ -19,11 +19,11 @@ By default, the image expects a single volume, located at ``/data``, that will h
 * the appservices configuration.
 
 You are free to use separate volumes depending on storage endpoints at your
-disposal. For instance, ``/data/media`` could be stored on a large but low
+disposal. For instance, `/data/media` could be stored on a large but low
 performance hdd storage while other files could be stored on high performance
 endpoints.
 
-In order to setup an application service, simply create an ``appservices``
+In order to setup an application service, simply create an `appservices`
 directory in the data volume and write the application service Yaml
 configuration file there. Multiple application services are supported.
 
@@ -56,6 +56,8 @@ The following environment variables are supported in `generate` mode:
 * `SYNAPSE_SERVER_NAME` (mandatory): the server public hostname.
 * `SYNAPSE_REPORT_STATS` (mandatory, `yes` or `no`): whether to enable
   anonymous statistics reporting.
+* `SYNAPSE_HTTP_PORT`: the port Synapse should listen on for http traffic.
+      Defaults to `8008`.
 * `SYNAPSE_CONFIG_DIR`: where additional config files (such as the log config
   and event signing key) will be stored. Defaults to `/data`.
 * `SYNAPSE_CONFIG_PATH`: path to the file to be generated. Defaults to
@@ -76,6 +78,8 @@ docker run -d --name synapse \
     matrixdotorg/synapse:latest
 ```
 
+(assuming 8008 is the port Synapse is configured to listen on for http traffic.)
+
 You can then check that it has started correctly with:
 
 ```
@@ -211,4 +215,4 @@ healthcheck:
 ## Using jemalloc
 
 Jemalloc is embedded in the image and will be used instead of the default allocator.
-You can read about jemalloc by reading the Synapse [README](../README.md)
+You can read about jemalloc by reading the Synapse [README](../README.md).

docker/conf-workers/nginx.conf.j2

@@ -0,0 +1,27 @@
+# This file contains the base config for the reverse proxy, as part of ../Dockerfile-workers.
+# configure_workers_and_start.py uses and amends to this file depending on the workers
+# that have been selected.
+
+{{ upstream_directives }}
+
+server {
+    # Listen on an unoccupied port number
+    listen 8008;
+    listen [::]:8008;
+
+    server_name localhost;
+
+    # Nginx by default only allows file uploads up to 1M in size
+    # Increase client_max_body_size to match max_upload_size defined in homeserver.yaml
+    client_max_body_size 100M;
+
+{{ worker_locations }}
+
+    # Send all other traffic to the main process
+    location ~* ^(\\/_matrix|\\/_synapse) {
+        proxy_pass http://localhost:8080;
+        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

@@ -0,0 +1,9 @@
+# This file contains the base for the shared homeserver config file between Synapse workers,
+# as part of ./Dockerfile-workers.
+# configure_workers_and_start.py uses and amends to this file depending on the workers
+# that have been selected.
+
+redis:
+    enabled: true
+
+{{ shared_worker_config }}

docker/conf-workers/supervisord.conf.j2

@@ -0,0 +1,41 @@
+# This file contains the base config for supervisord, as part of ../Dockerfile-workers.
+# configure_workers_and_start.py uses and amends to this file depending on the workers
+# that have been selected.
+[supervisord]
+nodaemon=true
+user=root
+
+[program:nginx]
+command=/usr/sbin/nginx -g "daemon off;"
+priority=500
+stdout_logfile=/dev/stdout
+stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
+username=www-data
+autorestart=true
+
+[program:redis]
+command=/usr/bin/redis-server /etc/redis/redis.conf --daemonize no
+priority=1
+stdout_logfile=/dev/stdout
+stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
+username=redis
+autorestart=true
+
+[program:synapse_main]
+command=/usr/local/bin/python -m synapse.app.homeserver --config-path="{{ main_config_path }}" --config-path=/conf/workers/shared.yaml
+priority=10
+# Log startup failures to supervisord's stdout/err
+# Regular synapse logs will still go in the configured data directory
+stdout_logfile=/dev/stdout
+stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
+autorestart=unexpected
+exitcodes=0
+
+# Additional process blocks
+{{ worker_config }}

docker/conf-workers/worker.yaml.j2

@@ -0,0 +1,26 @@
+# This is a configuration template for a single worker instance, and is
+# used by Dockerfile-workers.
+# Values will be change depending on whichever workers are selected when
+# running that image.
+
+worker_app: "{{ app }}"
+worker_name: "{{ name }}"
+
+# The replication listener on the main synapse process.
+worker_replication_host: 127.0.0.1
+worker_replication_http_port: 9093
+
+worker_listeners:
+  - type: http
+    port: {{ port }}
+{% if listener_resources %}
+    resources:
+      - names:
+{%- for resource in listener_resources %}
+        - {{ resource }}
+{%- endfor %}
+{% endif %}
+
+worker_log_config: {{ worker_log_config_filepath }}
+
+{{ worker_extra_conf }}

docker/conf/homeserver.yaml

@@ -40,7 +40,9 @@ listeners:
         compress: false
   {% endif %}
 
-  - port: 8008
+  # Allow configuring in case we want to reverse proxy 8008
+  # using another process in the same container
+  - port: {{ SYNAPSE_HTTP_PORT or 8008 }}
     tls: false
     bind_addresses: ['::']
     type: http

docker/conf/log.config

@@ -2,9 +2,34 @@ version: 1
 
 formatters:
   precise:
-   format: '%(asctime)s - %(name)s - %(lineno)d - %(levelname)s - %(request)s - %(message)s'
+{% if worker_name %}
+    format: '%(asctime)s - worker:{{ worker_name }} - %(name)s - %(lineno)d - %(levelname)s - %(request)s - %(message)s'
+{% else %}
+    format: '%(asctime)s - %(name)s - %(lineno)d - %(levelname)s - %(request)s - %(message)s'
+{% endif %}
 
 handlers:
+  file:
+    class: logging.handlers.TimedRotatingFileHandler
+    formatter: precise
+    filename: {{ LOG_FILE_PATH or "homeserver.log" }}
+    when: "midnight"
+    backupCount: 6  # Does not include the current log file.
+    encoding: utf8
+
+  # Default to buffering writes to log file for efficiency. This means that
+  # there will be a delay for INFO/DEBUG logs to get written, but WARNING/ERROR
+  # logs will still be flushed immediately.
+  buffer:
+    class: logging.handlers.MemoryHandler
+    target: file
+    # The capacity is the number of log lines that are buffered before
+    # being written to disk. Increasing this will lead to better
+    # performance, at the expensive of it taking longer for log lines to
+    # be written to disk.
+    capacity: 10
+    flushLevel: 30  # Flush for WARNING logs as well
+
   console:
     class: logging.StreamHandler
     formatter: precise
@@ -17,6 +42,11 @@ loggers:
 
 root:
     level: {{ SYNAPSE_LOG_LEVEL or "INFO" }}
+
+{% if LOG_FILE_PATH %}
+    handlers: [console, buffer]
+{% else %}
     handlers: [console]
+{% endif %}
 
 disable_existing_loggers: false

docker/configure_workers_and_start.py

@@ -0,0 +1,558 @@
+#!/usr/bin/env python
+# Copyright 2021 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This script reads environment variables and generates a shared Synapse worker,
+# nginx and supervisord configs depending on the workers requested.
+#
+# The environment variables it reads are:
+#   * SYNAPSE_SERVER_NAME: The desired server_name of the homeserver.
+#   * SYNAPSE_REPORT_STATS: Whether to report stats.
+#   * SYNAPSE_WORKER_TYPES: A comma separated list of worker names as specified in WORKER_CONFIG
+#         below. Leave empty for no workers, or set to '*' for all possible workers.
+#
+# NOTE: According to Complement's ENTRYPOINT expectations for a homeserver image (as defined
+# in the project's README), this script may be run multiple times, and functionality should
+# continue to work if so.
+
+import os
+import subprocess
+import sys
+
+import jinja2
+import yaml
+
+MAIN_PROCESS_HTTP_LISTENER_PORT = 8080
+
+
+WORKERS_CONFIG = {
+    "pusher": {
+        "app": "synapse.app.pusher",
+        "listener_resources": [],
+        "endpoint_patterns": [],
+        "shared_extra_conf": {"start_pushers": False},
+        "worker_extra_conf": "",
+    },
+    "user_dir": {
+        "app": "synapse.app.user_dir",
+        "listener_resources": ["client"],
+        "endpoint_patterns": [
+            "^/_matrix/client/(api/v1|r0|unstable)/user_directory/search$"
+        ],
+        "shared_extra_conf": {"update_user_directory": False},
+        "worker_extra_conf": "",
+    },
+    "media_repository": {
+        "app": "synapse.app.media_repository",
+        "listener_resources": ["media"],
+        "endpoint_patterns": [
+            "^/_matrix/media/",
+            "^/_synapse/admin/v1/purge_media_cache$",
+            "^/_synapse/admin/v1/room/.*/media.*$",
+            "^/_synapse/admin/v1/user/.*/media.*$",
+            "^/_synapse/admin/v1/media/.*$",
+            "^/_synapse/admin/v1/quarantine_media/.*$",
+        ],
+        "shared_extra_conf": {"enable_media_repo": False},
+        "worker_extra_conf": "enable_media_repo: true",
+    },
+    "appservice": {
+        "app": "synapse.app.appservice",
+        "listener_resources": [],
+        "endpoint_patterns": [],
+        "shared_extra_conf": {"notify_appservices": False},
+        "worker_extra_conf": "",
+    },
+    "federation_sender": {
+        "app": "synapse.app.federation_sender",
+        "listener_resources": [],
+        "endpoint_patterns": [],
+        "shared_extra_conf": {"send_federation": False},
+        "worker_extra_conf": "",
+    },
+    "synchrotron": {
+        "app": "synapse.app.generic_worker",
+        "listener_resources": ["client"],
+        "endpoint_patterns": [
+            "^/_matrix/client/(v2_alpha|r0)/sync$",
+            "^/_matrix/client/(api/v1|v2_alpha|r0)/events$",
+            "^/_matrix/client/(api/v1|r0)/initialSync$",
+            "^/_matrix/client/(api/v1|r0)/rooms/[^/]+/initialSync$",
+        ],
+        "shared_extra_conf": {},
+        "worker_extra_conf": "",
+    },
+    "federation_reader": {
+        "app": "synapse.app.generic_worker",
+        "listener_resources": ["federation"],
+        "endpoint_patterns": [
+            "^/_matrix/federation/(v1|v2)/event/",
+            "^/_matrix/federation/(v1|v2)/state/",
+            "^/_matrix/federation/(v1|v2)/state_ids/",
+            "^/_matrix/federation/(v1|v2)/backfill/",
+            "^/_matrix/federation/(v1|v2)/get_missing_events/",
+            "^/_matrix/federation/(v1|v2)/publicRooms",
+            "^/_matrix/federation/(v1|v2)/query/",
+            "^/_matrix/federation/(v1|v2)/make_join/",
+            "^/_matrix/federation/(v1|v2)/make_leave/",
+            "^/_matrix/federation/(v1|v2)/send_join/",
+            "^/_matrix/federation/(v1|v2)/send_leave/",
+            "^/_matrix/federation/(v1|v2)/invite/",
+            "^/_matrix/federation/(v1|v2)/query_auth/",
+            "^/_matrix/federation/(v1|v2)/event_auth/",
+            "^/_matrix/federation/(v1|v2)/exchange_third_party_invite/",
+            "^/_matrix/federation/(v1|v2)/user/devices/",
+            "^/_matrix/federation/(v1|v2)/get_groups_publicised$",
+            "^/_matrix/key/v2/query",
+        ],
+        "shared_extra_conf": {},
+        "worker_extra_conf": "",
+    },
+    "federation_inbound": {
+        "app": "synapse.app.generic_worker",
+        "listener_resources": ["federation"],
+        "endpoint_patterns": ["/_matrix/federation/(v1|v2)/send/"],
+        "shared_extra_conf": {},
+        "worker_extra_conf": "",
+    },
+    "event_persister": {
+        "app": "synapse.app.generic_worker",
+        "listener_resources": ["replication"],
+        "endpoint_patterns": [],
+        "shared_extra_conf": {},
+        "worker_extra_conf": "",
+    },
+    "background_worker": {
+        "app": "synapse.app.generic_worker",
+        "listener_resources": [],
+        "endpoint_patterns": [],
+        # This worker cannot be sharded. Therefore there should only ever be one background
+        # worker, and it should be named background_worker1
+        "shared_extra_conf": {"run_background_tasks_on": "background_worker1"},
+        "worker_extra_conf": "",
+    },
+    "event_creator": {
+        "app": "synapse.app.generic_worker",
+        "listener_resources": ["client"],
+        "endpoint_patterns": [
+            "^/_matrix/client/(api/v1|r0|unstable)/rooms/.*/redact",
+            "^/_matrix/client/(api/v1|r0|unstable)/rooms/.*/send",
+            "^/_matrix/client/(api/v1|r0|unstable)/rooms/.*/(join|invite|leave|ban|unban|kick)$",
+            "^/_matrix/client/(api/v1|r0|unstable)/join/",
+            "^/_matrix/client/(api/v1|r0|unstable)/profile/",
+        ],
+        "shared_extra_conf": {},
+        "worker_extra_conf": "",
+    },
+    "frontend_proxy": {
+        "app": "synapse.app.frontend_proxy",
+        "listener_resources": ["client", "replication"],
+        "endpoint_patterns": ["^/_matrix/client/(api/v1|r0|unstable)/keys/upload"],
+        "shared_extra_conf": {},
+        "worker_extra_conf": (
+            "worker_main_http_uri: http://127.0.0.1:%d"
+            % (MAIN_PROCESS_HTTP_LISTENER_PORT,),
+        ),
+    },
+}
+
+# Templates for sections that may be inserted multiple times in config files
+SUPERVISORD_PROCESS_CONFIG_BLOCK = """
+[program:synapse_{name}]
+command=/usr/local/bin/python -m {app} \
+    --config-path="{config_path}" \
+    --config-path=/conf/workers/shared.yaml \
+    --config-path=/conf/workers/{name}.yaml
+autorestart=unexpected
+priority=500
+exitcodes=0
+stdout_logfile=/dev/stdout
+stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
+"""
+
+NGINX_LOCATION_CONFIG_BLOCK = """
+    location ~* {endpoint} {
+        proxy_pass {upstream};
+        proxy_set_header X-Forwarded-For $remote_addr;
+        proxy_set_header X-Forwarded-Proto $scheme;
+        proxy_set_header Host $host;
+    }
+"""
+
+NGINX_UPSTREAM_CONFIG_BLOCK = """
+upstream {upstream_worker_type} {
+{body}
+}
+"""
+
+
+# Utility functions
+def log(txt: str):
+    """Log something to the stdout.
+
+    Args:
+        txt: The text to log.
+    """
+    print(txt)
+
+
+def error(txt: str):
+    """Log something and exit with an error code.
+
+    Args:
+        txt: The text to log in error.
+    """
+    log(txt)
+    sys.exit(2)
+
+
+def convert(src: str, dst: str, **template_vars):
+    """Generate a file from a template
+
+    Args:
+        src: Path to the input file.
+        dst: Path to write to.
+        template_vars: The arguments to replace placeholder variables in the template with.
+    """
+    # Read the template file
+    with open(src) as infile:
+        template = infile.read()
+
+    # Generate a string from the template. We disable autoescape to prevent template
+    # variables from being escaped.
+    rendered = jinja2.Template(template, autoescape=False).render(**template_vars)
+
+    # Write the generated contents to a file
+    #
+    # We use append mode in case the files have already been written to by something else
+    # (for instance, as part of the instructions in a dockerfile).
+    with open(dst, "a") as outfile:
+        # In case the existing file doesn't end with a newline
+        outfile.write("\n")
+
+        outfile.write(rendered)
+
+
+def add_sharding_to_shared_config(
+    shared_config: dict,
+    worker_type: str,
+    worker_name: str,
+    worker_port: int,
+) -> None:
+    """Given a dictionary representing a config file shared across all workers,
+    append sharded worker information to it for the current worker_type instance.
+
+    Args:
+        shared_config: The config dict that all worker instances share (after being converted to YAML)
+        worker_type: The type of worker (one of those defined in WORKERS_CONFIG).
+        worker_name: The name of the worker instance.
+        worker_port: The HTTP replication port that the worker instance is listening on.
+    """
+    # The instance_map config field marks the workers that write to various replication streams
+    instance_map = shared_config.setdefault("instance_map", {})
+
+    # Worker-type specific sharding config
+    if worker_type == "pusher":
+        shared_config.setdefault("pusher_instances", []).append(worker_name)
+
+    elif worker_type == "federation_sender":
+        shared_config.setdefault("federation_sender_instances", []).append(worker_name)
+
+    elif worker_type == "event_persister":
+        # Event persisters write to the events stream, so we need to update
+        # the list of event stream writers
+        shared_config.setdefault("stream_writers", {}).setdefault("events", []).append(
+            worker_name
+        )
+
+        # Map of stream writer instance names to host/ports combos
+        instance_map[worker_name] = {
+            "host": "localhost",
+            "port": worker_port,
+        }
+
+    elif worker_type == "media_repository":
+        # The first configured media worker will run the media background jobs
+        shared_config.setdefault("media_instance_running_background_jobs", worker_name)
+
+
+def generate_base_homeserver_config():
+    """Starts Synapse and generates a basic homeserver config, which will later be
+    modified for worker support.
+
+    Raises: CalledProcessError if calling start.py returned a non-zero exit code.
+    """
+    # start.py already does this for us, so just call that.
+    # note that this script is copied in in the official, monolith dockerfile
+    os.environ["SYNAPSE_HTTP_PORT"] = str(MAIN_PROCESS_HTTP_LISTENER_PORT)
+    subprocess.check_output(["/usr/local/bin/python", "/start.py", "migrate_config"])
+
+
+def generate_worker_files(environ, config_path: str, data_dir: str):
+    """Read the desired list of workers from environment variables and generate
+    shared homeserver, nginx and supervisord configs.
+
+    Args:
+        environ: _Environ[str]
+        config_path: Where to output the generated Synapse main worker config file.
+        data_dir: The location of the synapse data directory. Where log and
+            user-facing config files live.
+    """
+    # Note that yaml cares about indentation, so care should be taken to insert lines
+    # into files at the correct indentation below.
+
+    # shared_config is the contents of a Synapse config file that will be shared amongst
+    # the main Synapse process as well as all workers.
+    # It is intended mainly for disabling functionality when certain workers are spun up,
+    # and adding a replication listener.
+
+    # 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.
+    listeners = [
+        {
+            "port": 9093,
+            "bind_address": "127.0.0.1",
+            "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")
+        if original_listeners:
+            listeners += original_listeners
+
+    # The shared homeserver config. The contents of which will be inserted into the
+    # base shared worker jinja2 template.
+    #
+    # This config file will be passed to all workers, included Synapse's main process.
+    shared_config = {"listeners": listeners}
+
+    # The supervisord config. The contents of which will be inserted into the
+    # base supervisord jinja2 template.
+    #
+    # Supervisord will be in charge of running everything, from redis to nginx to Synapse
+    # and all of its worker processes. Load the config template, which defines a few
+    # services that are necessary to run.
+    supervisord_config = ""
+
+    # Upstreams for load-balancing purposes. This dict takes the form of a worker type to the
+    # ports of each worker. For example:
+    # {
+    #   worker_type: {1234, 1235, ...}}
+    # }
+    # and will be used to construct 'upstream' nginx directives.
+    nginx_upstreams = {}
+
+    # A map of: {"endpoint": "upstream"}, where "upstream" is a str representing what will be
+    # placed after the proxy_pass directive. The main benefit to representing this data as a
+    # dict over a str is that we can easily deduplicate endpoints across multiple instances
+    # of the same worker.
+    #
+    # An nginx site config that will be amended to depending on the workers that are
+    # spun up. To be placed in /etc/nginx/conf.d.
+    nginx_locations = {}
+
+    # Read the desired worker configuration from the environment
+    worker_types = environ.get("SYNAPSE_WORKER_TYPES")
+    if worker_types is None:
+        # No workers, just the main process
+        worker_types = []
+    else:
+        # Split type names by comma
+        worker_types = worker_types.split(",")
+
+    # Create the worker configuration directory if it doesn't already exist
+    os.makedirs("/conf/workers", exist_ok=True)
+
+    # Start worker ports from this arbitrary port
+    worker_port = 18009
+
+    # A counter of worker_type -> int. Used for determining the name for a given
+    # worker type when generating its config file, as each worker's name is just
+    # worker_type + instance #
+    worker_type_counter = {}
+
+    # For each worker type specified by the user, create config values
+    for worker_type in worker_types:
+        worker_type = worker_type.strip()
+
+        worker_config = WORKERS_CONFIG.get(worker_type)
+        if worker_config:
+            worker_config = worker_config.copy()
+        else:
+            log(worker_type + " is an unknown worker type! It will be ignored")
+            continue
+
+        new_worker_count = worker_type_counter.setdefault(worker_type, 0) + 1
+        worker_type_counter[worker_type] = new_worker_count
+
+        # Name workers by their type concatenated with an incrementing number
+        # e.g. federation_reader1
+        worker_name = worker_type + str(new_worker_count)
+        worker_config.update(
+            {"name": worker_name, "port": worker_port, "config_path": config_path}
+        )
+
+        # Update the shared config with any worker-type specific options
+        shared_config.update(worker_config["shared_extra_conf"])
+
+        # Check if more than one instance of this worker type has been specified
+        worker_type_total_count = worker_types.count(worker_type)
+        if worker_type_total_count > 1:
+            # Update the shared config with sharding-related options if necessary
+            add_sharding_to_shared_config(
+                shared_config, worker_type, worker_name, worker_port
+            )
+
+        # Enable the worker in supervisord
+        supervisord_config += SUPERVISORD_PROCESS_CONFIG_BLOCK.format_map(worker_config)
+
+        # Add nginx location blocks for this worker's endpoints (if any are defined)
+        for pattern in worker_config["endpoint_patterns"]:
+            # Determine whether we need to load-balance this worker
+            if worker_type_total_count > 1:
+                # Create or add to a load-balanced upstream for this worker
+                nginx_upstreams.setdefault(worker_type, set()).add(worker_port)
+
+                # Upstreams are named after the worker_type
+                upstream = "http://" + worker_type
+            else:
+                upstream = "http://localhost:%d" % (worker_port,)
+
+            # Note that this endpoint should proxy to this upstream
+            nginx_locations[pattern] = upstream
+
+        # Write out the worker's logging config file
+
+        # Check whether we should write worker logs to disk, in addition to the console
+        extra_log_template_args = {}
+        if environ.get("SYNAPSE_WORKERS_WRITE_LOGS_TO_DISK"):
+            extra_log_template_args["LOG_FILE_PATH"] = "{dir}/logs/{name}.log".format(
+                dir=data_dir, name=worker_name
+            )
+
+        # Render and write the file
+        log_config_filepath = "/conf/workers/{name}.log.config".format(name=worker_name)
+        convert(
+            "/conf/log.config",
+            log_config_filepath,
+            worker_name=worker_name,
+            **extra_log_template_args,
+        )
+
+        # Then a worker config file
+        convert(
+            "/conf/worker.yaml.j2",
+            "/conf/workers/{name}.yaml".format(name=worker_name),
+            **worker_config,
+            worker_log_config_filepath=log_config_filepath,
+        )
+
+        worker_port += 1
+
+    # Build the nginx location config blocks
+    nginx_location_config = ""
+    for endpoint, upstream in nginx_locations.items():
+        nginx_location_config += NGINX_LOCATION_CONFIG_BLOCK.format(
+            endpoint=endpoint,
+            upstream=upstream,
+        )
+
+    # Determine the load-balancing upstreams to configure
+    nginx_upstream_config = ""
+    for upstream_worker_type, upstream_worker_ports in nginx_upstreams.items():
+        body = ""
+        for port in upstream_worker_ports:
+            body += "    server localhost:%d;\n" % (port,)
+
+        # Add to the list of configured upstreams
+        nginx_upstream_config += NGINX_UPSTREAM_CONFIG_BLOCK.format(
+            upstream_worker_type=upstream_worker_type,
+            body=body,
+        )
+
+    # Finally, we'll write out the config files.
+
+    # Shared homeserver config
+    convert(
+        "/conf/shared.yaml.j2",
+        "/conf/workers/shared.yaml",
+        shared_worker_config=yaml.dump(shared_config),
+    )
+
+    # Nginx config
+    convert(
+        "/conf/nginx.conf.j2",
+        "/etc/nginx/conf.d/matrix-synapse.conf",
+        worker_locations=nginx_location_config,
+        upstream_directives=nginx_upstream_config,
+    )
+
+    # Supervisord config
+    convert(
+        "/conf/supervisord.conf.j2",
+        "/etc/supervisor/conf.d/supervisord.conf",
+        main_config_path=config_path,
+        worker_config=supervisord_config,
+    )
+
+    # Ensure the logging directory exists
+    log_dir = data_dir + "/logs"
+    if not os.path.exists(log_dir):
+        os.mkdir(log_dir)
+
+
+def start_supervisord():
+    """Starts up supervisord which then starts and monitors all other necessary processes
+
+    Raises: CalledProcessError if calling start.py return a non-zero exit code.
+    """
+    subprocess.run(["/usr/bin/supervisord"], stdin=subprocess.PIPE)
+
+
+def main(args, environ):
+    config_dir = environ.get("SYNAPSE_CONFIG_DIR", "/data")
+    config_path = environ.get("SYNAPSE_CONFIG_PATH", config_dir + "/homeserver.yaml")
+    data_dir = environ.get("SYNAPSE_DATA_DIR", "/data")
+
+    # override SYNAPSE_NO_TLS, we don't support TLS in worker mode,
+    # this needs to be handled by a frontend proxy
+    environ["SYNAPSE_NO_TLS"] = "yes"
+
+    # Generate the base homeserver config if one does not yet exist
+    if not os.path.exists(config_path):
+        log("Generating base homeserver config")
+        generate_base_homeserver_config()
+
+    # This script may be run multiple times (mostly by Complement, see note at top of file).
+    # Don't re-configure workers in this instance.
+    mark_filepath = "/conf/workers_have_been_configured"
+    if not os.path.exists(mark_filepath):
+        # Always regenerate all other config files
+        generate_worker_files(environ, config_path, data_dir)
+
+        # Mark workers as being configured
+        with open(mark_filepath, "w") as f:
+            f.write("")
+
+    # Start supervisord, which will start Synapse, all of the configured worker
+    # processes, redis, nginx etc. according to the config we created above.
+    start_supervisord()
+
+
+if __name__ == "__main__":
+    main(sys.argv, os.environ)