Merge branch 'chrusher-database-replication'

INSTALL.md

@@ -32,9 +32,9 @@ Alternatively if you have `git` installed you can clone the git repository:
 
 You can edit the `docker-compose.jsonnet` file to set the configuration options. Important options include:
 
-* `channel`, the Twitch channel to capture from
-* `segments_path`, the local path to save segments to
-* `db_args`, the arguments for connecting to the wubloader database
+* `channel`, the Twitch channel to capture from.
+* `segments_path`, the local path to save segments to.
+* `db_args`, the arguments for connecting to the wubloader database. You will likely need to update the `user`, `password` and `host` to match the database node that you are connecting to.
 * `ports`, the ports to expose each service on. Only the `nginx` port (default on port 80) needs to be externally accessible for a non-database node as all the other services are routed through `nginx`.
 
 To generate the `docker-compose.yml` file used by `docker-compose`, run `generate-docker-compose`
@@ -43,7 +43,7 @@ To generate the `docker-compose.yml` file used by `docker-compose`, run `generat
   
 After making any changes to `docker-compose.jsonnet`, you will need to rerun `generate-docker-compose`.
 
-By default the `downloader`, `restreamer`, `backfiller`, `cutter`, `thrimshim` and `nginx` services of the wubloader will be run. To change which services are run edit the `enabled` object in `docker-compose.jsonnet`. A complete wubloader set up also requires one and only one `database` service (though having a backup database is a good idea) and one and only one `sheetsync` service. TODO: explain how to setup database 
+By default the `downloader`, `restreamer`, `backfiller`, `cutter`, `thrimshim` and `nginx` services of the wubloader will be run. To change which services are run edit the `enabled` object in `docker-compose.jsonnet`. A complete wubloader set up also requires one and only one `database` service (though having a backup database is a good idea) and one and only one `sheetsync` service.
 
 ## Running the wubloader
 
@@ -54,5 +54,48 @@ To start the wubloader, simply run
 To stop the wubloader and clean up, simply run
 
   `docker-compose down`
-  
-To backfill from a node, the other nodes need to know about it. The best way to do this is to add the node to the database's nodes table.
+
+## Database setup
+
+When setting up a database node, a number of database specific options can be set.
+
+* `database_path`, the local path to save the database to. If this directory is empty then the database setups scripts will be run to create a new database. Otherwise, the database container will load the database stored in this folder.
+* `db_args.user`, `db_args.password`, the username and password for the database user that the rest of the wubloader will connect to.
+* `db_super_user`, `super_password`, the username and password for the database superuser that is only accessible from the local machine. 
+* `db_replication_user`, `db_replication_password`, the username and password for the database user other nodes can connect as to replicate the database. If `db_replication_user` is an empty string, remote replication will be disabled.
+* `db_standby`, If true this database node will replicate the database node given by `db_args.host`. 
+
+It is recommended that the passwords be changed from the defaults in production.
+A database node needs to expose its database on a port. By default this is `5432` but the port exposed to the outside can be changed in the `ports` object.
+
+The `events` table will be automatically populated by the `sheetsync`. If creating a new database, the startup script will attempt to populate the `nodes` and `editors` tables from the `nodes.csv` and `editors.csv` files in `segments_path` directory. The expected format for these files is:
+
+```
+nodes.csv
+
+name,url,backfill_from
+example,http://example.com,TRUE
+```
+
+```
+editors.csv
+
+name,email
+example,example@gmail.com
+```
+
+Alternatively, nodes can be added manually to the database's `nodes` table:
+
+`wubloader=> INSERT INTO nodes (name, url) VALUES ('example_name', 'http://example.com');`
+
+and editors to the database's `editors` table:
+
+`wubloader=> INSERT INTO editors (name, email) VALUES ('example', 'example@gmail.com');`
+
+### Promoting the standby server
+
+To promote the standby server to primary touch the trigger file in the docker container:
+
+`docker exec wubloader_postgres_1 touch /tmp/touch_to_promote_to_master`
+
+Be careful to prevent the original primary from restarting as another primary.

build

@@ -8,7 +8,7 @@ set -eu
 # Pass PUSH=true to also push the resulting images, or PUSH=latest to push them as :latest tag
 
 # The different images we can build
-COMPONENTS=(downloader restreamer backfiller thrimshim cutter sheetsync nginx)
+COMPONENTS=(downloader restreamer backfiller thrimshim cutter sheetsync nginx postgres)
 
 # Define push if not already defined
 PUSH=${PUSH:-}

cutter/cutter/main.py

@@ -495,7 +495,16 @@ def main(dbconnect, youtube_creds_file, name=None, base_dir=".", metrics_port=80
 	# We have two independent jobs to do - to perform cut jobs (cutter),
 	# and to check the status of transcoding videos to see if they're done (transcode checker).
 	# We want to error if either errors, and shut down if either exits.
-	dbmanager = DBManager(dsn=dbconnect)
+	dbmanager = None
+	stopping = gevent.event.Event()
+	while dbmanager is None:
+		try:
+			dbmanager = DBManager(dsn=dbconnect)
+		except Exception:
+			delay = common.jitter(10)
+			logging.info('Cannot connect to database. Retrying in {:.0f} s'.format(delay))
+			stop.wait(delay)
+
 	youtube_creds = json.load(open(youtube_creds_file))
 	youtube = Youtube(
 		client_id=youtube_creds['client_id'],

db_scripts/db_dump

@@ -0,0 +1,16 @@
+#! /bin/bash
+NAME=${1:-postgres}
+BUCKET=${2:-wubloader-db}
+CONTAINER=$(docker ps --format "{{.Names}}" | grep "$NAME")
+if [ -z "$CONTAINER" ]; then
+	echo "Container not found"
+	exit 1
+fi
+if [ "$(wc -l <<<"$CONTAINER")" -ne 1 ]; then
+	echo "Multiple containers found"
+	exit 1
+fi
+
+FILE="wubloader-$(date +%Y-%m-%dT%H:%M:%S).sql"
+echo "Dumping $CONTAINER to $FILE"
+docker exec $CONTAINER pg_dump wubloader -U postgres | aws s3 cp - "s3://$BUCKET/$FILE"

db_scripts/db_restore

@@ -0,0 +1,28 @@
+#! /bin/bash
+
+if [ -z "$1" ]; then
+	echo "No SQL script file"
+	echo "USAGE: $0 SQL_SCRIPT"
+	exit 1
+fi
+
+NAME=${2:-postgres}
+CONTAINER=$(docker ps --format "{{.Names}}" | grep "$NAME")
+if [ -z "$CONTAINER" ]; then
+	echo "Container not found"
+	exit 1
+fi
+if [ "$(wc -l <<<"$CONTAINER")" -ne 1 ]; then
+	echo "Multiple containers found"
+	exit 1
+fi
+
+docker cp "./$1" "$CONTAINER:/tmp.sql"
+
+# this will fail if there are active sessions by users other than 'postgres'
+# make sure all wubloader components are disconnected
+docker exec "$CONTAINER" dropdb wubloader -U postgres --if-exists
+docker exec "$CONTAINER" createdb wubloader -U postgres
+# this assumes that the vst role is in the postgres database
+docker exec "$CONTAINER" psql -d wubloader -f tmp.sql -U postgres
+docker exec "$CONTAINER" rm tmp.sql

docker-compose.jsonnet

@@ -7,7 +7,7 @@
   // Change these to configure the services.
 
   // Image tag (application version) to use.
-  // Note: "latest" is not reccomended in production, as you can't be sure what version
+  // Note: "latest" is not recommended in production, as you can't be sure what version
   // you're actually running, and must manually re-pull to get an updated copy.
   image_tag:: "latest",
 
@@ -33,8 +33,11 @@
   // On OSX you need to change this to /private/var/lib/wubloader
   segments_path:: "/var/lib/wubloader/",
 
-  // Local path to save database to. Full path must already exist. Cannot contain ':'.
-  database_path:: "/var/lib/wubloader_postgres/",
+  // Local path to save database to. Full path must already exist. Cannot
+  // contain ':'. If this directory is non-empty, the database will start with
+  // the database in this directory and not run the setup scripts to create a 
+  // new database.
+  database_path:: "/private/var/lib/wubloader_postgres/",
 
   // The host's port to expose each service on.
   // Only nginx (and postgres if that is being deployed) needs to be externally accessible - the other non-database ports are routed through nginx.
@@ -53,9 +56,9 @@
   // You can exec into the container and telnet to this port to get a python shell.
   backdoor_port:: 1234,
 
-  // Other nodes to backfill from. You should not include the local node.
+  // Other nodes to always backfill from. You should not include the local node.
+  // If you are using the database to find peers, you should leave this empty.
   peers:: [
-    "http://wubloader.codegunner.com/"
   ],
 
   authentication:: true, // set to false to disable auth in thrimshim
@@ -63,13 +66,20 @@
   // Connection args for the database.
   // If database is defined in this config, host and port should be postgres:5432.
   db_args:: {
-    user: "postgres",
-    password: "postgres",
+    user: "vst",
+    password: "dbfh2019", // don't use default in production. Must not contain ' or \ as these are not escaped.
     host: "postgres",
     port: 5432,
     dbname: "wubloader",
   },
 
+  // Other database arguments
+  db_super_user:: "postgres", // only accessible from localhost
+  db_super_password:: "postgres", // Must not contain ' or \ as these are not escaped.
+  db_replication_user:: "replicate", // if empty, don't allow replication
+  db_replication_password:: "standby", // don't use default in production. Must not contain ' or \ as these are not escaped.
+  db_standby:: false, // set to true to have this database replicate another server
+
   // Path to a JSON file containing google credentials as keys
   // 'client_id', 'client_secret' and 'refresh_token'.
   google_creds:: "./google_creds.json",
@@ -139,7 +149,7 @@
         "--qualities", std.join(",", $.qualities),
         "--static-nodes", std.join(",", $.peers),
         "--backdoor-port", std.toString($.backdoor_port),
-		"--node-database", $.db_connect,
+        "--node-database", $.db_connect,
       ],
       // Mount the segments directory at /mnt
       volumes: ["%s:/mnt" % $.segments_path],
@@ -233,14 +243,22 @@
     },
 
     [if $.enabled.postgres then "postgres"]: {
-      image: "postgres:latest",
+      image: "quay.io/ekimekim/wubloader-postgres:%s" % $.image_tag,
       restart: "on-failure",
       [if "postgres" in $.ports then "ports"]: ["%s:5432" % $.ports.postgres],
       environment: {
-        POSTGRES_USER: $.db_args.user,
-        POSTGRES_PASSWORD: $.db_args.password,
+        POSTGRES_USER: $.db_super_user,
+        POSTGRES_PASSWORD: $.db_super_password,
         POSTGRES_DB: $.db_args.dbname,
+        PGDATA: "/mnt/database",
+        WUBLOADER_USER: $.db_args.user,
+        WUBLOADER_PASSWORD: $.db_args.password,
+        REPLICATION_USER: $.db_replication_user,
+        REPLICATION_PASSWORD: $.db_replication_password,
+        MASTER_NODE: $.db_args.host,
       },
+      volumes: ["%s:/mnt/database" % $.database_path, "%s:/mnt/wubloader" % $.segments_path],
+      [if $.db_standby then "command"]: ["/standby_setup.sh"],
     },
 
   },

postgres/Dockerfile

@@ -0,0 +1,5 @@
+FROM postgres:latest
+COPY postgres/setup.sh /docker-entrypoint-initdb.d/setup.sh
+RUN chmod 0666 /docker-entrypoint-initdb.d/setup.sh
+COPY postgres/standby_setup.sh /standby_setup.sh
+RUN chmod 0700 /standby_setup.sh

postgres/setup.sh

@@ -0,0 +1,59 @@
+#! /bin/bash
+
+set -e
+
+# only allow the $WUBLOADER_USER to connect remotely rather than all users
+sed -i "/host all all all/d" "$PGDATA/pg_hba.conf"
+echo "host all $WUBLOADER_USER all md5" >> "$PGDATA/pg_hba.conf"
+
+echo "Creating $WUBLOADER_USER"
+psql -v ON_ERROR_STOP=1 -U postgres <<-EOSQL
+
+CREATE USER $WUBLOADER_USER LOGIN PASSWORD '$WUBLOADER_PASSWORD';
+
+EOSQL
+
+
+if [ -n "$REPLICATION_USER" ]; then
+	echo "Creating $REPLICATION_USER"
+	# allow the $REPLICATION user to replicate remotely
+	echo "host replication $REPLICATION_USER all md5" >> "$PGDATA/pg_hba.conf"
+	psql -v ON_ERROR_STOP=1 -U postgres <<-EOSQL
+
+	CREATE USER $REPLICATION_USER LOGIN REPLICATION PASSWORD '$REPLICATION_PASSWORD';
+
+	EOSQL
+
+	cat >> ${PGDATA}/postgresql.conf <<-EOF
+	wal_level = replica
+	archive_mode = on
+	archive_command = 'cd .'
+	max_wal_senders = 8
+	wal_keep_segments = 8
+	EOF
+
+fi
+
+if [ -a /mnt/wubloader/nodes.csv ]; then
+	echo "Loading nodes from nodes.csv"
+	psql -v -U postgres -d ${POSTGRES_DB} <<-EOF
+	CREATE TABLE IF NOT EXISTS nodes (
+		name TEXT PRIMARY KEY,
+		url TEXT NOT NULL,
+		backfill_from BOOLEAN NOT NULL DEFAULT TRUE);
+	COPY nodes FROM '/mnt/wubloader/nodes.csv' DELIMITER ',' CSV HEADER;
+	ALTER TABLE nodes OWNER TO $WUBLOADER_USER;
+	EOF
+fi
+
+if [ -a /mnt/wubloader/editors.csv ]; then
+	echo "Loading editors from editors.csv"
+	psql -v -U postgres -d ${POSTGRES_DB} <<-EOF
+	CREATE TABLE IF NOT EXISTS editors (
+		email TEXT PRIMARY KEY,
+		name TEXT NOT NULL);
+	COPY editors FROM '/mnt/wubloader/editors.csv' DELIMITER ',' CSV HEADER;
+	ALTER TABLE editors OWNER TO $WUBLOADER_USER;
+	EOF
+fi
+

postgres/standby_setup.sh

@@ -0,0 +1,24 @@
+#! /bin/bash
+
+set -e
+
+# if postgres database does not exist in $PGDATA
+if [ ! -s "$PGDATA/PG_VERSION" ]; then
+
+	# get a binary backup of the database on $MASTER_NODE
+	pg_basebackup -d "host=$MASTER_NODE password=$REPLICATION_PASSWORD port=5432 user=$REPLICATION_USER" -D ${PGDATA} -vP
+	
+	cat > ${PGDATA}/recovery.conf <<-EOF
+	standby_mode = on
+	primary_conninfo = 'host=$MASTER_NODE password=$REPLICATION_PASSWORD port=5432 user=$REPLICATION_USER'
+	# touch this file to promote this node to master
+	trigger_file = '/tmp/touch_to_promote_to_master'
+	EOF
+
+	chown postgres. ${PGDATA} -R
+	chmod 700 ${PGDATA} -R
+
+fi
+
+# start postgres 
+gosu postgres postgres

sheetsync/sheetsync/main.py

@@ -280,7 +280,15 @@ def main(dbconnect, sheets_creds_file, edit_url, bustime_start, sheet_id, worksh
 
 	logging.info("Starting up")
 
-	dbmanager = DBManager(dsn=dbconnect)
+	dbmanager = None
+	while dbmanager is None:
+		try:
+			dbmanager = DBManager(dsn=dbconnect)
+		except Exception:
+			delay = common.jitter(10)
+			logging.info('Cannot connect to database. Retrying in {:.0f} s'.format(delay))
+			stop.wait(delay)
+
 	sheets_creds = json.load(open(sheets_creds_file))
 
 	sheets = Sheets(

thrimshim/thrimshim/main.py

@@ -3,6 +3,7 @@ from functools import wraps
 import json
 import logging
 import signal
+import sys
 
 import argh
 import flask
@@ -13,7 +14,8 @@ import prometheus_client
 import psycopg2
 from psycopg2 import sql
 
-from common import database, PromLogCountsHandler, install_stacksampler
+import common
+from common import database
 from common.flask_stats import request_stats, after_request
 
 import google.oauth2.id_token
@@ -271,16 +273,32 @@ def main(connection_string, host='0.0.0.0', port=8004, backdoor_port=0,
 	no_authentication=False):
 	"""Thrimshim service."""
 	server = WSGIServer((host, port), cors(app))
-	app.db_manager = database.DBManager(dsn=connection_string)
+
 	app.no_authentication = no_authentication
 
+	stopping = gevent.event.Event()
 	def stop():
-		logging.info('Shutting down')
-		server.stop()
+		logging.info("Shutting down")
+		stopping.set()
+		# handle when the server is running
+		if hasattr(server, 'socket'):
+			server.stop()
+		# and when not
+		else:
+			sys.exit()
 	gevent.signal(signal.SIGTERM, stop)
 
-	PromLogCountsHandler.install()
-	install_stacksampler()
+	app.db_manager = None
+	while app.db_manager is None and not stopping.is_set():
+		try:
+			app.db_manager = database.DBManager(dsn=connection_string)
+		except Exception:
+			delay = common.jitter(10)
+			logging.info('Cannot connect to database. Retrying in {:.0f} s'.format(delay))
+			stopping.wait(delay)
+
+	common.PromLogCountsHandler.install()
+	common.install_stacksampler()
 
 	if backdoor_port:
 		gevent.backdoor.BackdoorServer(('127.0.0.1', backdoor_port), locals=locals()).start()