Autopause update: allow network interface selection (#711)

* Cherry-pick: Fixing AUTOPAUSE on Raspberry Pi 4s (#708)

* implement autopause interface selection

* concentrate ps calls in function file

* wording and add note about PAPER's etc watchdogs

Dockerfile

@@ -74,7 +74,8 @@ ENV UID=1000 GID=1000 \
   TYPE=VANILLA VERSION=LATEST FORGEVERSION=RECOMMENDED SPONGEBRANCH=STABLE SPONGEVERSION= FABRICVERSION=LATEST LEVEL=world \
   PVP=true DIFFICULTY=easy ENABLE_RCON=true RCON_PORT=25575 RCON_PASSWORD=minecraft \
   LEVEL_TYPE=DEFAULT SERVER_PORT=25565 ONLINE_MODE=TRUE SERVER_NAME="Dedicated Server" \
-  ENABLE_AUTOPAUSE=false AUTOPAUSE_TIMEOUT_EST=3600 AUTOPAUSE_TIMEOUT_KN=120 AUTOPAUSE_TIMEOUT_INIT=600 AUTOPAUSE_PERIOD=10
+  ENABLE_AUTOPAUSE=false AUTOPAUSE_TIMEOUT_EST=3600 AUTOPAUSE_TIMEOUT_KN=120 AUTOPAUSE_TIMEOUT_INIT=600 \
+  AUTOPAUSE_PERIOD=10 AUTOPAUSE_KNOCK_INTERFACE=eth0
 
 COPY start* /
 COPY health.sh /

README.md

@@ -182,22 +182,22 @@ healthy
 
 Some orchestration systems, such as Portainer, don't allow for disabling the default `HEALTHCHECK` declared by this image. In those cases you can approximate the disabling of healthchecks by setting the environment variable `DISABLE_HEALTHCHECK` to `true`.
 
-## Autopause (experimental)
+## Autopause
 
 ### Description
 
-> EXPERIMENTAL: this feature only works with default bridge networking using official Docker distributions. Host networking and container management software, such as Portainer, and NAS solutions do not seem to provide compatible networking.
-
 There are various bug reports on [Mojang](https://bugs.mojang.com) about high CPU usage of servers with newer versions, even with few or no clients connected (e.g. [this one](https://bugs.mojang.com/browse/MC-149018), in fact the functionality is based on [this comment in the thread](https://bugs.mojang.com/browse/MC-149018?focusedCommentId=593606&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-593606)).
 
 An autopause functionality has been added to this image to monitor whether clients are connected to the server. If for a specified time no client is connected, the Java process is stopped. When knocking on the server port (e.g. by the ingame Multiplayer server overview), the process is resumed. The experience for the client does not change.
 
 Of course, even loaded chunks are not ticked when the process is stopped.
 
-From the server's point of view, the pausing causes a single tick to take as long as the process is stopped, so the server watchdog might intervene after the process is continued, possibly forcing a container restart. To prevent this, ensure that the `max-tick-time` in the `server.properties` file is set correctly.
+From the server's point of view, the pausing causes a single tick to take as long as the process is stopped, so the server watchdog might intervene after the process is continued, possibly forcing a container restart. To prevent this, ensure that the `max-tick-time` in the `server.properties` file is set correctly. Non-vanilla versions might have their own configuration file, you might have to disable their watchdogs separately (e.g. PAPER Servers).
 
 On startup the `server.properties` file is checked and, if applicable, a warning is printed to the terminal. When the server is created (no data available in the persistent directory), the properties file is created with the Watchdog disabled.
 
+The utility used to wake the server (`knock(d)`) works at network interface level. So the correct interface has to be set using the `AUTOPAUSE_KNOCK_INTERFACE` variable when using non-default networking environments (e.g. host-networking, Portainer oder NAS solutions). See the description of the variable below.
+
 A starting, example compose file has been provided in [examples/docker-compose-autopause.yml](examples/docker-compose-autopause.yml).
 
 ### Enabling Autopause
@@ -208,7 +208,7 @@ Enable the Autopause functionality by setting:
 -e ENABLE_AUTOPAUSE=TRUE
 ```
 
-There are 4 more environment variables that define the behaviour:
+The following environment variables define the behaviour of auto-pausing:
 * `AUTOPAUSE_TIMEOUT_EST`, default `3600` (seconds)
 describes the time between the last client disconnect and the pausing of the process (read as timeout established)
 * `AUTOPAUSE_TIMEOUT_INIT`, default `600` (seconds)
@@ -217,6 +217,8 @@ describes the time between server start and the pausing of the process, when no
 describes the time between knocking of the port (e.g. by the main menu ping) and the pausing of the process, when no client connects inbetween (read as timeout knocked)
 * `AUTOPAUSE_PERIOD`, default `10` (seconds)
 describes period of the daemonized state machine, that handles the pausing of the process (resuming is done independently)
+* `AUTOPAUSE_KNOCK_INTERFACE`, default `eth0`
+<br>Describes the interface passed to the `knockd` daemon. If the default interface does not work, run the `ifconfig` command inside the container and derive the interface receiving the incoming connection from its output. The passed interface must exist inside the container. Using the loopback interface (`lo`) does likely not yield the desired results.
 
 ## Deployment Templates and Examples
 

files/autopause/autopause-daemon.sh

@@ -2,23 +2,48 @@
 
 . /autopause/autopause-fcns.sh
 
-. /start-utils
+. ${SCRIPTS:-/}start-utils
 
-sudo /usr/sbin/knockd -c /tmp/knockd-config.cfg -d
-if [ $? -ne 0 ] ; then
+
+autopause_error_loop() {
+  logAutopause "Available interfaces within the docker container:"
+  INTERFACES=$(echo /sys/class/net/*)
+  INTERFACES=${INTERFACES//\/sys\/class\/net\//}
+  logAutopause "  $INTERFACES"
+  logAutopause "Please set the environment variable AUTOPAUSE_KNOCK_INTERFACE to the interface that handles incoming connections."
+  logAutopause "If unsure which interface to choose, run the ifconfig command in the container."
+  logAutopause "Autopause failed to initialize. This log entry will be printed every 30 minutes."
   while :
   do
-    if [[ -n $(ps -o comm | grep java) ]] ; then
-      break
-    fi
-    sleep 0.1
+    sleep 1800
+    logAutopause "Autopause failed to initialize."
   done
+}
+
+# wait for java process to be started
+while :
+do
+  if java_process_exists ; then
+    break
+  fi
+  sleep 0.1
+done
+
+# check for interface existence
+if [[ -z "$AUTOPAUSE_KNOCK_INTERFACE" ]] ; then
+  logAutopause "AUTOPAUSE_KNOCK_INTERFACE variable must not be empty!"
+  autopause_error_loop
+fi
+if ! [[ -d "/sys/class/net/$AUTOPAUSE_KNOCK_INTERFACE" ]] ; then
+  logAutopause "Selected interface \"$AUTOPAUSE_KNOCK_INTERFACE\" does not exist!"
+  autopause_error_loop
+fi
+
+sudo /usr/sbin/knockd -c /tmp/knockd-config.cfg -d -i "$AUTOPAUSE_KNOCK_INTERFACE"
+if [ $? -ne 0 ] ; then
   logAutopause "Failed to start knockd daemon."
-  logAutopause "Possible cause: docker's host network mode."
-  logAutopause "Recreate without host mode or disable autopause functionality."
-  logAutopause "Stopping server."
-  killall -SIGTERM java
-  exit 1
+  logAutopause "Probable cause: Unable to attach to interface \"$AUTOPAUSE_KNOCK_INTERFACE\"."
+  autopause_error_loop
 fi
 
 STATE=INIT

files/autopause/autopause-fcns.sh

@@ -8,6 +8,10 @@ java_running() {
   [[ $( ps -a -o stat,comm | grep 'java' | awk '{ print $1 }') =~ ^S.*$ ]]
 }
 
+java_process_exists() {
+  [[ -n "$(ps -a -o comm | grep 'java')" ]]
+}
+
 rcon_client_exists() {
   [[ -n "$(ps -a -o comm | grep 'rcon-cli')" ]]
 }

files/autopause/pause.sh

@@ -17,5 +17,5 @@ if [[ $( ps -a -o stat,comm | grep 'java' | awk '{ print $1 }') =~ ^S.*$ ]] ; th
 
   # finally pause the process
   logAutopauseAction "Pausing Java process"
-  killall -q -STOP java
+  pkill -STOP java
 fi

files/autopause/resume.sh

@@ -4,5 +4,5 @@
 
 if [[ $( ps -a -o stat,comm | grep 'java' | awk '{ print $1 }') =~ ^T.*$ ]] ; then
   logAutopauseAction "Knocked, resuming Java process"
-  killall -q -CONT java
+  pkill -CONT java
 fi

files/sudoers-mc

@@ -1,2 +1,2 @@
-%minecraft ALL=(ALL) NOPASSWD:/usr/bin/killall
+%minecraft ALL=(ALL) NOPASSWD:/usr/bin/pkill
 %minecraft ALL=(ALL) NOPASSWD:/usr/sbin/knockd

start-autopause

@@ -46,10 +46,15 @@ if ! [[ $AUTOPAUSE_TIMEOUT_INIT =~ ^[0-9]+$ ]] ; then
   export AUTOPAUSE_TIMEOUT_INIT
   log "Warning: AUTOPAUSE_TIMEOUT_INIT is not numeric, set to 600 (seconds)"
 fi
+if [[ "$AUTOPAUSE_KNOCK_INTERFACE" == "lo" ]] ; then
+  log "Warning: AUTOPAUSE_KNOCK_INTERFACE is set to the local loopback interface."
+  log "         This is not advisable, as incoming connections are likely not picked up there."
+  log "         Continuing with this setting."
+fi
 
-if [[ -n $MAX_TICK_TIME ]] ; then
+if [[ -n "$MAX_TICK_TIME" && "$MAX_TICK_TIME" != "-1" ]] ; then
   log "Warning: MAX_TICK_TIME is non-default, for autopause to work properly, this check should be disabled (-1 for versions >= 1.8.1)"
-else
+elif [[ -z "$MAX_TICK_TIME" ]] ; then
   if versionLessThan 1.8.1; then
     # 10 years
     MAX_TICK_TIME=315360000000