Difference between revisions of "User:Lwoggardner"

From MythTV Official Wiki
Jump to: navigation, search
m (Docker/Containers)
m (Docker/Containers)
 
Line 618: Line 618:
 
== Containers ==
 
== Containers ==
  
The ip addresses and ports, both for connecting *and for listening* are stored in the central database, if there are any remote frontends or backends then these ports must be exposed consistently.
+
The SQL database contains the ip addresses and ports, both for connecting *and for listening*, if there are any remote frontends or backends then these ports must be exposed consistently. Hostnames are important for frontends to find their own settings - but these can be locally defined.
 +
 
 +
Each set of containers for master backend, slave backends + frontends gets it own container network via tunnels.
 +
 
 +
When the backend container starts it could rewrite the MasterServerIP(6) and BackendServerIP(6) as a straightforward sql update?
 +
When the frontend containers start, (from a launch command on the master backend) they should rewrite config.xml with the IP address of the relevant master backend.
  
Therefore we probably want to use the overlay network where each mythtv environment will have an overlay network on the same subnet so the configuration data is portable between environments.
 
  
 
=== Master Backend ===
 
=== Master Backend ===
  
Ports used for communication between frontend/backend/mysql are not exposed externally and so we will use the IP addresses from the overlay network.
+
Acts as the master LXD daemon from which the containers are launched, and the network gateway for the container networks.
  
Port for mythweb is exposed externally, but can be proxied / forwarded.
+
UPnP would require multicast into the container (looks possible http://bartsimons.me/lxc-lxd-port-forwarding-containers-using-upnp/)
  
UPnP would require multicast into the container?
+
Multiple backends can run simultaneously, but dvb/usb tuners can only be accessed by one backend at a time, by choosing which tuners are exposed to each backend
  
Multiple backends can run simultaneously, but dvb/usb tuners can only be accessed by one backend at a time.
+
Apps - Mysql, Mythbackend, Shepherd all in same lxd container
  
Apps - Mysql, Mythbackend, Shepherd all in same container
+
Filesystem - base LXD filesystem covers the main directory. Still use other LVM volumes for the recordings directory. Or perhaps just use ZFS for all of it. Need to ensure the test network doesn't consume the whole disk?
  
* WORKING
 
** able to use dvb apps to scan in a container - uses -v /dev/dvb:/dev/dvb --privileged.
 
** launch mysql in baseimage, with mythtv user
 
** launch mythbackend as mythtv user (run second runit service as mythtv user, which then runs the backend)
 
** use nsenter (via docker-bash from phusion baseimage) to run mythtv-setup and scan channels.  Need to copy xauth from host into the container.
 
  
* TODO
+
? Multicast support for upnp
** script to initialise storage directories
 
** xmltv/mythfilldatabase (via shepherd in australia) - run in backend container because updating channel details is linked with mythtv
 
** https://github.com/dotcloud/docker/issues/3043 - multicast support required for upnp on 1900?
 
** mythweb
 
  
 
=== Frontend ===
 
=== Frontend ===
Line 676: Line 671:
 
==== Create container network for new version ====
 
==== Create container network for new version ====
  
# Create and mount LVM snapshot of mythtv filesystem.
+
# Create and mount LVM snapshot of mythtv storage filesystem.
# Create a new overlay network
+
# Clone the existing container to create "new_master" backend container, with volumes linked to snapshot
# Create "new_master" backend container, with volumes linked to snapshot, nics from overlay
+
# Update MasterServerIP / BackendServerIP
 
# Restore the database from latest backup (also from CONFIG backup if necessary)
 
# Restore the database from latest backup (also from CONFIG backup if necessary)
 
# Create "new_frontend"(s) attached to "new_master", ports etc automatically mapped to 99xx
 
# Create "new_frontend"(s) attached to "new_master", ports etc automatically mapped to 99xx

Latest revision as of 09:03, 14 January 2017

I was keeping track of where I was at with Myth on my home wiki, but then figured I might as well share it here!

My Setup

MythBuntu 12.04 with separated front-end/backend

  • Live TV/TV recording Free to air HD/SD (from Sydney, Australia)
  • Multi-rec
  • DTS/AC3 passthrough including 5,1 surround digital out
  • VDPAU
  • MythMusic
  • MythGallery
  • MythVideo
  • Frontend Suspend to RAM + Wake On USB
  • Backend Suspend to RAM

Backend

Hardware

  • HDDs totalling 1.2Tb
  • Intel S775 Core 2 Q6600 Quad-Core CPU
  • Gigabyte S775 GA-G33M-S2 Motherboard
  • 4GB RAM
  • PCI Hauppauge Nova-T-500 MCE(OEM) Dual DVBT tuner
  • PCI Hauppauge Nova-TD 400 Dual DVBT tuner

Software

Configuration

  • Channel 9, Gem, GO will not tune. The dtv_multiplex.network_id is null - set to 4114 seems to fix things.

Shepherd grabber

Standard install except cron job is disabled and allow mythtv to run mythfilldatabase. This is to allow mythtv to also be in control of suspend. Mythtv is configured to run mythfilldatabase daily at 5 minutes after the router wakes the server up for daily processing

Suspend to RAM on idle

Why Suspend to RAM and not just a full shutdown that would "just work"? Partly due to boot speed and partly just because.

Overview:

  • mythbackend controls idle detection, supported by a custom script to check other reasons to stay up, and sudo rules to allow passwordless root activity
  • suspend needs to unload dvb/usb modules which requires stopping mythbackend which means triggering suspend via upstart
  • wake up for recordings triggered by /sys/class/rtc as per ACPI_Wakeup, hwlock-save update disabled via upstart override
  • wake up for daily processing (anacron) triggered by cron running on my (always on) DD-WRT enabled router issuing a WOL
  • wake up for serving adhoc web requests (mythweb et al) triggered by [custom router script]

Settings (set via mythweb as server is basically headless)

Name Setting Value Why
Pre ShutdownWakeUp Check Command preSDWUCheckCommand /home/mythtv/bin/check-idle.sh check if i'm logged in, or machine is in use for some other reason
Block ShutdownWakeUp without client blockSDWUwithoutClient 0 because this is a backend only machine
Set Wakeup time command SetWakeuptimeCommand sudo /usr/local/sbin/set-wakeup.sh $time
Server Halt command ServerHaltCommand sudo /usr/local/sbin/emit-suspend.sh can't suspend directly because we need to shutdown mythbackend during suspend
Wakeup time format WakeupTimeFormat time_t as per ACPI wakeup page to use /sys/class/rtc
/home/mythtv/bin/check_idle.sh
#!/bin/bash

ACCESS_LOG="/var/log/apache2/access.log"
LOG_AGE="5 minutes ago"

check_frontend() {
# Not this doesn't care about what the FE is doing, just that it is up
expect - $1 <<'END_EXPECT'
#exp_internal 1
set timeout 2
log_user 0
set prompt "# "
set cmd "query location\r"
set frontend [lindex $argv 0]
spawn netcat $frontend 6546
  expect {
        timeout { send_user "\nNo prompt\n"; exit 1 }
        eof { send_user "\nNo output\n"; exit 1 }
        $prompt
  }

  send $cmd
  expect  {
       timeout { send_user "\nNo response\n"; exit 1 } 
       eof { send_user "\nPremature EOF\n"; exit 1 }
       -re "query location\r\n(.*)\r\r\n# "
  }       
  set location $expect_out(1,string)
  send "exit\r"
  expect eof
  send_user "MythTVLocation:$location\n"
END_EXPECT

}

log() {
  logger -s -t "mythtv-checkidle"
}

# Regardless of WOL settings and going into standbymode the 0.25 front-end
# simply cannot cope with the backend dropping out unexpectedly so we
# keep the BE up until the FE suspends itself
check_frontend mythion && {
    echo "Frontend is still up" | log
    exit 1
}

# anacron is running daily, weekly or monthly jobs
pgrep 'anacron' && {
    echo "Anacron job is running" | log
    exit 1
}

# apache request in last 5 minutes
if [ -f $ACCESS_LOG -a `stat -c %Y $ACCESS_LOG` -gt `date --date "$LOG_AGE" "+%s"` ] ; then
    echo "Web server in use" | log
    exit 1
fi

# users are logged in
if [ `who | wc -l` -gt 0 ] ; then
    echo "Users logged in" | log
    who | log
    exit 1
fi

/usr/bin/mythshutdown -v --check | log

exit ${PIPESTATUS[0]}
/etc/sudoers.d/mythtv
#Enable mythtv to set the real time clock and emit the suspending event without a password prompt

%mythtv ALL = NOPASSWD: /usr/local/sbin/emit-suspend.sh, /usr/local/sbin/set-wakeup.sh
/usr/local/sbin/set-wakeup.sh
#!/bin/sh
#$1 is the first argument to the script. It is the time in seconds since 1970
#this is defined in mythtv-setup with the time_t argument

# Save the hwclock
 . /etc/default/rcS
[ "$UTC" = "yes" ] && tz="--utc" || tz="--localtime"
[ "$BADYEAR" = "yes" ] && badyear="--badyear"
hwclock --rtc=/dev/rtc0 --systohc $tz --noadjfile $badyear

echo 0 > /sys/class/rtc/rtc0/wakealarm      #this clears your alarm.
echo $1 > /sys/class/rtc/rtc0/wakealarm     #this writes your alarm

echo "Set Alarm to Wakeup at $1 :  $(date -d "@$1")" | logger -t "set-wakeup"
cat /proc/driver/rtc | logger -t "set-wakeup"
/usr/local/sbin/emit-suspend.sh
#!/bin/sh
# Suspend via upstart
initctl emit suspending
/etc/init/suspending.conf
start on (suspending)

script
   logger -t "suspending" "Suspending via upstart" 
   pm-suspend
end script
/etc/pm/sleep.d/50_mythtv
#!/bin/sh
# The dvb modules need to be unloaded to allow suspend/resume to work
# To unload the modules mythbackend needs to be stopped
# 
# NOTE: Suspend cannot be executed in a child process of mythbackend itself (eg via server halt command)
# since it will simply abort at the point we stop mythbackend below. If this is required use upstart or some
# other means to decouple pm-suspend from the mythbackend process.
#!/bin/bash

LOG=/var/log/mythtv/mythsuspend.log
MODULE="dvb_usb_dib0700"

echo "Will now $1 mythtv" | tee -a $LOG

case $1 in
  suspend|hibernate)
        
        stop mythtv-backend 2>&1 | tee -a $LOG
      
        while [ `lsmod | grep "^${MODULE}" | wc -l` -gt 0 ]
        do 
            sleep 1
            modprobe -r $MODULE 2>&1 | tee -a $LOG 
        done
  ;;
  resume|thaw)
        
        modprobe $MODULE
        start mythtv-backend | tee -a $LOG
  ;;
esac

Frontend

Hardware

  • ZOTAC ionitx-d-e board, 2G memory
  • Szrealan E-2011 case (tiny!)
  • External DVD player
  • 1Tb external e-Sata drive
  • BenQ PE8700 DLP Projector
  • Sony Bravia TV over HDMI
  • CambridgeAudio AV receiver


Software

  • MythBuntu 12.04 64bit
  • Synergy to allow laptop to be the keyboard/mouse.

Configuration

Enable Wake on USB

We need to enable USB0 in /proc/acpi/wakeup

Note if USB2 is also enabled, then the board immediately wakes up regardless, unsure if this is due to another USB device

Seems to be enabled by default for Mythbuntu 12.04 but Current cheap as Chinese remote does not send the wakeup signal from the power button unfortunately

Auto-start X

  • disabled lightdm in via upstart override (echo 'manual' > /etc/init/lightdm.override)
  • Simple mingetty startup with a tty definition and a startx script in the console
  • user .xsession linked to /usr/share/mythbuntu/session.sh so we get the same xfce startup.
/etc/init/tty7.conf
# tty7 - mythtv
#
start on runlevel [23]
stop on runlevel [!23]

respawn
exec /sbin/mingetty --autologin=mythtv tty7
~mythtv/.profile

Appended to detect if we are on tty7, in which case we start-up X with an optional layout specified in ~/.mythrc

Note that when X is stopped we automatically logout and respawn

start-frontend() {
MYTHRC="$HOME/.mythrc"
if [ -r $MYTHRC ] ; then
   . $MYTHRC
fi

LAYOUT_SPEC=
[ ! -z "$LAYOUT" ] && LAYOUT_SPEC="-layout $LAYOUT"

startx --  $LAYOUT_SPEC -logverbose 5
logout
}

tty | grep tty7 && start-frontend
xorg.conf

For PE8700 Samsung LCD. See also the tips for the NVidiaProprietaryDriver

The FX5200 cards were the cause of my DVI problems on my projector. Apparently they freak out a PCB component causing the EDID data to become corrupt. (a Gigabyte card lasted from Aug 2006 to Dec 2007. Fortunately the new nvidia driver allow you to use a CustomEDID to load the data from a file.

Option ExactModeTimingsDVI is also important for the projector.

Option "UseEDIDDpi" "false" or some equivalent control of DPI is important for the Samsung television otherwise you get really small fonts (fixed in 0.21)

There are two layouts, the default uses outputs via VGA to a Samsung LCD television, the second uses DVI to the PE8700 projector.

Section "ServerLayout"
    Identifier     "Main" # Output to Samsung LCD television
#LCD[0] VGA, LCD[1] HDMI
    Screen         0 "LCD[1]" 0 0
    Option         "Xinerama" "off"
EndSection

Section "ServerLayout"
    Identifier     "Projector"
    Screen         0 "Projector[0]" 0 0
    Option         "Xinerama" "off"
EndSection
 
Section "Module"
        Load    "glx"
EndSection

Section "Device"
    Identifier     "ION-GPU[0]"
    Driver         "nvidia"
    BusID          "PCI:3:0:0"
    Screen         0
    #This prevents the cursor being displayed on resume from suspend
    Option         "HWCursor" "off"
EndSection

Section "Screen"
    Identifier     "LCD[1]"
    Device         "ION-GPU[0]"
    Monitor        "AnyMonitor"
    DefaultDepth    24
    Option         "UseDisplayDevice" "DFP-1" # HDMI
    Option         "ConnectedMonitor" "DFP-1"
EndSection

Section "Screen"
    Identifier     "LCD[0]"
    Device         "ION-GPU[0]"
    Monitor        "AnyMonitor"
    DefaultDepth    24
    Option         "ConnectedMonitor" "CRT-1"
    Option         "UseEDIDDpi" "FALSE" #Computed DPI values not so good for TV display
    Option         "UseDisplayDevice" "CRT-1" # VGA out
    SubSection     "Display"
        Depth       24
        Modes       "1360x768" "800x600"
    EndSubSection
EndSection

Section "Screen"
    Identifier     "Projector[0]"
    Device         "ION-GPU[0]"
    Monitor        "AnyMonitor"
    Option         "ConnectedMonitor" "DFP"
    Option         "UseDisplayDevice" "DFP" # DVI out
    Option         "CustomEDID" "DFP-0:/etc/X11/benq-c.bin"

    DefaultDepth    24
    SubSection     "Display"
        Depth       24
        Modes       "1280x720" "800x600"
    EndSubSection
EndSection

Section "Monitor"
    Identifier   "AnyMonitor"
    Option       "DPMS"
    Option       "ExactModeTimingsDVI" "true"
    Option       "RenderAccel" "true"
EndSection

~mythtv/.mythtv/session
#!/bin/sh
LAPTOP=

export | egrep "DBUS_SESSION_BUS_ADDRESS|DISPLAY" > ~/.xsession-export
#unclutter -root -idle 2 -display $DISPLAY &

# Start a synergy client
synergyc $LAPTOP

# Start an idle detection loop
idle_mythtv.sh & 
~mythtv/bin/idle_mythtv.sh
#!/bin/bash

WAIT_STARTUP=300
WAIT_CHECK=60

# Don't idle immediately!
sleep $WAIT_STARTUP 

while true
do
        if check_idle.sh ; then
	    suspend_mythtv.sh
            # we only do this because although irexec responds fine, mythtv's
            # lirc connection can't reset itself on resume from suspend
            kill_mythtv.sh

	    # if kill_mythtv above works then we'll be dead here
            sleep $WAIT_STARTUP
        else
	    sleep $WAIT_CHECK
        fi
done
~mythtv/bin/check_idle.sh
#!/bin/bash

check_frontend() {

expect - $1 <<'END_EXPECT'
#exp_internal 1
set timeout 2
log_user 0
set prompt "# "
set cmd "query location\r"
set frontend [lindex $argv 0]
spawn netcat $frontend 6546
  expect {
        timeout { send_user "\nNo prompt\n"; exit 1 }
        eof { send_user "\nNo output\n"; exit 1 }
        $prompt
  }

  send $cmd
  expect  {
       timeout { send_user "\nNo response\n"; exit 1 } 
       eof { send_user "\nPremature EOF\n"; exit 1 }
       -re "query location\r\n(.*)\r\r\n# "
  }       
  set location $expect_out(1,string)
  send "exit\r"
  expect eof
  send_user "MythTVLocation:$location\n"
END_EXPECT

}

# Is frontend in standbymode?
check_frontend localhost | grep "standbymode"
if [ ${PIPESTATUS[0]} -eq 0 -a ${PIPESTATUS[1]} -ne 0 ]; then
  echo "Frontend is up and not in standbymode"
  exit 1
fi

OTHER_USERS=$(who | grep -v "mythtv" |  wc -l)
if [ ${OTHER_USERS} -gt 0 ] ; then
  who | grep -v "mythtv"
  exit 1
fi

# Backups in progress
if [ -f /var/local/backups/mythion_daily.inprogress ] ; then
   echo "Jenix/Mythion backups in progress"
   exit 1
fi

# anacron is running daily, weekly or monthly jobs
pgrep 'anacron' && {
    echo "Anacron job is running"
    exit 1
}
exit 0
~mythtv/bin/suspend_mythtv.sh
#!/bin/sh

# We use dbus-send because it seems to block until resume which is what we want
echo "Suspending"

dbus-send --system --print-reply --dest=org.freedesktop.Hal /org/freedesktop/Hal/devices/computer \
        org.freedesktop.Hal.Device.SystemPowerManagement.Suspend int32:2

echo "Resumed"
~mythtv/bin/kill_mythtv.sh

Clean restart of the front end

Kill mythtv and logout of xfce (respawn of tty causes immediate restart)

#!/bin/sh

[ -f $HOME/.xsession-export ] && . $HOME/.xsession-export
echo "killing mythfrontend"
pkill mythfrontend
#pkill unclutter
echo "xfce logout $DISPLAY"
xfce4-session-logout --logout --fast

LIRC Remote Control

Standard mythbuntu setup for MCE remote.

power button is mapped vie irexec to kill_mythtv.sh

Sound

Onboard sound

Settings for using the receiver. When connected to TV just use HDMI audio, requires "SPDIF1" to be unmuted

~/.asoundrc

Could probably do better 48/44kHz switching on the analog output between music and video but if that is important we just go digital passthrough and let the receiver do the work.

In myth config, you can see that default sound device is ALSA:myth-video, the passthrough device is ALSA:myth-passthru, and mythmusic uses ALSA:myth-music. the AC3PassThru and DTSPassThru settings are managed by the ruby program

#First line comment

pcm.myth-music {
  type copy
  slave.pcm "default"
}

ctl.myth-music {
  type hw
  card 0
}

pcm.myth-video {
  type copy 
  slave.pcm "default"
}

ctl.myth-video {
  type hw
  card 0
}

pcm.myth-passthru {
  type copy
  slave.pcm = "spdif"
}

ctl.myth-passthru {
  type hw
  card 0
}

minimyth.conf

TODO - Works nicely through the WRT310N router etc, but we have a hard drive have decided to stick with Mythbuntu for now.

Upgrades

Mythbuntu 12.04

Reconfigured from scratch for no particular reason

Mythbuntu 10.10

100% clean on server

front end was missing non-versioned symlink for vdpau nvidia library

ln -s /etc/alternatives/libvdpau_nvidia.so /usr/lib/libvdpau_nvidia.so

Mythbuntu 9.10

100% clean on server

front end needed /etc/events.d/tty7 moved to /etc/init/tty7.conf

Mythbuntu 8.10

0.21-fixes

upgrade smooth

Docker/Containers

Thinking about how to run MythTV in containers for portability, helping with testing and upgrades, alternate configurations. None of this exists yet.

Images

Based on ubuntu because that's the reference build environment

  • mythtv-common - includes mythbuntu PPA, and installs mythtv-common.
  • mythtv-backend - phusion/baseimage + install backend from PPA (results in 1G image), init, sshd, mysql, cron (for mythfilldb and backups), mythbackend,
  • mythtv-frontend - simple install from PPA
  • mythweb - install from PPA

Containers

The SQL database contains the ip addresses and ports, both for connecting *and for listening*, if there are any remote frontends or backends then these ports must be exposed consistently. Hostnames are important for frontends to find their own settings - but these can be locally defined.

Each set of containers for master backend, slave backends + frontends gets it own container network via tunnels.

When the backend container starts it could rewrite the MasterServerIP(6) and BackendServerIP(6) as a straightforward sql update? When the frontend containers start, (from a launch command on the master backend) they should rewrite config.xml with the IP address of the relevant master backend.


Master Backend

Acts as the master LXD daemon from which the containers are launched, and the network gateway for the container networks.

UPnP would require multicast into the container (looks possible http://bartsimons.me/lxc-lxd-port-forwarding-containers-using-upnp/)

Multiple backends can run simultaneously, but dvb/usb tuners can only be accessed by one backend at a time, by choosing which tuners are exposed to each backend

Apps - Mysql, Mythbackend, Shepherd all in same lxd container

Filesystem - base LXD filesystem covers the main directory. Still use other LVM volumes for the recordings directory. Or perhaps just use ZFS for all of it. Need to ensure the test network doesn't consume the whole disk?


? Multicast support for upnp

Frontend

Since the frontend is attached to a physical display there can really only be one frontend running at any one time on a given host.

Video: Run X display on host, just run the app in the container forwarding X over a Unix socket. OpenGL & VDPAU/VAAPI should be fine in this arrangement.

ALSA: Pass /dev/snd, enabled devices in cgroups - lxc config (libcontainer?)

PulseAudio: run PA on host and enable networking over unix socket?

https://www.flockport.com/run-gui-apps-in-lxc-containers/

LIRC: run daemon in frontend container, passing devices, exposing unix socket

Proposed Upgrade approach using containers

The whole point of using is to enable rollback after an upgrade attempt.

Restoring a CONFIG backup

    1. Database restored from latest backup in mythtv filesystem
    2. Schema upgraded by mythtv-setup
    3. Take a new UPGRADED backup
    4. Restoring from CONFIG backup
    5. running a SQL script to empty partial restore tables (change to mythconverge_restore.pl?)
    6. running mythconverg_restore.pl to perform a partial restore, with plugin-data, from UPGRADED


Create container network for new version

  1. Create and mount LVM snapshot of mythtv storage filesystem.
  2. Clone the existing container to create "new_master" backend container, with volumes linked to snapshot
  3. Update MasterServerIP / BackendServerIP
  4. Restore the database from latest backup (also from CONFIG backup if necessary)
  5. Create "new_frontend"(s) attached to "new_master", ports etc automatically mapped to 99xx
  6. Configure/Test, recordings, guide data, front end playback, remote controls, music, images, dvds etc.
  7. --- time passes, family keeps using PROD, I keep testing until all is working (or we run out of space in the snapshot)
  8. Copy backups from this environment to somewhere outside of the snapshot to be used as CONFIG for subsequent iterations.
  9. Drop the snapshot, and containers
  10. Repeat as necessary to refresh data from PROD

Cutover

  1. Stop master backend, take a FINAL backup, stop all prod container(s)
  2. Create an LVM snapshot of mythtv filesystem (readonly for rollback)
  3. Create "new_prod" backend, with volumes linked to prod mythtv filesystem, and last good CONFIG backup from above.
  4. Start backend,front-ends - and away we go.

Rollback

  1. Stop and discard the new prod containers.
  2. Restore the LVM snapshot.
  3. Restart the old prod containers.

Cleanup

  1. Remove the LVM snapshot
  2. Discard old prod containers