Merge "initial work to enable systemd service running"

This commit is contained in:
Jenkins 2017-03-28 18:41:38 +00:00 committed by Gerrit Code Review
commit d7bc9486aa
7 changed files with 298 additions and 10 deletions

177
SYSTEMD.rst Normal file
View File

@ -0,0 +1,177 @@
===========================
Using Systemd in DevStack
===========================
.. note::
This is an in progress document as we work out the way forward here
with DevStack and systemd.
DevStack can be run with all the services as systemd unit
files. Systemd is now the default init system for nearly every Linux
distro, and systemd encodes and solves many of the problems related to
poorly running processes.
Why this instead of screen?
===========================
The screen model for DevStack was invented when the number of services
that a DevStack user was going to run was typically < 10. This made
screen hot keys to jump around very easy. However, the landscape has
changed (not all services are stoppable in screen as some are under
Apache, there are typically at least 20 items)
There is also a common developer workflow of changing code in more
than one service, and needing to restart a bunch of services for that
to take effect.
To enable this add the following to your local.conf::
USE_SYSTEMD=True
Unit Structure
==============
.. note::
Originally we actually wanted to do this as user units, however
there are issues with running this under non interactive
shells. For now, we'll be running as system units. Some user unit
code is left in place in case we can switch back later.
All DevStack user units are created as a part of the DevStack slice
given the name ``devstack@$servicename.service``. This lets us do
certain operations at the slice level.
Manipulating Units
==================
Assuming the unit ``n-cpu`` to make the examples more clear.
Enable a unit (allows it to be started)::
sudo systemctl enable devstack@n-cpu.service
Disable a unit::
sudo systemctl disable devstack@n-cpu.service
Start a unit::
sudo systemctl start devstack@n-cpu.service
Stop a unit::
sudo systemctl stop devstack@n-cpu.service
Restart a unit::
sudo systemctl restart devstack@n-cpu.service
See status of a unit::
sudo systemctl status devstack@n-cpu.service
Querying Logs
=============
One of the other major things that comes with systemd is journald, a
consolidated way to access logs (including querying through structured
metadata). This is accessed by the user via ``journalctl`` command.
Logs can be accessed through ``journalctl``. journalctl has powerful
query facilities. We'll start with some common options.
Follow logs for a specific service::
journalctl -f --unit devstack@n-cpu.service
Following logs for multiple services simultaneously::
journalctl -f --unit devstack@n-cpu.service --user-unit
devstack@n-cond.service
Use higher precision time stamps::
journalctl -f -o short-precise --unit devstack@n-cpu.service
Known Issues
============
Be careful about systemd python libraries. There are 3 of them on
pypi, and they are all very different. They unfortunately all install
into the ``systemd`` namespace, which can cause some issues.
- ``systemd-python`` - this is the upstream maintained library, it has
a version number like systemd itself (currently ``233``). This is
the one you want.
- ``systemd`` - a python 3 only library, not what you want.
- ``python-systemd`` - another library you don't want. Installing it
on a system will break ansible's ability to run.
If we were using user units, the ``[Service]`` - ``Group=`` parameter
doesn't seem to work with user units, even though the documentation
says that it should. This means that we will need to do an explicit
``/usr/bin/sg``. This has the downside of making the SYSLOG_IDENTIFIER
be ``sg``. We can explicitly set that with ``SyslogIdentifier=``, but
it's really unfortunate that we're going to need this work
around. This is currently not a problem because we're only using
system units.
Future Work
===========
oslo.log journald
-----------------
Journald has an extremely rich mechanism for direct logging including
structured metadata. We should enhance oslo.log to take advantage of
that. It would let us do things like::
journalctl REQUEST_ID=......
journalctl INSTANCE_ID=......
And get all lines related to the request id or instance id.
sub targets/slices
------------------
We might want to create per project slices so that it's easy to
follow, restart all services of a single project (like swift) without
impacting other services.
log colorizing
--------------
We lose log colorization through this process. We might want to build
a custom colorizer that we could run journalctl output through
optionally for people.
user units
----------
It would be great if we could do services as user units, so that there
is a clear separation of code being run as not root, to ensure running
as root never accidentally gets baked in as an assumption to
services. However, user units interact poorly with devstack-gate and
the way that commands are run as users with ansible and su.
Maybe someday we can figure that out.
References
==========
- Arch Linux Wiki - https://wiki.archlinux.org/index.php/Systemd/User
- Python interface to journald -
https://www.freedesktop.org/software/systemd/python-systemd/journal.html
- Systemd documentation on service files -
https://www.freedesktop.org/software/systemd/man/systemd.service.html
- Systemd documentation on exec (can be used to impact service runs) -
https://www.freedesktop.org/software/systemd/man/systemd.exec.html

View File

@ -575,7 +575,9 @@ function vercmp {
function setup_logging {
local conf_file=$1
local other_cond=${2:-"False"}
if [ "$LOG_COLOR" == "True" ] && [ "$SYSLOG" == "False" ] && [ "$other_cond" == "False" ]; then
if [[ "$USE_SYSTEMD" == "True" ]]; then
setup_systemd_logging $conf_file
elif [ "$LOG_COLOR" == "True" ] && [ "$SYSLOG" == "False" ] && [ "$other_cond" == "False" ]; then
setup_colorized_logging $conf_file
else
setup_standard_logging_identity $conf_file
@ -601,6 +603,17 @@ function setup_colorized_logging {
iniset $conf_file $conf_section logging_exception_prefix "%(color)s%(asctime)s.%(msecs)03d TRACE %(name)s %(instance)s"
}
function setup_systemd_logging {
local conf_file=$1
local conf_section="DEFAULT"
local project_var="project_name"
local user_var="user_name"
iniset $conf_file $conf_section logging_context_format_string "%(levelname)s %(name)s [%(request_id)s %("$project_var")s %("$user_var")s] %(instance)s%(message)s"
iniset $conf_file $conf_section logging_default_format_string "%(levelname)s %(name)s [-] %(instance)s%(color)s%(message)s"
iniset $conf_file $conf_section logging_debug_format_suffix "from (pid=%(process)d) %(funcName)s %(pathname)s:%(lineno)d"
iniset $conf_file $conf_section logging_exception_prefix "ERROR %(name)s %(instance)s"
}
function setup_standard_logging_identity {
local conf_file=$1
iniset $conf_file DEFAULT logging_user_identity_format "%(project_name)s %(user_name)s"

View File

@ -1443,6 +1443,59 @@ function _run_process {
exit 0
}
function write_user_unit_file {
local service=$1
local command="$2"
local group=$3
local user=$4
local extra=""
if [[ -n "$group" ]]; then
extra="Group=$group"
fi
local unitfile="$SYSTEMD_DIR/$service"
mkdir -p $SYSTEMD_DIR
iniset -sudo $unitfile "Unit" "Description" "Devstack $service"
iniset -sudo $unitfile "Service" "User" "$user"
iniset -sudo $unitfile "Service" "ExecStart" "$command"
if [[ -n "$group" ]]; then
iniset -sudo $unitfile "Service" "Group" "$group"
fi
iniset -sudo $unitfile "Install" "WantedBy" "multi-user.target"
# changes to existing units sometimes need a refresh
$SYSTEMCTL daemon-reload
}
function _run_under_systemd {
local service=$1
local command="$2"
local cmd=$command
local systemd_service="devstack@$service.service"
local group=$3
local user=${4:-$STACK_USER}
write_user_unit_file $systemd_service "$cmd" "$group" "$user"
$SYSTEMCTL enable $systemd_service
$SYSTEMCTL start $systemd_service
_journal_log $service $systemd_service
}
function _journal_log {
local service=$1
local unit=$2
local logfile="${service}.log.${CURRENT_LOG_TIME}"
local real_logfile="${LOGDIR}/${logfile}"
if [[ -n ${LOGDIR} ]]; then
$JOURNALCTL_F $2 > "$real_logfile" &
bash -c "cd '$LOGDIR' && ln -sf '$logfile' ${service}.log"
if [[ -n ${SCREEN_LOGDIR} ]]; then
# Drop the backward-compat symlink
ln -sf "$real_logfile" ${SCREEN_LOGDIR}/screen-${service}.log
fi
fi
}
# Helper to remove the ``*.failure`` files under ``$SERVICE_DIR/$SCREEN_NAME``.
# This is used for ``service_check`` when all the ``screen_it`` are called finished
# Uses globals ``SCREEN_NAME``, ``SERVICE_DIR``
@ -1478,16 +1531,24 @@ function run_process {
local service=$1
local command="$2"
local group=$3
local subservice=$4
local user=$4
local name=${subservice:-$service}
local name=$service
time_start "run_process"
if is_service_enabled $service; then
if [[ "$USE_SCREEN" = "True" ]]; then
if [[ "$USE_SYSTEMD" = "True" ]]; then
_run_under_systemd "$name" "$command" "$group" "$user"
elif [[ "$USE_SCREEN" = "True" ]]; then
if [[ "$user" == "root" ]]; then
command="sudo $command"
fi
screen_process "$name" "$command" "$group"
else
# Spawn directly without screen
if [[ "$user" == "root" ]]; then
command="sudo $command"
fi
_run_process "$name" "$command" "$group" &
fi
fi
@ -1618,6 +1679,14 @@ function stop_process {
if is_service_enabled $service; then
# Kill via pid if we have one available
if [[ "$USE_SYSTEMD" == "True" ]]; then
# Only do this for units which appear enabled, this also
# catches units that don't really exist for cases like
# keystone without a failure.
$SYSTEMCTL stop devstack@$service.service
$SYSTEMCTL disable devstack@$service.service
fi
if [[ -r $SERVICE_DIR/$SCREEN_NAME/$service.pid ]]; then
pkill -g $(cat $SERVICE_DIR/$SCREEN_NAME/$service.pid)
# oslo.service tends to stop actually shutting down

View File

@ -24,12 +24,12 @@ function start_dstat {
# To enable memory_tracker add:
# enable_service memory_tracker
# to your localrc
run_process memory_tracker "sudo $TOP_DIR/tools/memory_tracker.sh"
run_process memory_tracker "$TOP_DIR/tools/memory_tracker.sh" "" "root"
# remove support for the old name when it's no longer used (sometime in Queens)
if is_service_enabled peakmem_tracker; then
deprecated "Use of peakmem_tracker in devstack is deprecated, use memory_tracker instead"
run_process peakmem_tracker "sudo $TOP_DIR/tools/memory_tracker.sh"
run_process peakmem_tracker "$TOP_DIR/tools/memory_tracker.sh" "" "root"
fi
}

View File

@ -602,8 +602,11 @@ function start_keystone {
tail_log key /var/log/$APACHE_NAME/keystone.log
tail_log key-access /var/log/$APACHE_NAME/keystone_access.log
else # uwsgi
run_process key "$KEYSTONE_BIN_DIR/uwsgi $KEYSTONE_PUBLIC_UWSGI_FILE" "" "key-p"
run_process key "$KEYSTONE_BIN_DIR/uwsgi $KEYSTONE_ADMIN_UWSGI_FILE" "" "key-a"
# TODO(sdague): we should really get down to a single keystone here
enable_service key-p
enable_service key-a
run_process key-p "$KEYSTONE_BIN_DIR/uwsgi $KEYSTONE_PUBLIC_UWSGI_FILE" ""
run_process key-a "$KEYSTONE_BIN_DIR/uwsgi $KEYSTONE_ADMIN_UWSGI_FILE" ""
fi
echo "Waiting for keystone to start..."

View File

@ -38,6 +38,15 @@ fi
# Set up default directories
GITDIR["python-swiftclient"]=$DEST/python-swiftclient
# Swift virtual environment
if [[ ${USE_VENV} = True ]]; then
PROJECT_VENV["swift"]=${SWIFT_DIR}.venv
SWIFT_BIN_DIR=${PROJECT_VENV["swift"]}/bin
else
SWIFT_BIN_DIR=$(get_python_exec_prefix)
fi
SWIFT_DIR=$DEST/swift
SWIFT_AUTH_CACHE_DIR=${SWIFT_AUTH_CACHE_DIR:-/var/cache/swift}
SWIFT_APACHE_WSGI_DIR=${SWIFT_APACHE_WSGI_DIR:-/var/www/swift}
@ -807,10 +816,10 @@ function start_swift {
local proxy_port=${SWIFT_DEFAULT_BIND_PORT}
start_tls_proxy swift '*' $proxy_port $SERVICE_HOST $SWIFT_DEFAULT_BIND_PORT_INT
fi
run_process s-proxy "swift-proxy-server ${SWIFT_CONF_DIR}/proxy-server.conf -v"
run_process s-proxy "$SWIFT_BIN_DIR/swift-proxy-server ${SWIFT_CONF_DIR}/proxy-server.conf -v"
if [[ ${SWIFT_REPLICAS} == 1 ]]; then
for type in object container account; do
run_process s-${type} "swift-${type}-server ${SWIFT_CONF_DIR}/${type}-server/1.conf -v"
run_process s-${type} "$SWIFT_BIN_DIR/swift-${type}-server ${SWIFT_CONF_DIR}/${type}-server/1.conf -v"
done
fi

17
stackrc
View File

@ -87,6 +87,23 @@ HORIZON_APACHE_ROOT="/dashboard"
# be disabled for automated testing by setting this value to False.
USE_SCREEN=$(trueorfalse True USE_SCREEN)
# Whether to use SYSTEMD to manage services
USE_SYSTEMD=$(trueorfalse False USE_SYSTEMD)
USER_UNITS=$(trueorfalse False USER_UNITS)
if [[ "$USER_UNITS" == "True" ]]; then
SYSTEMD_DIR="$HOME/.local/share/systemd/user"
SYSTEMCTL="systemctl --user"
JOURNALCTL_F="journalctl -f -o short-precise --user-unit"
else
SYSTEMD_DIR="/etc/systemd/system"
SYSTEMCTL="sudo systemctl"
JOURNALCTL_F="journalctl -f -o short-precise --unit"
fi
if [[ "$USE_SYSTEMD" == "True" ]]; then
USE_SCREEN=False
fi
# When using screen, should we keep a log file on disk? You might
# want this False if you have a long-running setup where verbose logs
# can fill-up the host.