+list Thanks,
Guido ---------- Forwarded message ---------- From: Guido Trotter <[email protected]> Date: Mon, Apr 22, 2013 at 8:37 AM Subject: Re: [PATCH 1/2] Added experimental resource agent (OCF) for Ganeti Master role in for pacemaker/corosync. To: Giuseppe Paterno' <[email protected]> On Sun, Apr 21, 2013 at 11:09 PM, Giuseppe Paterno' <[email protected]> wrote: Hi, Thanks for helping with HA integration! > From: Giuseppe Paterno <[email protected]> > > --- > doc/examples/GanetiMaster.in | 285 > +++++++++++++++++++++++++++++++++++++++++++ Why writing a new one instead of patching the current one? > 1 file changed, 285 insertions(+) > create mode 100755 doc/examples/GanetiMaster.in > > diff --git a/doc/examples/GanetiMaster.in b/doc/examples/GanetiMaster.in > new file mode 100755 > index 0000000..55c8cfc > --- /dev/null > +++ b/doc/examples/GanetiMaster.in > @@ -0,0 +1,285 @@ > +#!/bin/sh > +# > +# Resource Agent for managing Ganeti master role. > +# > +# License: GNU General Public License (GPL) > +# > +# (c) 2010-2013 Giuseppe Paterno' <[email protected]> > +# Guido Trotter <[email protected]> > +# > +# > +# This resource agent will ensure that ganeti-masterd and > +# ganeti-rapi are running in the master node. > +# > +# *WARNING*: This has not been extensively tested > +# > +# Copy in: > +# /usr/lib/ocf/resource.d/ganeti/ > +# > + > +# Agent version > +VERSION="0.3" > + > +: ${OCF_ROOT=/usr/lib/ocf} > +# Load standard OCF functions > +: ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/resource.d/heartbeat} > +. ${OCF_FUNCTIONS_DIR}/.ocf-shellfuncs > + > +#/usr/lib/ganeti/daemon-util check-config > + > +## Adding in path Ganeti > +PATH=$PATH:@PKGLIBDIR@ > + I would leave SHELL_ENV_INIT around, even if this for now will not be used with vcluster. > +## Binaries > +DAEMON_UTIL="daemon-util" > +MASTERD="ganeti-masterd" > +CLUSTER_CLI="gnt-cluster" > +RAPI="ganeti-rapi" > + > +## Not using Ganeti's voting system > +## as pacemaker will take care of voting and quorum > +MASTERD_OPTIONS="--no-voting --yes-do-it" Yes, pacemaker can, but what's the problem with leaving both enabled? (note that ganeti won't "decide" for a different master, it will just "vote" to verify that the cluster agrees) > + > +## Who am I > +HOSTNAME=$(hostname -f) > + > +## List of master candidates > +MASTER_CND_FILE="/var/lib/ganeti/ssconf_master_candidates" > + > +## RAPI endpoint > +RAPI_URL="https://localhost:5080/2/info"; > + > +# Pre-flight checks > +ganeti_master_validate_all() { > + > + # Check binaries > + check_binary $DAEMON_UTIL > + check_binary $MASTERD > + check_binary $RAPI > + check_binary $CLUSTER_CLI > + > + # Check if I'm master/master-candidate > + grep -Fx $HOSTNAME $MASTER_CND_FILE >/dev/null > + > + if [ $? -ne 0 ] ; then > + ocf_log err "$HOSTNAME is not Ganeti master neither master candidate" > + return $OCF_ERR_INSTALLED > + fi > + > + return $OCF_SUCCESS > + > +} > + > +# Script usage :) > +ganeti_master_usage() { > + > + cat << END > +usage: $0 action > + > +action: > + start : start Ganeti master role > + stop : stop Ganeti mater role > + status : return the status of Ganeti master daemon > + monitor : return if Ganeti Master daemon is working > + meta-data : show meta data message > + validate-all: validate the instance parameters > + > +END > + > + return $OCF_ERR_ARGS > +} > + > +# Return meta-data information > +ganeti_master_meta_data() { > + cat <<END > +<?xml version="1.0"?> > +<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd"> > +<resource-agent name="GanetiMaster"> > +<version>$VERSION</version> > + > +<longdesc lang="en"> > +OCF script to manage the ganeti master role in a cluster. > +Can be used to failover the ganeti master between master candidate nodes. > +</longdesc> > + > +<shortdesc lang="en">Manages the ganeti master role</shortdesc> > + > +<parameters> > +</parameters> > + > +<actions> > +<action name="start" timeout="60s" /> > +<action name="stop" timeout="60s" /> > +<action name="monitor" depth="0" timeout="20s" interval="30s" /> > +<action name="meta-data" timeout="5s" /> > +<action name="recover" timeout="20s" /> > +<action name="reload" timeout="30s" /> > +</actions> > +</resource-agent> > +END > + > + return $OCF_SUCCESS > +} > + > +# Start Master > +ganeti_master_start() { > + # exit immediately if configuration is not valid > + ganeti_master_validate_all || exit $? > + > + # Check if masterd is running and start if not > + is_masterd_running > + > + if [ $? -ne 0 ] > + then > + > + # Check who's the previous master > + CUR_MASTER=$($CLUSTER_CLI getmaster) > + ocf_log debug "Current master is $CUR_MASTER" > + > + # Executing failover if we are not the master > + if [ "x$CUR_MASTER" != "x$HOSTNAME" ] ; then > + ocf_log info "$CUR_MASTER is no longer the master, taking over ..." > + $CLUSTER_CLI master-failover $MASTERD_OPTIONS > + else > + # Start masterd > + ocf_log info "Starting $MASTERD" > + ocf_run $DAEMON_UTIL start $MASTERD $MASTERD_OPTIONS > + fi > + > + # Distribute configuration if I wasn't > + # the previous master. I cannot guarantee > + # the conf of a dead node Here an below, we should use impersonal comments, rather than personal ones "The configuration on a dead node cannot be guaranteed, rather than..." > + ocf_log info "Redistributing config in the Ganeti cluster ..." > + $CLUSTER_CLI redist-conf > + > + ## I should check the exit code > + ## but masterd might fail activating IP > + > + fi > + > + # Check if RAPI endpoint is running and start if not > + is_rapi_running > + > + if [ $? -ne 0 ] > + then > + > + # Start RAPI > + ocf_log info "Starting $RAPI" > + ocf_run $DAEMON_UTIL start $RAPI || exit $OCF_ERR_GENERIC > + > + fi > + > + return $OCF_SUCCESS > + > +} > + > +# Stop Master > +ganeti_master_stop() { > + > + # Stopping masterd > + ocf_log info "Sopping $MASTERD" > + ocf_run $DAEMON_UTIL stop $MASTERD > + > + # Stopping RAPI > + ocf_log info "Stopping $RAPI" > + ocf_run $DAEMON_UTIL stop $RAPI > + > + return $OCF_SUCCESS > + > +} > + > +## Check if RAPI is running > +is_rapi_running() { > + > + curl -k $RAPI_URL >/dev/null 2>/dev/null > + > + if [ $? -ne 0 ] > + then > + ocf_log info "RAPI endpoint not running." > + return $OCF_NOT_RUNNING > + else > + ocf_log info "RAPI endpoint running." > + return $OCF_SUCCESS > + fi > + > +} > + > +## Check if masterd is running > +is_masterd_running() { > + > + $CLUSTER_CLI master-ping > + > + if [ $? -ne 0 ] > + then > + ocf_log info "Master daemon not running." > + return $OCF_NOT_RUNNING > + else > + ocf_log info "Master daemon running." > + return $OCF_SUCCESS > + fi > + > +} > + > +# Monitor masterd status > +ganeti_master_monitor() { > + > + is_masterd_running > + master_status=$? > + > + is_rapi_running > + rapi_status=$? > + > + if [ $master_status -eq 0 -a $rapi_status -eq 0 ] > + then > + ocf_log info "Ganeti master service is running" > + return $OCF_SUCCESS > + else > + ocf_log info "Ganeti master service is not running" > + return $OCF_NOT_RUNNING > + fi > + This will make it very hard to temporarily stop ganeti, which sometimes one may want to do. The old approach was just to make sure one node had the master "role" or status, but didn't verify the service at all. This allowed easily to stop ganeti for an upgrade or other reason without any failure. Here if one stops pacemaker the master will immediately be failed over, but if one stops masterd it will be immediately restarted as well. I'm not saying this is inherently wrong, but perhaps a choice could be given? > +} > + > + > +# Make sure meta-data and usage always succeed > +case $__OCF_ACTION in > + > + meta-data) ganeti_master_meta_data > + exit $OCF_SUCCESS > + ;; > + usage|help) ganeti_master_usage > + exit $OCF_SUCCESS > + ;; > + > +esac > + > +# Anything other than meta-data and usage must pass validation > +ganeti_master_validate_all || exit $? > + > +# Translate each action into the appropriate function call > +case $__OCF_ACTION in > + > + start) ganeti_master_start > + ;; > + stop) ganeti_master_stop > + ;; > + status|monitor) ganeti_master_monitor > + ;; > + promote) exit $OCF_SUCCESS > + ;; > + demote) exit $OCF_SUCCESS > + ;; Why not non implemented? > + reload) ocf_log info "Reloading Ganeti Master..." > + ganeti_master_start > + ;; > + validate-all) ;; > + *) ganeti_master_usage > + exit $OCF_ERR_UNIMPLEMENTED > + ;; > + > +esac > +rc=$? > + > +# Final debug > +ocf_log debug "${OCF_RESOURCE_INSTANCE} $__OCF_ACTION returned $rc" > +exit $rc > -- > 1.7.12.4 (Apple Git-37) > Thanks!! Guido -- Guido Trotter Ganeti Engineering Google Germany GmbH Dienerstr. 12, 80331, München Registergericht und -nummer: Hamburg, HRB 86891 Sitz der Gesellschaft: Hamburg Geschäftsführer: Graham Law, Katherine Stephens Steuernummer: 48/725/00206 Umsatzsteueridentifikationsnummer: DE813741370
