hms_actions.sh #1

# hms_actions.sh
Version=1.0.35

#==============================================================================
# Copyright and license info is available in the LICENSE file included with
# the Server Deployment Package (SDP), and also available online:
# https://swarm.workshop.perforce.com/projects/perforce-software-sdp/view/main/LICENSE
#------------------------------------------------------------------------------

#==============================================================================
# HMS Library Functions.

#------------------------------------------------------------------------------
# change_component_status (start/stop, $component, $tag)
# $1 - start or stop
# $2 - fully qualified component name (<instance>:<component>, e.g.
# 1:p4broker01).
# $3 - optional tag, used only for brokers, e.g. "dfm"
#------------------------------------------------------------------------------
function change_component_status () {
   vvmsg "CALL: change_component_status ($*)"
   declare command=${1:-Unset}
   declare fqComponent=${2:-Unset}
   declare tag=${3:-""}
   declare componentType=
   declare componentCommand=
   declare componentMasterHost=
   declare instance=
   declare initScript=
   declare output=$P4U_TMPDIR/ccs.${fqComponent/:/_}.$RANDOM$RANDOM
   declare -i exitCode=0

   if [[ $fqComponent != *":"* ]]; then
      errmsg "Unable to start/stop entire instance yet; this will be implemented later."
      return 1
   fi

   instance=${fqComponent%%:*}

   if [[ $command =~ ^(start|stop)$ ]]; then
      componentType=${ComponentType[$fqComponent]:-Unset}
      componentMasterHost=${ComponentMasterHost[$fqComponent]:-Unset}

      case "$componentType" in
         (p4d-*) initScript="/p4/$instance/bin/p4d_${instance}_init";;
         (p4broker) initScript="/p4/$instance/bin/p4broker_${instance}_init";;
         (p4p) initScript="/p4/$instance/bin/p4p_${instance}_init";;
         (*)
            errmsg "Don't know how to start component [$fqComponent] of type [$componentType]."
            return 1
         ;;
      esac

      if [[ -n "$initScript" ]]; then
         componentCommand="$initScript $command $tag"
      else
         errmsg "No init script for component [$fqComponent]."
         return 1
      fi

      runRemoteCmd "$componentMasterHost" "$componentCommand" \
         "Executing $command on $componentType $fqComponent on $componentMasterHost" 1 1 0
      exitCode=$RCMDEXITCODE
      [[ $exitCode -eq 0 ]] || return 1
   else
      errmsg "Unrecognized start/stop command [$command]."
      return 1
   fi
   return 0
}

function get_replica_status () {
   vvmsg "CALL: get_replica_status ($*)"
   declare instance=${1:-Unset}
   declare -i summaryExitCode=0

   [[ $instance == Unset ]] && return 1

   msg "Getting list of managed components for instance $instance."
   componentList=$(get_component_list "$instance" 1)

   if [[ $? -ne 0 ]]; then
      errmsg "Could not get a list of valid components for instance $instance. Aborting replica status."
      return 1
   fi
   source /p4/common/bin/p4_vars $instance 
   for fqComponent in $componentList; do
      componentType=${ComponentType[$fqComponent]:-Unset}
      componentMasterHost=${ComponentMasterHost[$fqComponent]:-Unset}
      componentURL=${ComponentURL[$fqComponent]:-Unset}
      componentManaged=${ComponentManaged[$fqComponent]:-Unset}
      vvmsg "Type=$componentType Master=$componentMasterHost URL=$componentURL managed=$componentManaged"
      [[ $componentType == p4d-e ]] || [[ $componentType == p4d-r ]] || continue
      if [[ $componentURL == "ssl:"* ]]; then
                componentP4PORT="ssl:$componentMasterHost:${componentURL#ssl:}"
      else
                componentP4PORT="$componentMasterHost:$componentURL"
      fi
      componentStatusCmd="$P4BIN -s -p $componentP4PORT pull -lj "
      componentStatusCode=0
      run "$componentStatusCmd" \
         "3:Checking replica status for $fqComponent .|4:Checking with command: $componentStatusCmd" 1 1 ||\
         componentStatusCode=1
   done
}

function get_diskspace () {
   vvmsg "CALL: get_diskspace ($*)"
   declare instance=${1:-Unset}
   declare -i summaryExitCode=0

   [[ $instance == Unset ]] && return 1

   msg "Getting list of managed components for instance $instance."
   componentList=$(get_component_list "$instance" 1)

   if [[ $? -ne 0 ]]; then
      errmsg "Could not get a list of valid components for instance $instance. Aborting replica status."
      return 1
   fi
   source /p4/common/bin/p4_vars $instance 
   for fqComponent in $componentList; do
      componentType=${ComponentType[$fqComponent]:-Unset}
      componentMasterHost=${ComponentMasterHost[$fqComponent]:-Unset}
      componentURL=${ComponentURL[$fqComponent]:-Unset}
      componentManaged=${ComponentManaged[$fqComponent]:-Unset}
      [[ $componentType == p4d-e ]] || [[ $componentType == p4d-r ]] || continue
      if [[ $componentURL == "ssl:"* ]]; then
                componentP4PORT="ssl:$componentMasterHost:${componentURL#ssl:}"
      else
                componentP4PORT="$componentMasterHost:$componentURL"
      fi
      componentStatusCmd="$P4BIN -s -p $componentP4PORT df "
      componentStatusCode=0
      run "$componentStatusCmd" \
         "3:Checking diskspace for $fqComponent .|4:Checking with command: $componentStatusCmd" 1 1 ||\
         componentStatusCode=1
   done
}
#------------------------------------------------------------------------------
# Function: down_for_maintenance ($subCommand, $instance)
# $1 - on, off, or status
# $2 - Instance
#------------------------------------------------------------------------------
function down_for_maintenance () {
   vvmsg "CALL: down_for_maintenance ($*)"
   declare subCommand=${1:-Unset}
   declare instance=${2:-Unset}
   declare -i summaryExitCode=0

   [[ $instance == Unset || $subCommand == Unset ]] && return 1

   msg "Getting list of managed components for instance $instance."
   componentList=$(get_component_list "$instance" 1)

   if [[ $? -ne 0 ]]; then
      errmsg "Could not get a list of valid components for instance $instance. Aborting dfm."
      return 1
   fi

   # Put all brokers into Down For Maintenance mode (on), or bring
   # them out (off), or check on them (status).
   msg "Setting Down For Maintenance message on all available brokers."
   for fqComponent in $componentList; do
      componentType=${ComponentType[$fqComponent]:-Unset}
      [[ $componentType == p4broker ]] || continue
      componentMasterHost=${ComponentMasterHost[$fqComponent]:-Unset}
      if [[ $subCommand == on ]]; then
         change_component_status stop "$fqComponent"
         change_component_status start "$fqComponent" dfm
      elif [[ $subCommand == off ]]; then
         change_component_status stop "$fqComponent" dfm
         change_component_status start "$fqComponent"
      elif [[ $subCommand == status ]]; then
         get_component_status "$fqComponent" 1 dfm
      else
         errmsg "Unknown dfm subcommand [$subCommand]."
         return 1
      fi
   done
}

#------------------------------------------------------------------------------
# Function: execute_instance_failover ($instance, $path, $style)
#
# This function executes failover of all managed components for the instance.
#
# This is called by failover(), which does data verification.  This routine
# can thus dispense with some of the "defensive programming" done in
# failover(), allowing us to assume our data model (implemenented primarily
# with associative arrays) has what we need.
#------------------------------------------------------------------------------
function execute_instance_failover () {
   vvmsg "CALL: execute_instance_failover ($*)"

   declare instance=${1:-Unset}
   declare path=${2:-Unset}
   declare style=${3:-Unset}
   declare componentList=
   declare componentType=
   declare componentMasterHost=
   declare brokerDfmOutput=
   declare cmd=
   declare -i summaryExitCode=0

   [[ $instance == Unset || $path == Unset || $style == Unset ]] && return 1

   msg "Getting list of managed components for instance $instance."
   componentList=$(get_component_list "$instance" 1)
   if [[ $? -ne 0 ]]; then
      errmsg "Could not get a list of valid components for instance $instance. Aborting failover."
      return 1
   fi

   # Put all (available) brokers in Down For Maintenance mode.
   msg "Setting Down For Maintenance message on all available brokers."
   down_for_maintenance on $instance

   for fqComponent in $componentList; do
      componentType=${ComponentType[$fqComponent]:-Unset}
      [[ $componentType == p4d-e ]] || break

      failover_p4d_edge "$path" "$fqComponent" "$style" ||\
         summaryExitCode=1
   done

   for fqComponent in $componentList; do
      componentType=${ComponentType[$fqComponent]:-Unset}
      [[ $componentType == p4d-mc ]] || break

      failover_p4d_master "$path" "$fqComponent" "$style" ||\
         summaryExitCode=1
   done

   for fqComponent in $componentList; do
      componentType=${ComponentType[$fqComponent]:-Unset}
      case "$componentType" in
         (p4p|p4dtg|p4web|p4gf|swarm)
            warnmsg "Failover for component type [$componentType] not impelmented."
         ;; 
         (Unset)
            errmsg "ComponentType not defined for component $component."
            summaryExitCode=1
         ;; 
      esac
   done

   # Finally, bring all (available) brokers back online.
   down_for_maintenance off $instance

   return $summaryExitCode
}

#------------------------------------------------------------------------------
# Function: execute_edge_failover ($instance, $path, $style, $component)
#
# This function executes failover of an edge server.
#
# This is called by failover(), which does data verification.  This routine
# can thus dispense with some of the "defensive programming" done in
# failover(), allowing us to assume our data model (implemenented primarily
# with associative arrays) has what we need.
#------------------------------------------------------------------------------
function execute_edge_failover () {
   vvmsg "CALL: execute_edge_failover ($*)"

   declare instance=${1:-Unset}
   declare path=${2:-Unset}
   declare style=${3:-Unset}
   declare component=${4:-Unset}
   declare -i summaryExitCode=0

   [[ $instance == Unset || $path == Unset || $style == Unset || $component == Unset ]] && return 1

   fqComponent="${instance}:${component}"

   failover_p4d_edge "$path" "$fqComponent" "$style" ||\
      summaryExitCode=1

   return $summaryExitCode
}

#------------------------------------------------------------------------------
# Function: failover_p4d_edge ($path, $fqComponent, $style)
#------------------------------------------------------------------------------
function failover_p4d_edge () {
   vvmsg "CALL: failover_p4d_edge ($*)"
   declare path=${1:-Unset}
   declare fqComponent=${2:-Unset}
   declare style=${3:-Unset}

   declare backupHost=
   declare instance=

   declare edgeHost=
   declare backupHost=
   declare instance=
   declare fqPath=
   declare newServerID=
   declare -i exitCode=0 summaryExitCode=0

   [[ $path == Unset || $fqComponent == Unset || $style == Unset ]] && return 1

   instance=${fqComponent%%:*}
   fqPath=$instance:$path
   edgeHost=${FailoverMasterHost[$fqPath]:-Unset}
   backupHost=${FailoverBackupHost[$fqPath]:-Unset]}
   failoverType=${FailoverType[$fqPath]:-Unset}

   if [[ $failoverType == EdgeLocal ]]; then
      runRemoteCmd "$edgeHost" \
         "$HMS_SCRIPTS/failover_p4d_local.sh $instance $style $NO_OP" \
       "Failing over to offline dbs locally on $edgeHost. Running: failover_p4d_local.sh I=$instance S=$style NO_OP=$NO_OP" 1 1 0
      exitCode=$RCMDEXITCODE
      summaryExitCode=$exitCode
      if [[ $exitCode -eq 0 ]]; then
         msg "Local Failover on $edgeHost completed OK."
      else
         errmsg "Local Failover on $edgeHost FAILED."
         return 1
      fi
   elif [[ $failoverType == EdgeMO || $failoverType == EdgeFull ]]; then
      # The difference between Scheduled and Unscheduled failover is partly one
      # of expectations.  In Unscheduled Failover, we don't expect the edge
      # to respond.  But in either case, we do our best to avoid split brain,
      # and try to shut it down.
      if [[ $style == Scheduled ]]; then
         msg "Shutting down edge server for Scheduled Failover.  We expect it to shutdown cleanly." 
      else
         msg "Attempting shutdown of edge server for Unscheduled Failover.  It may or may not respond."
      fi

      runRemoteCmd "$edgeHost" \
         "$HMS_SCRIPTS/failover_p4d_from_this_host.sh $instance $style $failoverType $backupHost $NO_OP" \
       "Failing over from $edgeHost. Running: failover_p4d_from_this_host.sh I=$instance S=$style T=$failoverType BH=$backupHost NO_OP=$NO_OP" 1 1 0
      exitCode=$RCMDEXITCODE
      if [[ $exitCode -ne 0 ]]; then
         if [[ $style == Scheduled ]]; then
            errmsg "Shutdown of edge server FAILED during Scheduled Failover. Aborting failover of $fqComponent."
            return 1
         else
            warnmsg "Shutdown of edge server failed during Unscheduled Failover.  Continuing with failover of $fqComponent."
         fi
      fi

      # newServerID is the server spec we want the edge server to have,
      # the ServerID of the original edge sever.
      newServerID=${path#p4d_}
      newServerID="p4d_${path}"

      runRemoteCmd "$backupHost" \
         "$HMS_SCRIPTS/failover_p4d_to_this_host.sh $instance $style $failoverType $newServerID $NO_OP" \
         "Failing over to $backupHost. Running: $HMS_SCRIPTS/failover_p4d_to_this_host.sh I=$instance S=$style T=$failoverType SID=$newServerID NO_OP=$NO_OP" 1 1 0
      exitCode=$RCMDEXITCODE
      if [[ $exitCode -ne 0 ]]; then
         errmsg "Failover of $fqComponent to backup host [$backupHost] failed."
         summaryExitCode=1
      fi
   else
      errmsg "Unknown failover type ($failoverType), aborting failover of component $fqComponent."
      return 1
   fi

   return $summaryExitCode
}

#------------------------------------------------------------------------------
# Function: failover_p4d_master ($path, $fqComponent, $style)
#------------------------------------------------------------------------------
function failover_p4d_master () {
   vvmsg "CALL: failover_p4d_master ($*)"
   declare path=${1:-Unset}
   declare fqComponent=${2:-Unset}
   declare style=${3:-Unset}

   declare masterHost=
   declare backupHost=
   declare instance=
   declare fqPath=
   declare newServerID=
   declare -i exitCode=0 summaryExitCode=0

   [[ $path == Unset || $fqComponent == Unset || $style == Unset ]] && return 1

   instance=${fqComponent%%:*}
   fqPath=$instance:$path
   masterHost=${ComponentMasterHost[$fqComponent]:-Unset}
   backupHost=${FailoverBackupHost[$fqPath]:-Unset]}
   failoverType=${FailoverType[$fqPath]:-Unset}

   if [[ $failoverType == Local ]]; then
      runRemoteCmd "$masterHost" \
         "$HMS_SCRIPTS/failover_p4d_local.sh $instance $style $NO_OP" \
       "Failing over to offline dbs locally on $masterHost. Running: failover_p4d_local.sh I=$instance S=$style NO_OP=$NO_OP" 1 1 0
      exitCode=$RCMDEXITCODE
      summaryExitCode=$exitCode
      if [[ $exitCode -eq 0 ]]; then
         msg "Local Failover on $masterHost completed OK."
      else
         errmsg "Local Failover on $masterHost FAILED."
         return 1
      fi
   elif [[ $failoverType == MO || $failoverType == Full ]]; then
      # The difference between Scheduled and Unscheduled failover is partly one
      # of expectations.  In Unscheduled Failover, we don't expect the master
      # to respond.  But in either case, we do our best to avoid split brain,
      # and try to shut it down.
      if [[ $style == Scheduled ]]; then
         msg "Shutting down master server for Scheduled Failover.  We expect it to shutdown cleanly." 
      else
         msg "Attempting shutdown of master server for Unscheduled Failover.  It may or may not respond."
      fi

      runRemoteCmd "$masterHost" \
         "$HMS_SCRIPTS/failover_p4d_from_this_host.sh $instance $style $failoverType $backupHost $NO_OP" \
       "Failing over from $masterHost. Running: failover_p4d_from_this_host.sh I=$instance S=$style T=$failoverType BH=$backupHost NO_OP=$NO_OP" 1 1 0
      exitCode=$RCMDEXITCODE
      if [[ $exitCode -ne 0 ]]; then
         if [[ $style == Scheduled ]]; then
            errmsg "Shutdown of master server FAILED during Scheduled Failover. Aborting failover of $fqComponent."
            return 1
         else
            warnmsg "Shutdown of master server failed during Unscheduled Failover.  Continuing with failover of $fqComponent."
         fi
      fi

      # newServerID is the server spec we want the master server to have.  This is usually
      # 'master', but may have some other value.
      newServerID=${fqComponent#*:}

      runRemoteCmd "$backupHost" \
         "$HMS_SCRIPTS/failover_p4d_to_this_host.sh $instance $style $failoverType $newServerID $NO_OP" \
         "Failing over to $backupHost. Running: $HMS_SCRIPTS/failover_p4d_to_this_host.sh I=$instance S=$style T=$failoverType SID=$newServerID NO_OP=$NO_OP" 1 1 0
      exitCode=$RCMDEXITCODE
      if [[ $exitCode -ne 0 ]]; then
         errmsg "Failover of $fqComponent to backup host [$backupHost] failed."
         summaryExitCode=1
      fi
   else
      errmsg "Unknown failover type ($failoverType), aborting failover of component $fqComponent."
      return 1
   fi

   return $summaryExitCode
}

#------------------------------------------------------------------------------
# Function: failover ($path, $scope, $style)
#------------------------------------------------------------------------------
function failover () {
   vvmsg "CALL: failover ($*)"
   declare path=${1:-Unset}
   declare scope=${2:-Unset}
   declare style=${3:-Unset}

   declare scopeType=
   declare scopeHost=
   declare scopeInstance=
   declare fqPath=
   declare instance=
   declare instanceList=
   declare iPath=
   declare pathActive=
   declare failoverType=
   declare masterHost=
   declare backupHost=
   declare edgeHost=
   declare reply=Unset
   declare -i found
   declare -i status=0

   # Validate instance scope.
   if [[ $scope == "i:"* ]]; then
      scopeType=Instance
      scopeInstance=${scope#i:}
      found=0; for instance in ${!InstanceManaged[*]}; do
         if [[ $scopeInstance == $instance ]]; then
            found=1
            break
         fi
      done

      if [[ $found -eq 0 ]]; then
         errmsg "Invalid SDP instance specified with scope of $scope."
         return 1
      fi

   # Validate host scope.
   elif [[ $scope == "h:"* ]]; then
      scopeType=Host
      scopeHost=${scope#h:}
      found=0; for host in ${FailoverMasterHost[*]}; do
         if [[ $scopeHost == $host ]]; then
            found=1
            break
         fi
      done

      if [[ $found -eq 0 ]]; then
         errmsg "Invalid failover host specified with scope of $scope."
         return 1
      fi
   else
      errmsg "Badly formatted scope value [$scope]."
      return 1
   fi

   if [[ $scopeType == Instance ]]; then
      msg "Verified: Scope instance [$scopeInstance] is valid."
   else
      msg "Verified: Scope host [$scopeHost] is valid."
   fi

   # Verify the specified path is valid and active for each instance.
   if [[ $scopeType == Host ]]; then
      msg "Finding instances mastered on host $scopeHost."
      found=0; for instance in ${!InstanceManaged[*]}; do
         fqPath="$instance:$path"
         pathActive=${FailoverActive[$fqPath]:-Unset}

         instanceMasterHost="${InstanceMasterHost[$instance]:-Unset}"

         [[ $scopeHost == $instanceMasterHost ]] || continue

         if [[ $pathActive == 1 ]]; then
            msg "Verified: Failover path [$path] is active for instance [$instance]."
         elif [[ $pathActive == 0 ]]; then
            warnmsg "Failover path [$path] is inactive for instance [$instance]. Skipping it."
            continue
         else
            errmsg "No failover path [$path] is configured for instance [$instance]. Skipping it."
            status=1
            continue
         fi

         # Next, confirm we have a configured target failover host (unless
         # FailoverType is 'local').
         failoverType=${FailoverType[$fqPath]:-Unset}

         if [[ $failoverType =~ ^(Full|MO)$ ]]; then
            backupHost=${FailoverBackupHost[$fqPath]:-Unset]}
            if [[ $backupHost == Unset ]]; then
               errmsg "No backup host configured for instance:path [$fqPath]. Skipping it."
               status=1
               continue
            else
               msg "Instance $instance will failover from $masterHost to $backupHost."
            fi
         elif [[ $failoverType =~ ^(EdgeFull|EdgeMO)$ ]]; then
            backupHost=${FailoverBackupHost[$fqPath]:-Unset]}
            edgeHost=${FailoverMasterHost[$fqPath]:-Unset]}
            if [[ $backupHost == Unset ]]; then
               errmsg "No backup host configured for instance:path [$fqPath]. Skipping it."
               status=1
               continue
            else
               msg "Instance $instance will failover from $edgeHost to $backupHost."
            fi
         elif [[ $failoverType == Local ]]; then
            msg "Local Failover will use offline databases for instance $instance on host [$masterHost]."
            backupHost=localhost
         elif [[ $failoverType == EdgeLocal ]]; then
            msg "Edge Local Failover will use offline databases for instance $instance on host [$edgeHost]."
            backupHost=localhost
         else
            errmsg "Could not determine failover type for [$fqPath]. Skipping it."
            status=1
            continue
         fi

         if [[ $found -eq 0 ]]; then
            found+=1
            instanceList="$instance"
         else
            instanceList+=" $instance"
         fi
      done

      if [[ $found -gt 0 ]]; then
         msg "$found instance(s) mastered on $scopeHost are targeted for failover: $instanceList."
      else
         errmsg "No instances mastered on host $scopeHost. Aborting failover."
         return 1
      fi
   else
      instance=$scopeInstance
      instanceList="$instance"
      fqPath="$instance:$path"

      pathActive=${FailoverActive[$fqPath]:-Unset}

      if [[ $pathActive == 1 ]]; then
         msg "Verified: Failover path [$path] is active for instance [$instance]."
      elif [[ $pathActive == 0 ]]; then
         errmsg "Failover path [$path] is inactive for instance [$instance]. Aborting."
         return 1
      else
         errmsg "No failover path [$path] is configured for instance [$instance]. Aborting."
         return 1
      fi

      failoverType=${FailoverType[$fqPath]:-Unset}
      masterHost=${InstanceMasterHost[$instance]:-Unset}

      if [[ $masterHost == Unset ]]; then
         errmsg "No master host is configured for instance [$instance]. Aborting."
         return 1
      fi

      if [[ $failoverType =~ ^(Full|MO)$ ]]; then
         backupHost=${FailoverBackupHost[$fqPath]:-Unset]}
         if [[ $backupHost == Unset ]]; then
            errmsg "No backup host configured for instance:path [$fqPath]. Aborting."
            return 1
         else
            msg "Instance $instance will failover from $masterHost to $backupHost."
         fi
      elif [[ $failoverType =~ ^(EdgeFull|EdgeMO)$ ]]; then
         backupHost=${FailoverBackupHost[$fqPath]:-Unset]}
         edgeHost=${FailoverMasterHost[$fqPath]:-Unset]}
         if [[ $backupHost == Unset ]]; then
            errmsg "No backup host configured for instance:path [$fqPath]. Aborting."
            return 1
         else
            msg "Instance $instance will failover from $edgeHost to $backupHost."
         fi
      elif [[ $failoverType == Local ]]; then
         msg "Local Failover will use offline databases for instance $instance on host [$masterHost]."
         backupHost=localhost
      elif [[ $failoverType == EdgeLocal ]]; then
         msg "Edge Local Failover will use offline databases for instance $instance on host [$edgeHost]."
         backupHost=localhost
      else
         errmsg "Could not determine failover type for [$fqPath]. Skipping it."
         return 1
      fi
   fi

   # Confirm user intent.
   if [[ $Interactive -eq 1 ]]; then
      while [[ $reply == Unset ]]; do
         echo -n -e "\nConfirm your intent to initiate failover: [Y/y/N/n]: "
         read -e reply
         [[ ${reply^^} == Y || ${reply^^} == N ]] && break
         msg "\nInvalid input specified ('$reply'), specify Y or N only.\n"
         reply=Unset
      done
   else
      msg "Failover intent confirmed non-interactively with '-y'."
      reply=Y
   fi

   if [[ ${reply^^} == N ]]; then
      msg "Failover not confirmed.  Failover aborted."
      return 1
   fi

   msg "Starting Failover for SDP instance(s): $instanceList."
   for instance in $instanceList; do
      if [[ "$failoverType" != "Edge"* ]]; then
         execute_instance_failover "$instance" "$path" "$style"
      else
         execute_edge_failover "$instance" "$path" "$style" "$edgeHost"
      fi
   done

   return $status
}

#------------------------------------------------------------------------------
# Function: get_component_list ($target, $onlyManaged)
#
# Given a target as specified on the command line, return a valid list of
# components, or return 1 if we fail to return a valid list of components.
#
# The target may look like: all|<instance>[:<component>]
#
# If called with a specified component, i.e. something like 1:p4d01, then
# show a warning if that component is not managed.
#
# If called with 'all' or an instance name, get the list of all defined
# components (globally or for the specified instance), regardless of whether
# they are managed.
#
# Output: The output to stdout is the valid component list. Any warnings or
# errors are sent to stderr.  An exit code of 0 indicates valid components
# were returned.
#
# Sample Usage: list=$(get_component_list "abc") || bail "No valid instances!"
# 
#------------------------------------------------------------------------------
function get_component_list () {
   declare target=${1:-Unset}
   declare onlyManaged=${2:-0}
   declare component=
   declare fqComponent=
   declare instanceComponents=
   declare componentList=
   declare componentManaged=

   [[ $target == Unset ]] && return 1

   if [[ $target == "all" ]]; then
      componentList=""
      for component in ${!ComponentManaged[*]}; do
         # Build a list of all define components, or only managed ones,
         # depending on whether $onlyManaged.
         if [[ $onlyManaged -eq 1 ]]; then
            componentManaged=${ComponentManaged[$component]:-Invalid}
            if [[ $componentManaged == 1 ]]; then
               componentList+=" $component"
            elif [[ $componentManaged == Invalid ]]; then
               errnmsg "Component $target is invalid." >&2
            fi
         else
            componentList+=" $component"
         fi
      done
      componentList=$(echo $componentList)
   elif [[ $target == *":"* ]]; then
      # If target contains a colon, user specified a single component
      # named as "instance:component".
      componentList=$target
      componentManaged=${ComponentManaged[$target]:-Invalid}
      if [[ $componentManaged != Invalid ]]; then
         if [[ $componentManaged -eq 0 ]]; then
            warnmsg "Component $target is not managed by HMS." >&2
         fi
      else
         errmsg "The specified component $target is invalid." >&2
         return 1
      fi
   else
      # Get all components for a given instance.  In this case, the
      # $target was specified as just an SDP instance name.
      instanceComponents=${InstanceComponents[$target]:-Unset}
      if [[ $instanceComponents == Unset ]]; then
         errmsg "No components configured for instance $target." >&2
         return 1
      fi

      componentList=
      for component in $instanceComponents; do
         fqComponent=$target:$component

         # Build a list of all define components, or only managed ones,
         # depending on whether $onlyManaged.
         if [[ $onlyManaged -eq 1 ]]; then
            componentManaged=${ComponentManaged[$fqComponent]:-Invalid}
            if [[ $componentManaged == 1 ]]; then
               componentList+=" $fqComponent"
            elif [[ $componentManaged == Invalid ]]; then
               errnmsg "Component $fqComponent is invalid." >&2
            fi
         else
            componentList+=" $fqComponent"
         fi
      done
      componentList=$(echo $componentList)
   fi

   if [[ -n "$componentList" ]]; then
      echo $componentList
   else
      return 1
   fi

   return 0
}

#------------------------------------------------------------------------------
# Function: get_component_status ($component, $optimistic, $tag)
#
# Input:
# $1 - A fully qualified component name of the form <Instance>:<ComponentName>
#
# $2 - "Optimistic" setting, 1 or 0 (the default).  By default, components that
# don't yet have status logic coded report as failed.  If 1, components of
# known types report as OK.  Components of unknown type always report as
# failed.  (Any such components should be marked inactive in the topology
# config file).
#
# $3 - optional tag for broker components, e.g. 'dfm'.
#
# Exit Code:
# Returns a "happy" '0' if component is up/online. Otherwise, returns one of
# the following:
# 1  Component down.
# 2  Component offline (applies to broker only).
# 11 Status not yet implemented for this component type.
# 12 Unknown component type.
# 13 Component host not configured.
# 99 Unknown (internal error).
#
# Output: In all cases, a message is displayed indicating component status.
#
#------------------------------------------------------------------------------
function get_component_status () {
   vvmsg "CALL: get_component_status ($*)"
   declare component=${1:-Unset}
   declare optimistic=${2:-0}
   declare tag=${3:-default}
   declare instance=
   declare componentType=
   declare componentHost=
   declare componentURL=
   declare componentP4PORT=
   declare componentStatusCmd=
   declare -i componentStatusCode=
   declare componentStatusMsg=
   declare componentVersion=
   declare brokerStatusFile=$P4U_TMPDIR/broker_status.${component/:/_}.out

   [[ $component == Unset ]] && return 1

   get_component_version "$component"
   componentVersion=${ComponentVersion[$component]}

   instance=${component%%:*}
   componentType=${ComponentType[$component]:-Unset}
   componentURL=${ComponentURL[$component]:-Unset}
   componentManaged=${ComponentManaged[$component]:-Unset}
   componentMasterHost=${ComponentMasterHost[$component]:-Unset}

   if [[ "$componentMasterHost" == Unset ]]; then
      errmsg "No master host configured for component $component."
      componentStatusCode=13
   elif [[ $componentManaged == 1 ]]; then
      case "$componentType" in
         (p4d-mc|p4d-e|p4d-r)
            if [[ $componentURL == "ssl:"* ]]; then
                componentP4PORT="ssl:${ComponentMasterHost[$component]:-Unset}:${componentURL#ssl:}"
            else
                componentP4PORT="${ComponentMasterHost[$component]:-Unset}:$componentURL"
            fi

            componentStatusCmd="$P4CBIN/irun -i hms -L off -v1 $P4BIN -s -p $componentP4PORT info -s"
            componentStatusCode=0
            run "$componentStatusCmd" \
               "3:Checking status for $component of type $componentType.|4:Checking with command: $componentStatusCmd" 0 0 ||\
               componentStatusCode=1
         ;;
         (p4broker)
            componentStatusCmd="/p4/${instance}/bin/p4broker_${instance}_init status $tag"
            msg "Checking status of $component on $componentMasterHost"
            vmsg "Executing: ssh -q -n -l $OSUSER $componentMasterHost \"$componentStatusCmd\""
            ssh -q -n -l $OSUSER $componentMasterHost "$componentStatusCmd" > $brokerStatusFile 2>&1
            [[ $VERBOSITY -gt 3 ]] && cat $brokerStatusFile

            fgrep "is running as" $brokerStatusFile > /dev/null 2>&1
            if [[ $? -eq 0 ]]; then
               componentStatusCode=0
               msg "Verified: $component is OK."
            else
               fgrep "is NOT running" $brokerStatusFile > /dev/null 2>&1
               if [[ $? -eq 0 ]]; then
                  componentStatusCode=2
                  msg "Verified: $component is offline."
               else
                  componentStatusCode=1
                  errmsg "Component $component status unknown."
               fi
            fi
            rm -f $brokerStatusFile
         ;;
         (p4p|p4dtg|p4web|p4gf)
            warnmsg "Sorry, can't status components of type $componentType just yet."
            componentStatusCode=11
            [[ $optimistic -eq 1 ]] && componentStatusCode=0
         ;;
         (swarm)
            warnmsg "Sorry, can't status components of type swarm just yet."
            msg "Check the Swarm worker queue with wget using the configured URL ($componentURL)."
            componentStatusCode=11
            [[ $optimistic -eq 1 ]] && componentStatusCode=0
         ;;
         (*)
            warnmsg "Unknown component type ($componentType) configured."
            componentStatusCode=11
         ;;
         (Unset)
            errmsg "ComponentType not defined for component $component."
            componentStatusCode=12
         ;;
      esac
   else
      warnmsg "Component $component is not managed by HMS.  Assuming it is OK."
      componentStatusCode=0
   fi

   case "$componentStatusCode" in
      (0) componentStatusMsg=OK;;
      (1) componentStatusMsg=FAILED;;
      (2) componentStatusMsg=Offline;;
      (11) componentStatusMsg="Status not yet implemented for this component type.";;
      (12) componentStatusMsg="Unknown component type.";;
      (13) componentStatusMsg="Component host not configured.";;
      (*) componentStatusMsg="Unknown Status Code [$componentStatusCode]"; componentStatusCode=99;
   esac

   ComponentStatusCode[$component]=$componentStatusCode
   ComponentStatusMsg[$component]=$componentStatusMsg

   if [[ "$componentType" == p4broker ]]; then
      msg "Component $component v$componentVersion ($tag config) status is $componentStatusMsg (code $componentStatusCode)."
   else
      msg "Component $component v$componentVersion status is $componentStatusMsg (code $componentStatusCode)."
   fi

   return $componentStatusCode
}

#------------------------------------------------------------------------------
# Function: get_component_version ($component)
#
# Set values in ComponentMajorVersion, ComponentMinorVersion, and
# ComponentVersion associative arrays, indexed by the fully qualified
# component name (i.e. <instance>:<compoment_name>).
#
# Output: This get_component_version() is a quiet function.  It does its work
# silently.  In event of problems, it just sets the version values to UNKNOWN.
# It doesn't complain.  This is called by get_component_status, which takes
# care of notifying the user of problems.
#------------------------------------------------------------------------------
function get_component_version () {
   vvmsg "CALL: get_component_version ($*)"
   declare component=${1:-Unset}

   declare componentMasterHost=
   declare componentType=
   declare componentExe=
   declare componentMajorVersion=
   declare componentMinorVersion=
   declare componentVersionFile=$P4U_TMPDIR/component_version.${component/:/_}.out
   declare componentVersionCmd=

   instance=${component%%:*}
   componentType=${ComponentType[$component]:-Unset}
   componentManaged=${ComponentManaged[$component]:-Unset}
   componentMasterHost=${ComponentMasterHost[$component]:-Unset}

   case "$componentType" in
      (p4d-*|p4broker|p4p)
         if [[ $componentURL == "ssl:"* ]]; then
            componentP4PORT="ssl:${ComponentMasterHost[$component]:-Unset}:${componentURL#ssl:}"
         else
            componentP4PORT="${ComponentMasterHost[$component]:-Unset}:$componentURL"
         fi
         componentExe="/p4/$instance/bin/${componentType%-*}_${instance}"

         ssh -q -l $OSUSER $componentMasterHost "$componentExe -V" > $componentVersionFile 2>&1

         componentMajorVersion=$(fgrep "Rev. " $componentVersionFile 2>/dev/null)
         if [[ -n "$componentMajorVersion" ]]; then
            componentMinorVersion=$(echo $componentMajorVersion|cut -d '/' -f 4)
            componentMinorVersion=${componentMinorVersion%% *}
            componentMajorVersion=$(echo $componentMajorVersion|cut -d '/' -f 3)
         else
            componentMajorVersion="UNKNOWN"
            componentMinorVersion="UNKNOWN"
         fi
      ;;
      (p4dtg|p4web|p4gf)
         warnmsg "Sorry, can't get version for component of type $componentType just yet."
            componentMajorVersion="UNKNOWN"
            componentMinorVersion="UNKNOWN"
      ;;
      (swarm)
         ssh -q -l $OSUSER $componentMasterHost "cat /p4/$instance/swarm/Version" > $componentVersionFile 2>&1

         # A Swarm version file has 3 lines and looks like this:
         # RELEASE = 2016 1 ;
         # PATCHLEVEL = 1400259 ;
         # SUPPDATE = 2016 06 13 ;

         # Extract the RELEASE and PATCHLEVEL and normalized to our standard format.

         componentMajorVersion=$(fgrep RELEASE $componentVersionFile 2>/dev/null)
         if [[ -n "$componentMajorVersion" ]]; then
            componentMajorVersion=${componentMajorVersion##* = }
            componentMajorVersion=${componentMajorVersion% ;}
            componentMajorVersion=${componentMajorVersion/ /.}
         else
            componentMajorVersion="UNKNOWN"
         fi

         componentMinorVersion=$(fgrep PATCHLEVEL $componentVersionFile 2>/dev/null)
         if [[ -n "$componentMinorVersion" ]]; then
            componentMinorVersion=${componentMinorVersion##* = }
            componentMinorVersion=${componentMinorVersion% ;}
         else
            componentMinorVersion="UNKNOWN"
         fi

         rm -f $componentVersionFile
      ;;
   esac

   ComponentMajorVersion[$component]=$componentMajorVersion
   ComponentMinorVersion[$component]=$componentMinorVersion
   if [[ $componentMajorVersion != UNKNOWN ]]; then
      ComponentVersion[$component]="$componentMajorVersion.$componentMinorVersion"
   else
      ComponentVersion[$component]=UNKNOWN
   fi
}