mkrep.sh #14

#!/bin/bash
#==============================================================================
# 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
#------------------------------------------------------------------------------

#==============================================================================
# Declarations and Environment

if [[ ${BASH_VERSINFO[0]} -lt 4 ]]; then
   echo -e "\\n\\nERROR: $0 requires bash version 4.x+; current bash version is $BASH_VERSION."
   exit 1
fi

# Allow override of P4U_HOME, which is set only when testing P4U scripts.
export P4U_HOME=${P4U_HOME:-/p4/common/bin}
export SDP_ENV=${SDP_ENV:-/p4/common/bin/p4_vars}
export P4U_LIB=${P4U_LIB:-/p4/common/lib}
export P4U_ENV=$P4U_LIB/p4u_env.sh
export P4U_LOG=Unset
export VERBOSITY=${VERBOSITY:-3}
declare SSH_Opts="-oBatchMode=yes"
declare RandomPassword=

# Environment isolation.  For stability and security reasons, prepend
# PATH to include dirs where known-good scripts exist.
# known/tested PATH and, by implication, executables on the PATH.
export PATH=$P4U_HOME:$PATH:~/bin:.
export P4CONFIG=${P4CONFIG:-.p4config}
export P4ENVIRO=/dev/null/.p4enviro

[[ -r "$P4U_ENV" ]] || {
   echo -e "\\nError: Cannot load environment from: $P4U_ENV\\n\\n"
   exit 1
}

declare BASH_LIBS=$P4U_ENV
BASH_LIBS+=" $P4U_LIB/libcore.sh"
BASH_LIBS+=" $P4U_LIB/libp4u.sh"

for bash_lib in $BASH_LIBS; do
   # shellcheck disable=SC1090
   source "$bash_lib" ||\
      { echo -e "\\nFATAL: Failed to load bash lib [$bash_lib]. Aborting.\\n"; exit 1; }
done

declare Version=2.2.0
declare -i SilentMode=0
declare -i StartupCmdNumFirst StartupCmdNumLast
declare -i DoSSHChecks=1
RandomPassword=$(date +%s | sha256sum | base64 | head -c 32)
declare JournalPrefix=
export VERBOSITY=3

#==============================================================================
# Local Functions

#------------------------------------------------------------------------------
# Function: terminate
function terminate
{
   # Disable signal trapping.
   trap - EXIT SIGINT SIGTERM

   vvmsg "$THISSCRIPT: EXITCODE: $OverallReturnStatus"

   # Stop logging.
   [[ "${P4U_LOG}" == off ]] || stoplog

   # Don't litter.
   cleanTrash

   # With the trap removed, exit.
   exit "$OverallReturnStatus"
}

function bail {
   declare msg="${1:-Unknown Error}"
   declare -i rc

   rc="${2:-1}"
   echo -e "\\n$THISSCRIPT (line: ${BASH_LINENO[0]}): FATAL: $msg\\n\\n" >&2

   exit "$rc"
}

#------------------------------------------------------------------------------
# Function: usage (required function)
#
# Input:
# $1 - style, either -h (for short form) or -man (for man-page like format).
#------------------------------------------------------------------------------
function usage
{
   declare style=${1:--h}

   echo "USAGE for $THISSCRIPT v$Version:

$THISSCRIPT -i <SDP_Instance> -t <Type> -s <Site_Tag> -r <Replica_Host> [-p] [-ssh_opts=\"opts\"] [-skip_ssh] [-L <log>] [-si] [-v<n>] [-n] [-D]

or

$THISSCRIPT [-h|-man|-V]
"
   if [[ $style == -man ]]; then
      echo -e "
DESCRIPTION:
        This script creates makes a replica, and provides enough information to
        make it ready in all respects.

OPTIONS:
 -i <SDP_Instance>
        Specify the SDP Instance.

 -t <Type>
        Specify the replica type tag.  The type corresponds to the 'Type:' and
        'Services:' field of the server spec, which describes the type of services
        offered by a given replica.

        Valid values are:
        * ha:   High Availability mandatory standby replica, for 'p4 failover' (P4D 2018.2+)
        * ham:  High Availability metadata-only mandatory standby replica, for 'p4 failover' (P4D 2018.2+)
        * ro:   Read-Only standby replica.
        * rom:  Read-Only standby replica, Metadata only.
        * fr:   Forwarding Replica (Unfiltered).
        * fs:   Forwarding Standby (Unfiltered).
        * frm:  Forwarding Replica (Unfiltered,  Metadata only).
        * fsm:  Forwarding Standby (Unfiltered,  Metadata only).
        * ffr:  Filtered Forwarding Replica.  Not a valid failover target.
        * edge: Edge Server. Filtered by definition.

        Replicas with 'standby' are always unfiltered, and use the 'journalcopy'
        method of replication, which copies a byte-for-byte verbatim journal file
        rather than one that is merely logically equivalent.

        The tag has several purposes:
        1. Short Hand.  Each tag represents a combination of 'Type:' and fully
        qualified 'Services:' values used in server specs.

        2. Distillation. Only the most useful Type/Services combinations have a
        shorthand form.

        3. For forwarding replicas, the name includes the critical distinction of
        whether any replication filtering is used; as filtering of any kind disqualifies
        a replica from being a potential failover target.  (No such distinction is
        needed for edge servers, which are filtered by definition).

 -s <Site_Tag>
        Specify a geographic site tag indicating where the replica will physically be
        located.  Valid site tags are defined in the site tags file,
        $SiteTagsFile

        Current valid site tags are:
        $(grep -v '^#' "$SiteTagsFile" 2>&1|grep -v '$^'|sed 's:^:\t:g')

 -r <Replica_Host>
        Specify the target replica host.

 -p        This script performs a check to ensure that the Protections table grants
        super access to the group $ServiceUsersGroup.

        By default, an error is displayed if the check fails, i.e. if super user
        access for the group $ServiceUsersGroup cannot be verified.  This is
        becuase, by default, we want to avoid making changes to the Protections
        table in case the customer has local policies or custom automation that
        requires site-specific procedures to update the Protections table.

        If '-p' is specified, an attempt is made to append the Protections table
        an entry like:

        super group $ServiceUsersGroup * //...

 -ssh_opts=\"opts\"
        Specify '-ssh_opts' to pass paraemters on to the ssh command.  For example, to
        specify ssh operation on non-standard port 2222, specify '-ssh_opts=\"-p 2222\"'.

 -skip_ssh
        Specify '-skip_ssh' to skip SSH checks duing the preflight check.

        This is useful if you only intend to do the metadata prepaation phase
        of startup prior to SSH being setup.

 -v<n>        Set verbosity 1-5 (-v1 = quiet, -v5 = highest).

 -L <log>
        Specify the path to a log file, or the special value 'off' to disable
        logging.  By default, all output (stdout and stderr) goes in the logs
	directory referenced by \$LOGS.

        NOTE: This script is self-logging.  That is, output displayed on the screen
        is simultaneously captured in the log file.  Do not run this script with
        redirection operators like '> log' or '2>&1', and do not use 'tee.'

-si        Operate silently.  All output (stdout and stderr) is redirected to the log
        only; no output appears on the terminal.  This cannot be used with
        '-L off'.
      
 -n        No-Op.  Prints commands instead of running them.

 -D     Set extreme debugging verbosity.

 -f        Full Mode Setup:  The completes an edge servers setup so no additional steps 
        are required.  This setup requires an ssh connection from the master to the 
        edge to be in place first.  It also requires the depot log journal and /p4
        mounts to be in place and setup as expected.  This setup assumes a standard
        SDP setup.

HELP OPTIONS:
 -h        Display short help message
 -man        Display man-style help message
 -V        Dispay version info for this script and its libraries.

DEPENDENCIES:
        This script depends on ssh keys being defined to allow the Perforce
        operating system user ($OSUSER) to ssh to any necessary machines
        without a password.

        This script assumes the replica host already has the SDP fully
        configured.

FILES:
        This Site Tags file defines the list of valid geographic site tags:
        $SiteTagsFile

EXAMPLES:
        Prepare an edge server to run on host syc-helix-04:
        $THISSCRIPT -i acme -t edge -s syd -r syc-helix-04
"
   fi

   exit 1
}

#==============================================================================
# Command Line Processing

declare ReplicaHost=Unset
declare ReplicaTypeTag=Unset
declare ReplicaType=
declare SiteTag=Unset
declare SiteTagsFile=${P4CCFG:-/p4/common/config}/SiteTags.cfg
declare SDPInstance=Unset
#declare -i Interactive=1
declare -i MetadataOnly=0
declare -i shiftArgs=0
declare -i UpdateProtections=0

set +u
while [[ $# -gt 0 ]]; do
   case $1 in
      (-h) usage -h;;
      (-man) usage -man;;
      (-r) ReplicaHost=$2; shiftArgs=1;;
      (-t) ReplicaTypeTag=$2; shiftArgs=1;;
      (-i) SDPInstance=$2; shiftArgs=1;;
      (-s) SiteTag=$2; shiftArgs=1;;
      (-p) UpdateProtections=1;;
      (-V) show_versions; exit 1;;
      (-v1) export VERBOSITY=1;;
      (-v2) export VERBOSITY=2;;
      (-v3) export VERBOSITY=3;;
      (-v4) export VERBOSITY=4;;
      (-v5) export VERBOSITY=5;;
      (-ssh_opts) SSH_Opts="$2"; shiftArgs=1;;
      (-skip_ssh) DoSSHChecks=0;;
      (-L) export P4U_LOG="$2"; shiftArgs=1;;
      (-si) SilentMode=1;;
      (-n) export NO_OP=1;;
      (-D) set -x;; # Debug; use 'set -x' mode.
      (*) usageError "Unknown arg ($1).";;
   esac

   # Shift (modify $#) the appropriate number of times.
   shift; while [[ $shiftArgs -gt 0 ]]; do
      [[ $# -eq 0 ]] && usageError "Bad usage."
      shiftArgs=$shiftArgs-1
      shift
   done
done
set -u

#==============================================================================
# Command Line Verification

[[ $SilentMode -eq 1 && "$P4U_LOG" == off ]] && \
   usageError "Cannot use '-si' with '-L off'."

[[ "$P4U_LOG" == Unset ]] && \
   P4U_LOG=${LOGS:-/tmp}/mkrep.$(date +'%Y%m%d-%H%M').log

[[ $SDPInstance == Unset ]] && usageError "\\nThe '-i <SDP_Instance>' parameter is required."
[[ $ReplicaHost == Unset ]] && usageError "\\nThe '-r <Replica_Host>' parameter is required."
[[ $ReplicaTypeTag == Unset ]] && usageError "\\nThe '-t <Type>' parameter is required."
[[ $SiteTag == Unset ]] && usageError "\\nThe '-s <Site_Tag>' parameter is required."

# Set log file to point at SDP_INSTANCE.
P4U_LOG="/p4/${SDPInstance}/logs/mkrep.$(date +'%Y%m%d-%H%M%S').log"

case "$ReplicaTypeTag" in
   (ha) ReplicaType=standby;;                             # HA Mandatory Standby replica.
   (ham) ReplicaType=standby; MetadataOnly=1;;            # HA Mandatory Standby replica.
   (ro) ReplicaType=standby;;                             # Read-Only Standby replica.
   (rom) ReplicaType=standby; MetadataOnly=1;;            # Read-Only Standby replica, Metadata only.
   (fr) ReplicaType=forwarding-replica;;                  # Forwarding Replica (Unfiltered).
   (fs) ReplicaType=forwarding-standby;;                  # Forwarding Standby (Unfiltered).
   (frm) ReplicaType=forwarding-replica; MetadataOnly=1;; # Forwarding Replica (Unfiltered), Metdata only.
   (fsm) ReplicaType=forwarding-standby; MetadataOnly=1;; # Forwarding Standby (Unfiltered).
   (ffr) ReplicaType=forwarding-replica;;                 # Filtered Forwarding Replica
   (edge) ReplicaType=edge-server;;                       # Edge Server. Filtered by def'n, cannot be Metdata only.
   (*) usageError "The specified replica type tag [$ReplicaTypeTag] is invalid.";;
esac

declare -i tagFound=0
if [[ -r "$SiteTagsFile" ]]; then
   while read -r line; do
      [[ $line == "#*" ]] && continue
      # shellcheck disable=SC2086 disable=SC2116
      [[ -z "$(echo $line)" ]] && continue
      [[ "$line" == *":"* ]] || continue
      tag=${line%%:*}

      if [[ "$tag" == "$SiteTag" ]]; then
         tagFound=1
         break
      fi
   done < "$SiteTagsFile"
else
   bail "Missing site tag configuration file [$SiteTagsFile]. Aborting."
fi

[[ $tagFound -eq 1 ]] ||\
   bail "Failed to find specified site tag [$SiteTag] inite tag configuration file [$SiteTagsFile]. Aborting."

#==============================================================================
# Main Program

trap terminate EXIT SIGINT SIGTERM

declare -i OverallReturnStatus=0
declare ServerSpec=
declare ServerSpecFile=
declare ShortServerSpec=
declare ServiceUser=
declare ServiceUserSpecFile=
declare ServiceUsersGroup=ServiceUsers
declare TmpDir=$(mktemp -d)
declare ProtectsFile="$TmpDir/protect.p4s"
declare GroupSpecFile="$TmpDir/group.$ServiceUsersGroup.p4s"
declare TmpFile="$TmpDir/tmpFile.$THISSCRIPT"

GARBAGE+=" $TmpDir"

touch "${TmpFile}" || bail "Couldn't touch tmp file [$TmpFile]."

if [[ "${P4U_LOG}" != off ]]; then
   touch "${P4U_LOG}" || bail "Couldn't touch log file [${P4U_LOG}]."

   # Redirect stdout and stderr to a log file.
   if [[ "$SilentMode" -eq 0 ]]; then
      exec > >(tee "${P4U_LOG}")
      exec 2>&1
   else
      exec >"${P4U_LOG}"
      exec 2>&1
   fi

   initlog
fi

msg "Starting $THISSCRIPT v$Version at $(date)."

msg "${H}\\nPart 0: Environment Setup and Preflight Checks."

msg "Loading SDP environment for instance $SDPInstance."
# shellcheck disable=SC1090
source "$SDP_ENV" "$SDPInstance"
# shellcheck disable=SC1091
source /p4/common/bin/backup_functions.sh
check_vars
set_vars

# Version check: Require P4D 2018.1 for using 'standby' replica.
# shellcheck disable=SC2072
if [[ "$ReplicaType" == *"standby" && "$P4D_VERSION" > "2018.1" ]]; then
   msg "P4D is 2018.1+, as recommended for standby replicas."
elif [[ "$P4D_VERSION" > "2016.2" ]]; then
   msg "P4D is 2016.2+, as supported for $ReplicaType replicas."
else
   bail "P4D must be 2018.1+ if using journalcopy replicas, P4D_VERSION is $P4D_VERSION. Aborting."
fi

# Version check: Require P4D 2018.2 for using 'ha' replica, i.e. a 2018.2-style 'mandatory'
# standby replica with ReplicatingFrom field set.
if [[ "$ReplicaTypeTag" == "ha"* ]]; then
   # shellcheck disable=SC2072
   if [[ "$P4D_VERSION" > "2018.2" ]]; then
      msg "P4D is 2018.2+, as required for 'ha' type replicas that use 'p4 failover'."
   else
      bail "P4D must be 2018.2+ if using HA replicas, P4D_VERSION is $P4D_VERSION. Aborting."
   fi
fi

msg "Overriding P4CONFIG, P4ENVIRO, P4ALIASES, etc."
export P4ENVIRO=/dev/null/.p4enviro
export P4ALIASES=/dev/null/.p4aliases
export P4CONFIG="$TmpDir/.p4config"
export P4TICKETS="$TmpDir/.p4tickets"
export P4TRUST="$TmpDir/.p4trust"
export P4BIN="/p4/${SDPInstance}/bin/$P4SERVER"
echo "P4PORT=$P4MASTERPORT" > "$P4CONFIG"

run "$P4BIN -u $P4USER -p $P4MASTERPORT -s login -s" \
   "Verifying that user $P4USER can login to port $P4MASTERPORT." 1 1 0
if [[ $? -ne 0 ]]; then
   msg "Trying login with:\n$P4BIN -u $P4USER -p $P4MASTERPORT -s login -a < $SDP_ADMIN_PASSWORD_FILE\n"
   if [[ $NO_OP -eq 0 ]]; then
      $P4BIN -u $P4USER -p $P4MASTERPORT -s login -a < $SDP_ADMIN_PASSWORD_FILE
   else
      msg "NO_OP: Would execute: $P4BIN -u $P4USER -p $P4MASTERPORT -s login -a < $SDP_ADMIN_PASSWORD_FILE"
   fi
fi

if [[ $DoSSHChecks -eq 1 ]]; then
   msg "Checking ssh access to master host $P4MASTER."
   # shellcheck disable=SC2086
   ssh $SSH_Opts -q "$P4MASTER" /bin/ls > /dev/null 2>&1 || bail "Failed to ssh to host $P4MASTER."

   msg "${H}\\nCreating SSH Command."
   export SSH_CMD_R="ssh $SSH_Opts -q $ReplicaHost"
   export SSH_CMD_M="ssh $SSH_Opts -q $P4MASTER"
   vmsg "SSH_CMD_R=$SSH_CMD_R"
   vmsg "SSH_CMD_M=$SSH_CMD_M"

   msg "Checking ssh access to master host $P4MASTER."
   $SSH_CMD_M /bin/ls > /dev/null 2>&1 || bail "Failed to ssh to host $P4MASTER."
   msg "Verified: ssh access to master host $P4MASTER is OK."

   msg "Checking ssh access to replica host $ReplicaHost."
   $SSH_CMD_R /bin/ls > /dev/null 2>&1 || bail "Failed to ssh to host $ReplicaHost."
   msg "Verified: ssh access to replica host $ReplicaHost is OK."
else
   msg "Skipping SSH preflight checks due to -skip_ssh."
fi

#--------------------------------------------------------------
msg "${H}\\nPart 1: Preparation: Login."

if [[ "$P4MASTERPORT" == "ssl:"* ]]; then
   run "p4 trust -y"
fi

run "$P4BIN -u $P4USER -p $P4MASTERPORT -s login -s" \
   "Checking whether user $P4USER is logged into port $P4MASTERPORT." 1 1 0

if [[ "$CMDEXITCODE" -ne 0 ]]; then
   msg "Trying login with:\\n$P4BIN -u $P4USER -p $P4MASTERPORT -s login -a < $SDP_ADMIN_PASSWORD_FILE\\n"
   if [[ $NO_OP -eq 0 ]]; then
      "$P4BIN" -u "$P4USER" -p "$P4MASTERPORT" -s login -a < "$SDP_ADMIN_PASSWORD_FILE"
   else
      msg "NO_OP: Would execute: $P4BIN -u $P4USER -p $P4MASTERPORT -s login -a < $SDP_ADMIN_PASSWORD_FILE"
   fi
fi

run "$P4BIN -u $P4USER -p $P4MASTERPORT -s login -s" \
   "Verifying that user $P4USER can login to port $P4MASTERPORT." 1 1 0 ||\
   bail "Login verification failed."

#--------------------------------------------------------------
msg "${H}\\nPart 2: Define Server Spec."
#TODO Need to make ArchiveDataFilter appear here through some option
ShortServerSpec="${ReplicaTypeTag}_${SiteTag}"
ServerSpec="p4d_${ShortServerSpec}"

ServerSpecFile="$TmpDir/$ServerSpec.server.p4s"

# strip of ssl: or an host: element from front of value to just leave port - so strip twice
ReplicaPort=${P4MASTERPORT#*:}
ReplicaPort=${ReplicaPort#*:}

if [[ "$ReplicaTypeTag" == "ha"* ]]; then
   echo -e "ServerID: $ServerSpec\\n
Type: server\\n
Name: $ServerSpec\\n
Options: nomandatory\\n
ReplicatingFrom: master.${SDPInstance}\\n
Services: $ReplicaType\\n
ExternalAddress: $ReplicaHost:$ReplicaPort\\n
Description:" > "$ServerSpecFile" || bail "Failed to initialize server spec file [$ServerSpecFile]."

warnmsg "Server $ServerSpec is intended to have the Options: field value set to 'mandatory'.  However, it is initially configured as 'nomandatory' to prevent undue stalling of a global topology while a fresh new standby replica gets caught up.  As of 2019.1, new standby replicas cannot be made 'mandatory' until they are online.

After this replica is brought online and is seen to be replicating properly (and up-to-date), consider making it a 'mandatory' replica.

A 'mandatory' replica is one that you can trust is at least as current as all other replicas, and this is the optimal candidate for a smooth failover from the master server.  However, if the 'mandatory' replica fails, it cannot be easily ignored -- it stalls, the global topology does.

Should that ever occur, you can modify the server spec manually on the master server, changing the 'mandatory' value to 'nomandatory'. That will enable global replication to move on.  It should then be a high priority to figure out what went wrong with the replica so that it can be brought back online.\\n"

elif [[ "$ReplicaType" == *"standby" ]]; then
   echo -e "ServerID: $ServerSpec\\n
Type: server\\n
Name: $ServerSpec\\n
Options: nomandatory\\n
ReplicatingFrom: master.${SDPInstance}\\n
Services: $ReplicaType\\n
ExternalAddress: $ReplicaHost:$ReplicaPort\\n
Description:" > "$ServerSpecFile" || bail "Failed to initialize server spec file [$ServerSpecFile]."
else
   echo -e "ServerID: $ServerSpec\\n
Type: server\\n
Name: $ServerSpec\\n
Services: $ReplicaType\\n
ExternalAddress: $ReplicaHost:$ReplicaPort\\n
Description:" > "$ServerSpecFile" || bail "Failed to initialize server spec file [$ServerSpecFile]."
fi

case "$ReplicaTypeTag" in
   (ha) Desc="High Availability Mandatory Standby Replica (Unfiltered) in ${SiteTag^^}.";;
   (ham) Desc="High Availability Metadata-only Mandatory Standby Replica (Unfiltered) in ${SiteTag^^}.";;
   (ro) Desc="Read-Only Standby Replica (Unfiltered) in ${SiteTag^^}.";;
   (rom) Desc="Read-Only Standby Replica (Unfiltered, Metadata Only) in ${SiteTag^^}.";;
   (fr) Desc="Forwarding Replica (Unfiltered) in ${SiteTag^^}.";;
   (fs) Desc="Forwarding Standby Replica (Unfiltered) in ${SiteTag^^}.";;
   (frm) Desc="Forwarding Replica (Unfiltered, Metadata Only) in ${SiteTag^^}.";;
   (fsm) Desc="Forwarding Standby Replica (Unfiltered, Metadata Only) in ${SiteTag^^}.";;
   (ffr) Desc="Filtered Forwarding Replica in ${SiteTag^^}.";;
   (edge) Desc="Edge server in ${SiteTag^^}.";;
   (*) bail "\\nInternal Error: Unrecognized replica type tag [$ReplicaTypeTag].";;
esac

echo -e "\\t$Desc\\n" >> "$ServerSpecFile" || bail "Failed to complete server spec file [$ServerSpecFile]."

msg "Creating server spec $ServerSpec with these contents:"
msg "${H}"
cat "$ServerSpecFile"
msg "${H}"

if [[ $NO_OP -eq 0 ]]; then
   p4 server -i < "$ServerSpecFile" || bail "Failed to load server spec from file: $ServerSpecFile"
else
   msg "NO_OP: Would run: p4 server -i .LT. $ServerSpecFile"
fi

#--------------------------------------------------------------
msg "${H}\\nPart 3: Set configurables."

ServiceUser=svc_${ServerSpec}
ServiceUserSpecFile=$TmpDir/$ServiceUser.user.p4s
declare -i ConfigureOK=1

run "p4 configure set $ServerSpec#P4TARGET=$P4MASTERPORT" || ConfigureOK=0
run "p4 configure set $ServerSpec#db.replication=readonly" || ConfigureOK=0
run "p4 configure set $ServerSpec#rpl.forward.all=1" || ConfigureOK=0
run "p4 configure set $ServerSpec#rpl.compress=4" || ConfigureOK=0
run "p4 configure set $ServerSpec#server=4" || ConfigureOK=0
run "p4 configure set $ServerSpec#monitor=2" || ConfigureOK=0
run "p4 configure set $ServerSpec#serviceUser=$ServiceUser" || ConfigureOK=0

if [[ "$ReplicaType" == *"standby" ]] ; then
   run "p4 configure set $ServerSpec#rpl.journalcopy.location=1" || ConfigureOK=0
fi

# For master/commit servers, the journalPrefix is always:
# /p4/N/checkpoints/p4_N
#
# Regardless of what host that folder appears on, checkpoints
# from there will have originated on the master server.
#
# For replicas of all types, including edge servers, set journalPrefix
# to something like:
# /p4/N/checkpoints.bos_edge/p4_N.bos_edge
#
# where 'bos_edge' is a sample ServerID, less the 'p4d_' prefix.
# In the context of journalPrefix value and resulting checkpoint/journal
# file names, the leading 'p4d_' in the ServerID is redundant, and thus
# trimmed.
#
# The checkpoints and checkpoints.* folders are on the /hxdepots volume,
# which is always backed up.
#
# Incorporating the SeverID (server spec name) allows multiple
# "workspace servers" (edge servers used for horizontal scaling)
# to share storage on a SAN, while providing each edge a separate
# location for checkpoints.
#

# shellcheck disable=SC2153
JournalPrefix="$P4HOME/checkpoints.${ShortServerSpec}/p4_${SDPInstance}.${ShortServerSpec}"
run "p4 configure set $ServerSpec#journalPrefix=$JournalPrefix" || ConfigureOK=0

if [[ $NO_OP -eq 0 ]]; then
   # For 'journalcopy' replicas, i.e. those with 'Standby' in the name,
   # startup.1 is the 'journalcopy' command to pull the raw P4JOURNAL file from
   # the P4TARGET server, and startup.2 is a 'pull' command with the -L' flag
   # to replay P4JOURNAL records into the db.

   # With the SDP, the pulled journal appears as a file $LOGS/journal.N, where N
   # is the journal counter. The rpl.journalcopy.location=1 setting enables this
   # desired behavior.

   # For non-journalcopy replicas (including any filtered replicas, including
   # edge servers that are filtered by nature), startup.1 is a pull
   # command that both pulls journal chunks and replays them into the database.
   if [[ "$Desc" == *"Standby"* ]]; then
      vmsg "Executing: p4 configure set $ServerSpec#startup.1='journalcopy -i 0'"
      # shellcheck disable=SC2140
      p4 configure set "$ServerSpec#startup.1"="journalcopy -i 0" || ConfigureOK=0
      vmsg "Executing: p4 configure set $ServerSpec#startup.2='pull -i 1 -L'"
      # shellcheck disable=SC2140
      p4 configure set "$ServerSpec#startup.2"="pull -i 1 -L" || ConfigureOK=0
      StartupCmdNumFirst=3
      StartupCmdNumLast=7
   else
      vmsg "Executing: p4 configure set $ServerSpec#startup.1='pull -i 1'"
      # shellcheck disable=SC2140
      p4 configure set "$ServerSpec#startup.1"="pull -i 1" || ConfigureOK=0
      StartupCmdNumFirst=2
      StartupCmdNumLast=6
   fi
else
   if [[ "$Desc" == *"Standby"* ]]; then
      vmsg "NO_OP: Would execute: p4 configure set $ServerSpec#startup.1=\"journalcopy -i 0\""
      vmsg "NO_OP: Would execute: p4 configure set $ServerSpec#startup.2=\"pull -i 1 -L\""
      StartupCmdNumFirst=3
      StartupCmdNumLast=7
   else
      vmsg "NO_OP: Would execute: p4 configure set $ServerSpec#startup.1=\"pull -i 1\""
      StartupCmdNumFirst=2
      StartupCmdNumLast=6
   fi
fi

if [[ $MetadataOnly -eq 0 ]]; then
   run "p4 configure set $ServerSpec#lbr.replication=readonly" || ConfigureOK=0
   for i in $(seq $StartupCmdNumFirst $StartupCmdNumLast); do
      if [[ $NO_OP -eq 0 ]]; then
         vmsg "Executing: p4 configure set $ServerSpec#startup.$i='pull -i 1 -u'"
         # shellcheck disable=SC2140
         p4 configure set "$ServerSpec#startup.$i"="pull -i 1 -u" || ConfigureOK=0
      else
         vmsg "NO_OP: Would execute: p4 configure set $ServerSpec#startup.1='pull -i 1'"
      fi
   done
else
   run "p4 configure set $ServerSpec#lbr.replication=shared" || ConfigureOK=0
fi

if [[ $ConfigureOK -eq 1 ]]; then
   msg "Verified:  All configurables were set OK."
   run "p4 configure show allservers" "Showing all persistent configurables." 0 1 0
else
   bail "Errors encountered setting configurables.  See the output above. Aborting."
fi

#--------------------------------------------------------------
msg "${H}\\nPart 4: Create replica service user $ServiceUser."

echo -e "User: $ServiceUser\\n
Email: ${MAILFROM#\#}\\n
FullName: Replication Server User for $ServerSpec\\n
Type: service\\n
AuthMethod: perforce\\n" > "$ServiceUserSpecFile" || bail "Failed to initialize user spec file [$ServiceUserSpecFile]."

vmsg "Contents of $ServiceUserSpecFile:"
vmsg "${H}"
[[ $VERBOSITY -gt 2 ]] && cat "$ServiceUserSpecFile"
vmsg "${H}"

if [[ $NO_OP -eq 0 ]]; then
   p4 user -f -i < "$ServiceUserSpecFile" || \
      bail "Failed to load user spec from file: $ServiceUserSpecFile"
else
   msg "NO_OP: Would run: p4 user -f -i .LT. $ServiceUserSpecFile"
fi

PasswdFile="$P4CCFG/.p4passwd.$P4SERVER.$ServiceUser"
touch "$PasswdFile" || bail "Failed to initialize password file $PasswdFile."

chmod 600 "$PasswdFile"
echo "$RandomPassword" > "$PasswdFile"
echo "$RandomPassword" >> "$PasswdFile"

if [[ $NO_OP -eq 0 ]]; then
   msg "Setting password for service user $ServiceUser."
   p4 passwd "$ServiceUser" < "$PasswdFile"
else
   msg "NO_OP: Would run: p4 passwd $ServiceUser .LT. $PasswdFile"
fi

#--------------------------------------------------------------
msg "${H}\\nPart 5: Make replica service user a super user with unlimited timeout."

msg "Checking if Protections table references group $ServiceUsersGroup."

if [[ "$(p4 protects -g $ServiceUsersGroup -m)" == "super" ]]; then
   msg "Verified: Protections table grants super access to group $ServiceUsersGroup."
else
   if [[ "$UpdateProtections" -eq 1 ]]; then
      msg "Adding protections table entry to reference group $ServiceUsersGroup."
      p4 protect -o | grep -v '^#' > "$ProtectsFile" ||\
         bail "Failed to dump protections to tmp file: $ProtectsFile"
      echo -e "\\tsuper group $ServiceUsersGroup * //..." >> "$ProtectsFile" ||\
         bail "Failed to update tmp file $TmpFile."
         p4 protect -i < "$ProtectsFile" || bail "Failed to update Protecttions table from file $TmpFile."

      vmsg "Contents of $ProtectsFile:"
      vmsg "${H};"
      [[ $VERBOSITY -gt 3 ]] && cat "$ProtectsFile"
      vmsg "${H}"

      if [[ $NO_OP -eq 0 ]]; then
         p4 protect -i < "$ProtectsFile"
      else
         msg "NO_OP: Would run: p4 protect -i .LT. $ProtectsFile"
      fi
   else
      bail "The Protections table must grant super access to the group: $ServiceUsersGroup\\nConsider using the '-p' flag to update the table automatically, or else follow your site-specifc procedures for updating the Protections table to grant super access to this group."
   fi
fi

msg "Checking if serivce user $ServiceUser is in service users group $ServiceUsersGroup."

# shellcheck disable=SC2143
if [[ -n $(p4 groups "$ServiceUser" | grep "^$ServiceUsersGroup$") ]]; then
   msg "Verified: Serivce user $ServiceUser is in service users group $ServiceUsersGroup."
else
   # This logic will create the group spec for service users if it does not already exist,
   # or add our new service user to the group if it already exists.  The 'p4 group -o'
   # command generate a valid group spec whether the spec actually exists on the server or
   # not.
   p4 group -o $ServiceUsersGroup | grep -v '^#' |\
      sed "s:43200:unlimited:g;\$ s/.*/\t$ServiceUser/" > "$GroupSpecFile" ||\
      bail "Failed to update group spec file $GroupSpecFile."

   vmsg "Contents of $GroupSpecFile:"
   vmsg "${H}"
   [[ $VERBOSITY -gt 3 ]] && cat "$GroupSpecFile"
   vmsg "${H}"

   if [[ $NO_OP -eq 0 ]]; then
      p4 group -i < "$GroupSpecFile"
   else
      msg "NO_OP: Would run: p4 group -i .LT. $GroupSpecFile"
   fi
fi

#--------------------------------------------------------------
if [[ $OverallReturnStatus -eq 0 ]]; then
   msg "${H}\\nAll processing completed successfully.\\n"
   msg "\\nNext steps:"
   msg "STEP 1. Login as ${OSUSER}@${P4MASTER}."
   msg "STEP 1.1 Set your environment with:\\n\\tcd /p4/common/bin\\n\\tsource p4_vars $SDPInstance\\n"
   msg "STEP 1.2 If an Archive Filter is needed add it now into the server spec.  This is usually only needed for Export Controlled locations.\\n\\tp4 server $ServerSpec\\n"
   msg "STEP 2. Execute this command:\\n\\tnohup daily_checkpoint.sh $SDPInstance < /dev/null > /dev/null 2>&1 &\\n"
   msg "STEP 3. Monitor the checkpoint.log file, and confirm that the journal rotation\\nstep completes.  This generally occurs within the first few seconds or minutes\\nof starting daily_checkpoint.sh, even though it may take hours to complete for large\\ndata sets.  Optionally, wait for it to complete.\\n"
   msg "STEP 4. Login as ${OSUSER}@${ReplicaHost}."
   msg "STEP 5. Set your environment with:\\n\\tcd /p4/common/bin\\n\\tsource p4_vars $SDPInstance\\n"
   msg "STEP 5.1 Create local service user password file\\n\\t echo '$RandomPassword' > $P4CCFG/.p4passwd.${P4SERVER}.service"
   msg "STEP 6. Transfer the highest numbered completed checkpoint file from\\n${P4MASTER}:${CHECKPOINTS}.  Completed checkpoint files have a corresponding\\n*.md5 file.  If you waited for daily_checkpoint.sh to complete in STEP 3, you\\nneed only the latest checkpoint file.  Otherwise, you also need the journal\\nfile that has the same number.  Copy those files to:\\n$ReplicaHost:$CHECKPOINTS/.\\n\\nThe commands, as run from $ReplicaHost, will look like these sample commands:\\n\\tcd $CHECKPOINTS\\n\\tscp -p ${P4MASTER}:${CHECKPOINTS}/p4_${SDPInstance}.ckp.2550.gz .\\n\\tscp -p ${P4MASTER}:${CHECKPOINTS}/p4_${SDPInstance}.jnl.2550 .\\n\\nIf you have enabled ssh access from the commit server the commands will instead be from the commit:\\n\\tscp ${CHECKPOINTS}/p4_${SDPInstance}.ckp.2550.gz ${ReplicaHost}:${CHECKPOINTS}/\\n\\tscp ${CHECKPOINTS}/p4_${SDPInstance}.jnl.2550 ${ReplicaHost}:${CHECKPOINTS}/\\n"
   msg "STEP 7. Create $P4ROOT/server.id file like so:\\n\\techo $ServerSpec > $P4ROOT/server.id\\n"
   msg "STEP 7.1 Create $OFFLINE_DB/server.id file like so:\\n\\techo $ServerSpec > $OFFLINE_DB/server.id\\n"
   msg "STEP 8. Verify that you have enough disk space, e.g. with:\\n\\tdf -h $P4ROOT, at least 30x zipped checkpoint size is needed.\\n"
   # shellcheck disable=SC2153
   msg "STEP 9. Recover the checkpoint like so:\\n\\tcd $P4ROOT\\n\\t$P4DBIN -r $P4ROOT -z -jr ${CHECKPOINTS}/p4_${SDPInstance}.ckp.NNN.gz\\n\\nwhere NNN is the latest checkpoint from STEP 6.\\n\\nIf you did not wait for daily_checkpoint.sh to complete in STEP 3, then also replay the latest journal file copied in STEP 6, like so:\\n\\t$P4DBIN -r $P4ROOT -z -jr ${CHECKPOINTS}/${SDPInstance}.jnl.NNN"
   msg "STEP 9.1 Recover the checkpoint to the offline_db:\\n\\tcd $OFFLINE_DB\\n\\t$P4DBIN -r $OFFLINE_DB -z -jr ${CHECKPOINTS}/p4_${SDPInstance}.ckp.NNN.gz\\n\\nwhere NNN is the latest checkpoint from STEP 6."
   msg "STEP 9.2 Mark the Offline_db as usable:\\n\\techo Usable > offline_db_usable.txt"
   PasswdFile=$P4CCFG/.p4passwd.$SDPInstance.$ServiceUser
   msg "STEP 10.1 Login the service user to the master:\\n\\t$P4DBIN -u $ServiceUser -p $P4MASTERPORT login < $P4CCFG/.p4passwd.p4_$SDPInstance.service " 
   # shellcheck disable=SC2153
   msg "STEP 10.2 Start the replica like so:\\n\\t${P4DBIN}_init start\\n\\nWait several seconds, then do:\\n\\t$P4CBIN/p4login -v -service\\n\\nReview the $LOGS/p4login.log, then check replication status with:\\n\\tp4 pull -lj\\n"
   msg "STEP 11. Kick off a verify to pull over archive files:\\n\\tnohup p4verify.sh $SDPInstance < /dev/null > /dev/null 2>&1 &\\n"
   msg "STEP 12. Wait about one minute, then check, $LOGS/p4verify.log to ensure it got\\noff to a good start.  That will run for a while."

else
   msg "${H}\\nProcessing completed, but with errors.  Scan above output carefully.\\n" 
fi

# Illustrate using $SECONDS to display runtime of a script.
msg "That took $((SECONDS/3600)) hours $((SECONDS%3600/60)) minutes $((SECONDS%60)) seconds.\\n"

# See the terminate() function, which is really where this script exits.
exit "$OverallReturnStatus"