#!/usr/bin/env bash
#
# Copyright (c) 2016 10x Genomics, Inc. All rights reserved.
#
# Site Check for 10x Genomics pipelines.
#

eval "$(docopts -V " " -h - : "$@" <<EOF
The commands below should be preceded by '$TENX_PRODUCT':

Usage:
    sitecheck
    sitecheck -h | --help | --version

Options:
    -h --help           Show this message.
    --version           Show version.
EOF
)"

set -o pipefail

function check {
    echo "====================================================================="
    echo "$1"
    echo "$2"
    echo "---------------------------------------------------------------------"
    cmd=$2
    (
        exec 3>&1
        builtin read -t 20 < <(
            eval /usr/bin/env $cmd 1>&3 2> /dev/null
            builtin echo $?
        ) 3>&1 && exit $REPLY || {
            rc=$?
            builtin echo "timed out"
            builtin kill -s SIGHUP -- $! 2> /dev/null
            exit $rc
        }
    )
    ecode=$?
    echo "====================================================================="
    echo ""
}

function ifchk {
    if [ $ecode == 0 ]; then
        check "$1" "$2"
    fi
}


# Generate version banner
read -d '' BANNER <<EOF
$TENX_PRODUCT $TENX_SUBCMD ($TENX_VERSION)
$TENX_COPYRIGHT
-------------------------------------------------------------------------------
EOF
echo "$BANNER"

date
echo
check "System Info"         "uname -a"
check "Linux Distro"        "cat /etc/*-release | sort -u"
check "Kernel Build"        "cat /proc/version"
check "glibc version"       "ldd --version | head -n 1"
check "CPU Model"           "grep -m 1 'model name' /proc/cpuinfo | cut -d ':' -f 2 | sed 's/^[ \t]*//'"
check "CPU Sockets"         "grep 'physical id' /proc/cpuinfo | sort -u | wc -l"
check "CPU Cores"           "grep -c processor /proc/cpuinfo"
check "CPU Support"         "grep -m 1 'flags' /proc/cpuinfo | cut -d ':' -f 2 | sed 's/^\s*//'"
check "Memory Total"        "grep MemTotal /proc/meminfo | cut -d ':' -f 2 | sed 's/^[ \t]*//'"
check "Disk Space"          "df -Ph | awk '{print \$2, \$3, \$4}'"
check "Filesystem Options"  "mount | cut -d ' ' -f 5,6"
check "User Limits"         "bash -c 'ulimit -a'"
check "User Limits (hard)"  "bash -c 'ulimit -aH'"
check "Global File Limit"   "cat /proc/sys/fs/file-{max,nr}"

check "Memory config"       "sysctl vm"
check "THP memory config"   "cat /sys/kernel/mm/*transparent_hugepage/enabled"

check "cgroups"             "cat /proc/self/cgroup"
MEM_CGROUP=$(mount | grep '^cgroup\b.*\(.*\bmemory\b.*\)' | \
             cut -d ' ' -f 3)$(\
             grep -e ':memory:' /proc/self/cgroup | \
             cut -d ':' -f 3)
if [ ! -z "$MEM_CGROUP" ]; then
  check "cgroup mem stats"    "cat $MEM_CGROUP/memory.stat"
  ifchk "memory soft limit"   "cat $MEM_CGROUP/memory.*soft_limit_in_bytes"
  ifchk "memory hard limit"   "cat $MEM_CGROUP/memory.limit_in_bytes"
  ifchk "memory swap limit"   "cat $MEM_CGROUP/memory.memsw.limit_in_bytes"
fi

# docker container detection
check "Container"           "[ -e /.dockerenv ] || [ -e /.dockerinit ] \
|| [ ! -z \"\$container\" ] || grep -m 1 -E 'docker|lxc' /proc/1/cgroup \
> /dev/null && echo 'Detected'"
check "init process"        "head -n 1 /proc/1/sched | cut -d ' ' -f 1"

# cluster config
check "SGE Submit"          "which qsub"
ifchk "SGE CLUSTER_NAME"    "echo \$SGE_CLUSTER_NAME"
ifchk "SGE JOB_NAME"        "echo \$JOB_NAME"

check "LSF Submit"          "which bsub"
ifchk "LSF LSB_JOBNAME"     "echo \$LSB_JOBNAME"

check "BCL2FASTQ 1"         "which configureBclToFastq.pl"
ifchk "BCL2FASTQ 1 Version" "ls \$(dirname \$(which configureBclToFastq.pl))/../etc"
ifchk "Perl"                "which perl"
ifchk "Perl Version"        "perl -v"

check "BCL2FASTQ 2"         "which bcl2fastq"
ifchk "BCL2FASTQ 2 Version" "bcl2fastq --version"

check "Java"                "which java"
ifchk "Java Version"        "java -version 2>&1 | cat"

check "10X Refdata"         "echo \$TENX_REFDATA"
ifchk "10X Refdata Version" "cat \$TENX_REFDATA/version"

check "qconf"               "which qconf"
ifchk "qconf -sc"           "qconf -sc && qconf -sconf | grep -E '(shell_start_mode|login_shells|max_jobs)'"
ifchk "qhost"               "which qhost"
ifchk "qhost count"         "qhost -l \"mem_total=20G\" | tail -n +3 | wc -l"
ifchk "qhost -F"            "qhost -F -q -h \$(qhost | sort -h -k 5 -r | head -n 1 | cut -d \" \" -f 1)"

check "slurm info"          "sinfo -O nodes,maxcpuspernode,memory,time"

check "MRP"                 "mrp --version"
ifchk "mrp templates"       "ls \$(dirname \$(dirname \$(which mrp)))/jobmanagers/*.template"
if [ $ecode == 0 ]; then
  mrp_dir=$(dirname "$(which mrp)")
  jobmanagers=$(dirname "$mrp_dir")/jobmanagers
  for template in "$jobmanagers"/*.template; do
    ifchk "$template" "cat $template"
  done
fi