[ceph.git] / ceph / src / rocksdb / tools / regression_test.sh

#!/usr/bin/env bash
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
# The RocksDB regression test script.
# REQUIREMENT: must be able to run make db_bench in the current directory
#
# This script will do the following things in order:
#
# 1. check out the specified rocksdb commit.
# 2. build db_bench using the specified commit
# 3. setup test directory $TEST_PATH.  If not specified, then the test directory
#    will be "/tmp/rocksdb/regression_test"
# 4. run set of benchmarks on the specified host
#    (can be either locally or remotely)
# 5. generate report in the $RESULT_PATH.  If RESULT_PATH is not specified,
#    RESULT_PATH will be set to $TEST_PATH/current_time
#
# = Examples =
# * Run the regression test using rocksdb commit abcdef that outputs results
#   and temp files in "/my/output/dir"
#r
#   TEST_PATH=/my/output/dir COMMIT_ID=abcdef ./tools/regression_test.sh
#
# * Run the regression test on a remost host under "/my/output/dir" directory
#   and stores the result locally in "/my/benchmark/results" using commit
#   abcdef and with the rocksdb options specified in /my/path/to/OPTIONS-012345
#   with 1000000000 keys in each benchmark in the regression test where each
#   key and value are 100 and 900 bytes respectively:
#
#   REMOTE_USER_AT_HOST=yhchiang@my.remote.host \
#       TEST_PATH=/my/output/dir \
#       RESULT_PATH=/my/benchmark/results \
#       COMMIT_ID=abcdef \
#       OPTIONS_FILE=/my/path/to/OPTIONS-012345 \
#       NUM_KEYS=1000000000 \
#       KEY_SIZE=100 \
#       VALUE_SIZE=900 \
#       ./tools/regression_test.sh
#
# = Regression test environmental parameters =
#   DEBUG: If true, then the script will not checkout master and build db_bench
#       if db_bench already exists
#       Default: 0
#   TEST_MODE: If 1, run fillseqdeterminstic and benchmarks both
#       if 0, only run fillseqdeterministc
#       if 2, only run benchmarks
#       Default: 1
#   TEST_PATH: the root directory of the regression test.
#       Default: "/tmp/rocksdb/regression_test"
#   RESULT_PATH: the directory where the regression results will be generated.
#       Default: "$TEST_PATH/current_time"
#   REMOTE_USER_AT_HOST: If set, then test will run on the specified host under
#       TEST_PATH directory and outputs test results locally in RESULT_PATH
#       The REMOTE_USER_AT_HOST should follow the format user-id@host.name
#   DB_PATH: the path where the rocksdb database will be created during the
#       regression test.  Default:  $TEST_PATH/db
#   WAL_PATH: the path where the rocksdb WAL will be outputed.
#       Default:  $TEST_PATH/wal
#   OPTIONS_FILE:  If specified, then the regression test will use the specified
#       file to initialize the RocksDB options in its benchmarks.  Note that
#       this feature only work for commits after 88acd93 or rocksdb version
#       later than 4.9.
#   DELETE_TEST_PATH: If true, then the test directory will be deleted
#       after the script ends.
#       Default: 0
#
# = db_bench parameters =
#   NUM_THREADS:  The number of concurrent foreground threads that will issue
#       database operations in the benchmark.  Default: 16.
#   NUM_KEYS:  The key range that will be used in the entire regression test.
#       Default: 1G.
#   NUM_OPS:  The number of operations (reads, writes, or deletes) that will
#       be issued in EACH thread.
#       Default: $NUM_KEYS / $NUM_THREADS
#   KEY_SIZE:  The size of each key in bytes in db_bench.  Default: 100.
#   VALUE_SIZE:  The size of each value in bytes in db_bench.  Default: 900.
#   CACHE_SIZE:  The size of RocksDB block cache used in db_bench.  Default: 1G
#   STATISTICS:  If 1, then statistics is on in db_bench.  Default: 0.
#   COMPRESSION_RATIO:  The compression ratio of the key generated in db_bench.
#       Default: 0.5.
#   HISTOGRAM:  If 1, then the histogram feature on performance feature is on.
#   STATS_PER_INTERVAL:  If 1, then the statistics will be reported for every
#       STATS_INTERVAL_SECONDS seconds.  Default 1.
#   STATS_INTERVAL_SECONDS:  If STATS_PER_INTERVAL is set to 1, then statistics
#       will be reported for every STATS_INTERVAL_SECONDS.  Default 60.
#   MAX_BACKGROUND_FLUSHES:  The maxinum number of concurrent flushes in
#       db_bench.  Default: 4.
#   MAX_BACKGROUND_COMPACTIONS:  The maximum number of concurrent compactions
#       in db_bench.  Default: 16.
#   NUM_HIGH_PRI_THREADS:  The number of high-pri threads available for
#       concurrent flushes in db_bench.  Default: 4.
#   NUM_LOW_PRI_THREADS:  The number of low-pri threads available for
#       concurrent compactions in db_bench.  Default: 16.
#   SEEK_NEXTS:  Controls how many Next() will be called after seek.
#       Default: 10.
#   SEED:  random seed that controls the randomness of the benchmark.
#       Default: $( date +%s )

#==============================================================================
#  CONSTANT
#==============================================================================
TITLE_FORMAT="%40s,%25s,%30s,%7s,%9s,%8s,"
TITLE_FORMAT+="%10s,%13s,%14s,%11s,%12s,"
TITLE_FORMAT+="%7s,%11s,"
TITLE_FORMAT+="%9s,%10s,%10s,%10s,%10s,%10s,%5s,"
TITLE_FORMAT+="%5s,%5s,%5s" # time
TITLE_FORMAT+="\n"

DATA_FORMAT="%40s,%25s,%30s,%7s,%9s,%8s,"
DATA_FORMAT+="%10s,%13.0f,%14s,%11s,%12s,"
DATA_FORMAT+="%7s,%11s,"
DATA_FORMAT+="%9.0f,%10.0f,%10.0f,%10.0f,%10.0f,%10.0f,%5.0f,"
DATA_FORMAT+="%5.0f,%5.0f,%5.0f" # time
DATA_FORMAT+="\n"

MAIN_PATTERN="$1""[[:blank:]]+:.*[[:blank:]]+([0-9\.]+)[[:blank:]]+ops/sec"
PERC_PATTERN="Percentiles: P50: ([0-9\.]+) P75: ([0-9\.]+) "
PERC_PATTERN+="P99: ([0-9\.]+) P99.9: ([0-9\.]+) P99.99: ([0-9\.]+)"
#==============================================================================

function main {
  TEST_ROOT_DIR=${TEST_PATH:-"/tmp/rocksdb/regression_test"}
  init_arguments $TEST_ROOT_DIR

  build_db_bench_and_ldb

  setup_test_directory
  if [ $TEST_MODE -le 1 ]; then
      tmp=$DB_PATH
      DB_PATH=$ORIGIN_PATH
      test_remote "test -d $DB_PATH"
      if [[ $? -ne 0 ]]; then
          echo "Building DB..."
          # compactall alone will not print ops or threads, which will fail update_report
          run_db_bench "fillseq,compactall" $NUM_KEYS 1 0 0
      fi
      DB_PATH=$tmp
  fi
  if [ $TEST_MODE -ge 1 ]; then
      build_checkpoint
      run_db_bench "readrandom"
      run_db_bench "readwhilewriting"
      run_db_bench "deleterandom" $((NUM_KEYS / 10 / $NUM_THREADS))
      run_db_bench "seekrandom"
      run_db_bench "seekrandomwhilewriting"
  fi

  cleanup_test_directory $TEST_ROOT_DIR
  echo ""
  echo "Benchmark completed!  Results are available in $RESULT_PATH"
}

############################################################################
function init_arguments {
  K=1024
  M=$((1024 * K))
  G=$((1024 * M))

  current_time=$(date +"%F-%H:%M:%S")
  RESULT_PATH=${RESULT_PATH:-"$1/results/$current_time"}
  COMMIT_ID=`git log | head -n1 | cut -c 8-`
  SUMMARY_FILE="$RESULT_PATH/SUMMARY.csv"

  DB_PATH=${3:-"$1/db"}
  ORIGIN_PATH=${ORIGIN_PATH:-"$(dirname $(dirname $DB_PATH))/db"}
  WAL_PATH=${4:-""}
  if [ -z "$REMOTE_USER_AT_HOST" ]; then
    DB_BENCH_DIR=${5:-"."}
  else
    DB_BENCH_DIR=${5:-"$1/db_bench"}
  fi

  DEBUG=${DEBUG:-0}
  TEST_MODE=${TEST_MODE:-1}
  SCP=${SCP:-"scp"}
  SSH=${SSH:-"ssh"}
  NUM_THREADS=${NUM_THREADS:-16}
  NUM_KEYS=${NUM_KEYS:-$((1 * G))}  # key range
  NUM_OPS=${NUM_OPS:-$(($NUM_KEYS / $NUM_THREADS))}
  KEY_SIZE=${KEY_SIZE:-100}
  VALUE_SIZE=${VALUE_SIZE:-900}
  CACHE_SIZE=${CACHE_SIZE:-$((1 * G))}
  STATISTICS=${STATISTICS:-0}
  COMPRESSION_RATIO=${COMPRESSION_RATIO:-0.5}
  HISTOGRAM=${HISTOGRAM:-1}
  NUM_MULTI_DB=${NUM_MULTI_DB:-1}
  STATS_PER_INTERVAL=${STATS_PER_INTERVAL:-1}
  STATS_INTERVAL_SECONDS=${STATS_INTERVAL_SECONDS:-600}
  MAX_BACKGROUND_FLUSHES=${MAX_BACKGROUND_FLUSHES:-4}
  MAX_BACKGROUND_COMPACTIONS=${MAX_BACKGROUND_COMPACTIONS:-16}
  NUM_HIGH_PRI_THREADS=${NUM_HIGH_PRI_THREADS:-4}
  NUM_LOW_PRI_THREADS=${NUM_LOW_PRI_THREADS:-16}
  DELETE_TEST_PATH=${DELETE_TEST_PATH:-0}
  SEEK_NEXTS=${SEEK_NEXTS:-10}
  SEED=${SEED:-$( date +%s )}
}

# $1 --- benchmark name
# $2 --- number of operations.  Default: $NUM_KEYS
# $3 --- number of threads.  Default $NUM_THREADS
# $4 --- use_existing_db.  Default: 1
# $5 --- update_report. Default: 1
function run_db_bench {
  # this will terminate all currently-running db_bench
  find_db_bench_cmd="ps aux | grep db_bench | grep -v grep | grep -v aux | awk '{print \$2}'"

  ops=${2:-$NUM_OPS}
  threads=${3:-$NUM_THREADS}
  USE_EXISTING_DB=${4:-1}
  UPDATE_REPORT=${5:-1}
  echo ""
  echo "======================================================================="
  echo "Benchmark $1"
  echo "======================================================================="
  echo ""
  db_bench_error=0
  options_file_arg=$(setup_options_file)
  echo "$options_file_arg"
  # use `which time` to avoid using bash's internal time command
  db_bench_cmd="("'\$(which time)'" -p $DB_BENCH_DIR/db_bench \
      --benchmarks=$1 --db=$DB_PATH --wal_dir=$WAL_PATH \
      --use_existing_db=$USE_EXISTING_DB \
      --disable_auto_compactions \
      --threads=$threads \
      --num=$NUM_KEYS \
      --reads=$ops \
      --writes=$ops \
      --deletes=$ops \
      --key_size=$KEY_SIZE \
      --value_size=$VALUE_SIZE \
      --cache_size=$CACHE_SIZE \
      --statistics=$STATISTICS \
      $options_file_arg \
      --compression_ratio=$COMPRESSION_RATIO \
      --histogram=$HISTOGRAM \
      --seek_nexts=$SEEK_NEXTS \
      --stats_per_interval=$STATS_PER_INTERVAL \
      --stats_interval_seconds=$STATS_INTERVAL_SECONDS \
      --max_background_flushes=$MAX_BACKGROUND_FLUSHES \
      --num_multi_db=$NUM_MULTI_DB \
      --max_background_compactions=$MAX_BACKGROUND_COMPACTIONS \
      --num_high_pri_threads=$NUM_HIGH_PRI_THREADS \
      --num_low_pri_threads=$NUM_LOW_PRI_THREADS \
      --seed=$SEED) 2>&1"
  ps_cmd="ps aux"
  if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
    echo "Running benchmark remotely on $REMOTE_USER_AT_HOST"
    db_bench_cmd="$SSH $REMOTE_USER_AT_HOST \"$db_bench_cmd\""
    ps_cmd="$SSH $REMOTE_USER_AT_HOST $ps_cmd"
  fi

  ## make sure no db_bench is running
  # The following statement is necessary make sure "eval $ps_cmd" will success.
  # Otherwise, if we simply check whether "$(eval $ps_cmd | grep db_bench)" is
  # successful or not, then it will always be false since grep will return
  # non-zero status when there's no matching output.
  ps_output="$(eval $ps_cmd)"
  exit_on_error $? "$ps_cmd"

  # perform the actual command to check whether db_bench is running
  grep_output="$(eval $ps_cmd | grep db_bench | grep -v grep)"
  if [ "$grep_output" != "" ]; then
    echo "Stopped regression_test.sh as there're still db_bench processes running:"
    echo $grep_output
    echo "Clean up test directory"
    cleanup_test_directory $TEST_ROOT_DIR
    exit 2
  fi

  ## run the db_bench
  cmd="($db_bench_cmd || db_bench_error=1) | tee -a $RESULT_PATH/$1"
  exit_on_error $?
  echo $cmd
  eval $cmd
  exit_on_error $db_bench_error
  if [ $UPDATE_REPORT -ne 0 ]; then
    update_report "$1" "$RESULT_PATH/$1" $ops $threads
  fi
}

function build_checkpoint {
    cmd_prefix=""
    if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
        cmd_prefix="$SSH $REMOTE_USER_AT_HOST "
    fi
    if [ $NUM_MULTI_DB -gt 1 ]; then
        dirs=$($cmd_prefix find $ORIGIN_PATH -type d -links 2)
        for dir in $dirs; do
            db_index=$(basename $dir)
            echo "Building checkpoints: $ORIGIN_PATH/$db_index -> $DB_PATH/$db_index ..."
            $cmd_prefix $DB_BENCH_DIR/ldb checkpoint --checkpoint_dir=$DB_PATH/$db_index \
                        --db=$ORIGIN_PATH/$db_index 2>&1
        done
    else
        # checkpoint cannot build in directory already exists
        $cmd_prefix rm -rf $DB_PATH
        echo "Building checkpoint: $ORIGIN_PATH -> $DB_PATH ..."
        $cmd_prefix $DB_BENCH_DIR/ldb checkpoint --checkpoint_dir=$DB_PATH \
                    --db=$ORIGIN_PATH 2>&1
    fi
}

function multiply {
  echo "$1 * $2" | bc
}

# $1 --- name of the benchmark
# $2 --- the filename of the output log of db_bench
function update_report {
  main_result=`cat $2 | grep $1`
  exit_on_error $?
  perc_statement=`cat $2 | grep Percentile`
  exit_on_error $?

  # Obtain micros / op

  [[ $main_result =~ $MAIN_PATTERN ]]
  ops_per_s=${BASH_REMATCH[1]}

  # Obtain percentile information
  [[ $perc_statement =~ $PERC_PATTERN ]]
  perc[0]=${BASH_REMATCH[1]}  # p50
  perc[1]=${BASH_REMATCH[2]}  # p75
  perc[2]=${BASH_REMATCH[3]}  # p99
  perc[3]=${BASH_REMATCH[4]}  # p99.9
  perc[4]=${BASH_REMATCH[5]}  # p99.99

  # Parse the output of the time command
  real_sec=`tail -3 $2 | grep real | awk '{print $2}'`
  user_sec=`tail -3 $2 | grep user | awk '{print $2}'`
  sys_sec=`tail -3 $2 | grep sys | awk '{print $2}'`

  (printf "$DATA_FORMAT" \
    $COMMIT_ID $1 $REMOTE_USER_AT_HOST $NUM_MULTI_DB $NUM_KEYS $KEY_SIZE $VALUE_SIZE \
       $(multiply $COMPRESSION_RATIO 100) \
       $3 $4 $CACHE_SIZE \
       $MAX_BACKGROUND_FLUSHES $MAX_BACKGROUND_COMPACTIONS \
       $ops_per_s \
       $(multiply ${perc[0]} 1000) \
       $(multiply ${perc[1]} 1000) \
       $(multiply ${perc[2]} 1000) \
       $(multiply ${perc[3]} 1000) \
       $(multiply ${perc[4]} 1000) \
       $DEBUG \
       $real_sec \
       $user_sec \
       $sys_sec \
       >> $SUMMARY_FILE)
  exit_on_error $?
}

function exit_on_error {
  if [ $1 -ne 0 ]; then
    echo ""
    echo "ERROR: Benchmark did not complete successfully."
    if ! [ -z "$2" ]; then
      echo "Failure command: $2"
    fi
    echo "Partial results are output to $RESULT_PATH"
    echo "ERROR" >> $SUMMARY_FILE
    exit $1
  fi
}

function checkout_rocksdb {
  echo "Checking out commit $1 ..."

  git fetch --all
  exit_on_error $?

  git checkout $1
  exit_on_error $?
}

function build_db_bench_and_ldb {
  echo "Building db_bench & ldb ..."

  make clean
  exit_on_error $?

  DEBUG_LEVEL=0 PORTABLE=1 make db_bench ldb -j32
  exit_on_error $?
}

function run_remote {
  test_remote "$1"
  exit_on_error $? "$1"
}

function test_remote {
  if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
      cmd="$SSH $REMOTE_USER_AT_HOST '$1'"
  else
      cmd="$1"
  fi
  eval "$cmd"
}

function run_local {
  eval "$1"
  exit_on_error $?
}

function setup_options_file {
  if ! [ -z "$OPTIONS_FILE" ]; then
    if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
      options_file="$DB_BENCH_DIR/OPTIONS_FILE"
      run_local "$SCP $OPTIONS_FILE $REMOTE_USER_AT_HOST:$options_file"
    else
      options_file="$OPTIONS_FILE"
    fi
    echo "--options_file=$options_file"
  fi
  echo ""
}

function setup_test_directory {
  echo "Deleting old regression test directories and creating new ones"

  run_remote "rm -rf $DB_PATH"
  run_remote "rm -rf $DB_BENCH_DIR"
  run_local "rm -rf $RESULT_PATH"

  if ! [ -z "$WAL_PATH" ]; then
    run_remote "rm -rf $WAL_PATH"
    run_remote "mkdir -p $WAL_PATH"
  fi

  run_remote "mkdir -p $DB_PATH"

  run_remote "mkdir -p $DB_BENCH_DIR"
  run_remote "ls -l $DB_BENCH_DIR"

  if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
      run_local "$SCP ./db_bench $REMOTE_USER_AT_HOST:$DB_BENCH_DIR/db_bench"
      run_local "$SCP ./ldb $REMOTE_USER_AT_HOST:$DB_BENCH_DIR/ldb"
  fi

  run_local "mkdir -p $RESULT_PATH"

  (printf $TITLE_FORMAT \
      "commit id" "benchmark" "user@host" "num-dbs" "key-range" "key-size" \
      "value-size" "compress-rate" "ops-per-thread" "num-threads" "cache-size" \
      "flushes" "compactions" \
      "ops-per-s" "p50" "p75" "p99" "p99.9" "p99.99" "debug" \
      "real-sec" "user-sec" "sys-sec" \
      >> $SUMMARY_FILE)
  exit_on_error $?
}

function cleanup_test_directory {

  if [ $DELETE_TEST_PATH -ne 0 ]; then
    echo "Clear old regression test directories and creating new ones"
    run_remote "rm -rf $DB_PATH"
    run_remote "rm -rf $WAL_PATH"
    if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
      run_remote "rm -rf $DB_BENCH_DIR"
    fi
    run_remote "rm -rf $1"
  else
    echo "------------ DEBUG MODE ------------"
    echo "DB  PATH: $DB_PATH"
    echo "WAL PATH: $WAL_PATH"
  fi
}

############################################################################

# shellcheck disable=SC2068
main $@
Commit	Line	Data
11fdf7f2	1	#!/usr/bin/env bash
f67539c2	2	# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
7c673cae FG	3	# The RocksDB regression test script.
	4	# REQUIREMENT: must be able to run make db_bench in the current directory
	5	#
	6	# This script will do the following things in order:
	7	#
	8	# 1. check out the specified rocksdb commit.
	9	# 2. build db_bench using the specified commit
	10	# 3. setup test directory $TEST_PATH. If not specified, then the test directory
	11	# will be "/tmp/rocksdb/regression_test"
	12	# 4. run set of benchmarks on the specified host
	13	# (can be either locally or remotely)
	14	# 5. generate report in the $RESULT_PATH. If RESULT_PATH is not specified,
	15	# RESULT_PATH will be set to $TEST_PATH/current_time
	16	#
	17	# = Examples =
	18	# * Run the regression test using rocksdb commit abcdef that outputs results
	19	# and temp files in "/my/output/dir"
	20	#r
	21	# TEST_PATH=/my/output/dir COMMIT_ID=abcdef ./tools/regression_test.sh
	22	#
	23	# * Run the regression test on a remost host under "/my/output/dir" directory
	24	# and stores the result locally in "/my/benchmark/results" using commit
	25	# abcdef and with the rocksdb options specified in /my/path/to/OPTIONS-012345
	26	# with 1000000000 keys in each benchmark in the regression test where each
	27	# key and value are 100 and 900 bytes respectively:
	28	#
	29	# REMOTE_USER_AT_HOST=yhchiang@my.remote.host \
	30	# TEST_PATH=/my/output/dir \
	31	# RESULT_PATH=/my/benchmark/results \
	32	# COMMIT_ID=abcdef \
	33	# OPTIONS_FILE=/my/path/to/OPTIONS-012345 \
	34	# NUM_KEYS=1000000000 \
	35	# KEY_SIZE=100 \
	36	# VALUE_SIZE=900 \
	37	# ./tools/regression_test.sh
	38	#
	39	# = Regression test environmental parameters =
	40	# DEBUG: If true, then the script will not checkout master and build db_bench
	41	# if db_bench already exists
	42	# Default: 0
	43	# TEST_MODE: If 1, run fillseqdeterminstic and benchmarks both
	44	# if 0, only run fillseqdeterministc
	45	# if 2, only run benchmarks
	46	# Default: 1
	47	# TEST_PATH: the root directory of the regression test.
	48	# Default: "/tmp/rocksdb/regression_test"
	49	# RESULT_PATH: the directory where the regression results will be generated.
	50	# Default: "$TEST_PATH/current_time"
	51	# REMOTE_USER_AT_HOST: If set, then test will run on the specified host under
	52	# TEST_PATH directory and outputs test results locally in RESULT_PATH
	53	# The REMOTE_USER_AT_HOST should follow the format user-id@host.name
	54	# DB_PATH: the path where the rocksdb database will be created during the
	55	# regression test. Default: $TEST_PATH/db
	56	# WAL_PATH: the path where the rocksdb WAL will be outputed.
	57	# Default: $TEST_PATH/wal
	58	# OPTIONS_FILE: If specified, then the regression test will use the specified
	59	# file to initialize the RocksDB options in its benchmarks. Note that
	60	# this feature only work for commits after 88acd93 or rocksdb version
	61	# later than 4.9.
	62	# DELETE_TEST_PATH: If true, then the test directory will be deleted
	63	# after the script ends.
	64	# Default: 0
	65	#
	66	# = db_bench parameters =
67	# NUM_THREADS: The number of concurrent foreground threads that will issue
68	# database operations in the benchmark. Default: 16.
69	# NUM_KEYS: The key range that will be used in the entire regression test.
70	# Default: 1G.
71	# NUM_OPS: The number of operations (reads, writes, or deletes) that will
72	# be issued in EACH thread.
73	# Default: $NUM_KEYS / $NUM_THREADS
74	# KEY_SIZE: The size of each key in bytes in db_bench. Default: 100.
75	# VALUE_SIZE: The size of each value in bytes in db_bench. Default: 900.
76	# CACHE_SIZE: The size of RocksDB block cache used in db_bench. Default: 1G
77	# STATISTICS: If 1, then statistics is on in db_bench. Default: 0.
78	# COMPRESSION_RATIO: The compression ratio of the key generated in db_bench.
79	# Default: 0.5.
80	# HISTOGRAM: If 1, then the histogram feature on performance feature is on.
81	# STATS_PER_INTERVAL: If 1, then the statistics will be reported for every
82	# STATS_INTERVAL_SECONDS seconds. Default 1.
83	# STATS_INTERVAL_SECONDS: If STATS_PER_INTERVAL is set to 1, then statistics
84	# will be reported for every STATS_INTERVAL_SECONDS. Default 60.
85	# MAX_BACKGROUND_FLUSHES: The maxinum number of concurrent flushes in
86	# db_bench. Default: 4.
87	# MAX_BACKGROUND_COMPACTIONS: The maximum number of concurrent compactions
88	# in db_bench. Default: 16.
11fdf7f2 TL	89	# NUM_HIGH_PRI_THREADS: The number of high-pri threads available for
	90	# concurrent flushes in db_bench. Default: 4.
	91	# NUM_LOW_PRI_THREADS: The number of low-pri threads available for
	92	# concurrent compactions in db_bench. Default: 16.
7c673cae FG	93	# SEEK_NEXTS: Controls how many Next() will be called after seek.
	94	# Default: 10.
	95	# SEED: random seed that controls the randomness of the benchmark.
	96	# Default: $( date +%s )
	97
	98	#==============================================================================
	99	# CONSTANT
	100	#==============================================================================
	101	TITLE_FORMAT="%40s,%25s,%30s,%7s,%9s,%8s,"
	102	TITLE_FORMAT+="%10s,%13s,%14s,%11s,%12s,"
	103	TITLE_FORMAT+="%7s,%11s,"
	104	TITLE_FORMAT+="%9s,%10s,%10s,%10s,%10s,%10s,%5s,"
	105	TITLE_FORMAT+="%5s,%5s,%5s" # time
	106	TITLE_FORMAT+="\n"
	107
	108	DATA_FORMAT="%40s,%25s,%30s,%7s,%9s,%8s,"
	109	DATA_FORMAT+="%10s,%13.0f,%14s,%11s,%12s,"
	110	DATA_FORMAT+="%7s,%11s,"
	111	DATA_FORMAT+="%9.0f,%10.0f,%10.0f,%10.0f,%10.0f,%10.0f,%5.0f,"
	112	DATA_FORMAT+="%5.0f,%5.0f,%5.0f" # time
	113	DATA_FORMAT+="\n"
	114
	115	MAIN_PATTERN="$1""[[:blank:]]+:.*[[:blank:]]+([0-9\.]+)[[:blank:]]+ops/sec"
	116	PERC_PATTERN="Percentiles: P50: ([0-9\.]+) P75: ([0-9\.]+) "
	117	PERC_PATTERN+="P99: ([0-9\.]+) P99.9: ([0-9\.]+) P99.99: ([0-9\.]+)"
	118	#==============================================================================
	119
	120	function main {
11fdf7f2 TL	121	TEST_ROOT_DIR=${TEST_PATH:-"/tmp/rocksdb/regression_test"}
	122	init_arguments $TEST_ROOT_DIR
	123
	124	build_db_bench_and_ldb
7c673cae FG	125
	126	setup_test_directory
	127	if [ $TEST_MODE -le 1 ]; then
	128	tmp=$DB_PATH
	129	DB_PATH=$ORIGIN_PATH
	130	test_remote "test -d $DB_PATH"
11fdf7f2	131	if [[ $? -ne 0 ]]; then
7c673cae	132	echo "Building DB..."
11fdf7f2 TL	133	# compactall alone will not print ops or threads, which will fail update_report
11fdf7f2 TL	134	run_db_bench "fillseq,compactall" $NUM_KEYS 1 0 0
7c673cae FG	135	fi
	136	DB_PATH=$tmp
	137	fi
	138	if [ $TEST_MODE -ge 1 ]; then
	139	build_checkpoint
	140	run_db_bench "readrandom"
	141	run_db_bench "readwhilewriting"
	142	run_db_bench "deleterandom" $((NUM_KEYS / 10 / $NUM_THREADS))
	143	run_db_bench "seekrandom"
	144	run_db_bench "seekrandomwhilewriting"
	145	fi
	146
11fdf7f2	147	cleanup_test_directory $TEST_ROOT_DIR
7c673cae FG	148	echo ""
	149	echo "Benchmark completed! Results are available in $RESULT_PATH"
	150	}
	151
	152	############################################################################
	153	function init_arguments {
	154	K=1024
	155	M=$((1024 * K))
	156	G=$((1024 * M))
	157
	158	current_time=$(date +"%F-%H:%M:%S")
	159	RESULT_PATH=${RESULT_PATH:-"$1/results/$current_time"}
	160	COMMIT_ID=`git log \| head -n1 \| cut -c 8-`
	161	SUMMARY_FILE="$RESULT_PATH/SUMMARY.csv"
	162
	163	DB_PATH=${3:-"$1/db"}
	164	ORIGIN_PATH=${ORIGIN_PATH:-"$(dirname $(dirname $DB_PATH))/db"}
	165	WAL_PATH=${4:-""}
	166	if [ -z "$REMOTE_USER_AT_HOST" ]; then
	167	DB_BENCH_DIR=${5:-"."}
	168	else
	169	DB_BENCH_DIR=${5:-"$1/db_bench"}
	170	fi
	171
	172	DEBUG=${DEBUG:-0}
	173	TEST_MODE=${TEST_MODE:-1}
	174	SCP=${SCP:-"scp"}
	175	SSH=${SSH:-"ssh"}
	176	NUM_THREADS=${NUM_THREADS:-16}
	177	NUM_KEYS=${NUM_KEYS:-$((1 * G))} # key range
	178	NUM_OPS=${NUM_OPS:-$(($NUM_KEYS / $NUM_THREADS))}
	179	KEY_SIZE=${KEY_SIZE:-100}
	180	VALUE_SIZE=${VALUE_SIZE:-900}
	181	CACHE_SIZE=${CACHE_SIZE:-$((1 * G))}
	182	STATISTICS=${STATISTICS:-0}
	183	COMPRESSION_RATIO=${COMPRESSION_RATIO:-0.5}
	184	HISTOGRAM=${HISTOGRAM:-1}
	185	NUM_MULTI_DB=${NUM_MULTI_DB:-1}
	186	STATS_PER_INTERVAL=${STATS_PER_INTERVAL:-1}
	187	STATS_INTERVAL_SECONDS=${STATS_INTERVAL_SECONDS:-600}
	188	MAX_BACKGROUND_FLUSHES=${MAX_BACKGROUND_FLUSHES:-4}
	189	MAX_BACKGROUND_COMPACTIONS=${MAX_BACKGROUND_COMPACTIONS:-16}
11fdf7f2 TL	190	NUM_HIGH_PRI_THREADS=${NUM_HIGH_PRI_THREADS:-4}
11fdf7f2 TL	191	NUM_LOW_PRI_THREADS=${NUM_LOW_PRI_THREADS:-16}
7c673cae FG	192	DELETE_TEST_PATH=${DELETE_TEST_PATH:-0}
	193	SEEK_NEXTS=${SEEK_NEXTS:-10}
	194	SEED=${SEED:-$( date +%s )}
	195	}
	196
	197	# $1 --- benchmark name
	198	# $2 --- number of operations. Default: $NUM_KEYS
	199	# $3 --- number of threads. Default $NUM_THREADS
	200	# $4 --- use_existing_db. Default: 1
11fdf7f2	201	# $5 --- update_report. Default: 1
7c673cae FG	202	function run_db_bench {
	203	# this will terminate all currently-running db_bench
	204	find_db_bench_cmd="ps aux \| grep db_bench \| grep -v grep \| grep -v aux \| awk '{print \$2}'"
	205
7c673cae FG	206	ops=${2:-$NUM_OPS}
7c673cae FG	207	threads=${3:-$NUM_THREADS}
11fdf7f2 TL	208	USE_EXISTING_DB=${4:-1}
11fdf7f2 TL	209	UPDATE_REPORT=${5:-1}
7c673cae FG	210	echo ""
	211	echo "======================================================================="
	212	echo "Benchmark $1"
	213	echo "======================================================================="
	214	echo ""
	215	db_bench_error=0
	216	options_file_arg=$(setup_options_file)
	217	echo "$options_file_arg"
	218	# use `which time` to avoid using bash's internal time command
	219	db_bench_cmd="("'\$(which time)'" -p $DB_BENCH_DIR/db_bench \
	220	--benchmarks=$1 --db=$DB_PATH --wal_dir=$WAL_PATH \
	221	--use_existing_db=$USE_EXISTING_DB \
	222	--disable_auto_compactions \
	223	--threads=$threads \
	224	--num=$NUM_KEYS \
	225	--reads=$ops \
	226	--writes=$ops \
	227	--deletes=$ops \
	228	--key_size=$KEY_SIZE \
	229	--value_size=$VALUE_SIZE \
	230	--cache_size=$CACHE_SIZE \
	231	--statistics=$STATISTICS \
	232	$options_file_arg \
	233	--compression_ratio=$COMPRESSION_RATIO \
	234	--histogram=$HISTOGRAM \
	235	--seek_nexts=$SEEK_NEXTS \
	236	--stats_per_interval=$STATS_PER_INTERVAL \
	237	--stats_interval_seconds=$STATS_INTERVAL_SECONDS \
	238	--max_background_flushes=$MAX_BACKGROUND_FLUSHES \
	239	--num_multi_db=$NUM_MULTI_DB \
	240	--max_background_compactions=$MAX_BACKGROUND_COMPACTIONS \
11fdf7f2 TL	241	--num_high_pri_threads=$NUM_HIGH_PRI_THREADS \
11fdf7f2 TL	242	--num_low_pri_threads=$NUM_LOW_PRI_THREADS \
7c673cae FG	243	--seed=$SEED) 2>&1"
	244	ps_cmd="ps aux"
	245	if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
	246	echo "Running benchmark remotely on $REMOTE_USER_AT_HOST"
	247	db_bench_cmd="$SSH $REMOTE_USER_AT_HOST \"$db_bench_cmd\""
	248	ps_cmd="$SSH $REMOTE_USER_AT_HOST $ps_cmd"
	249	fi
	250
	251	## make sure no db_bench is running
	252	# The following statement is necessary make sure "eval $ps_cmd" will success.
	253	# Otherwise, if we simply check whether "$(eval $ps_cmd \| grep db_bench)" is
	254	# successful or not, then it will always be false since grep will return
	255	# non-zero status when there's no matching output.
	256	ps_output="$(eval $ps_cmd)"
	257	exit_on_error $? "$ps_cmd"
	258
	259	# perform the actual command to check whether db_bench is running
	260	grep_output="$(eval $ps_cmd \| grep db_bench \| grep -v grep)"
	261	if [ "$grep_output" != "" ]; then
	262	echo "Stopped regression_test.sh as there're still db_bench processes running:"
	263	echo $grep_output
11fdf7f2 TL	264	echo "Clean up test directory"
11fdf7f2 TL	265	cleanup_test_directory $TEST_ROOT_DIR
7c673cae FG	266	exit 2
	267	fi
	268
	269	## run the db_bench
	270	cmd="($db_bench_cmd \|\| db_bench_error=1) \| tee -a $RESULT_PATH/$1"
	271	exit_on_error $?
	272	echo $cmd
	273	eval $cmd
	274	exit_on_error $db_bench_error
11fdf7f2 TL	275	if [ $UPDATE_REPORT -ne 0 ]; then
	276	update_report "$1" "$RESULT_PATH/$1" $ops $threads
	277	fi
7c673cae FG	278	}
	279
	280	function build_checkpoint {
	281	cmd_prefix=""
	282	if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
	283	cmd_prefix="$SSH $REMOTE_USER_AT_HOST "
	284	fi
11fdf7f2 TL	285	if [ $NUM_MULTI_DB -gt 1 ]; then
	286	dirs=$($cmd_prefix find $ORIGIN_PATH -type d -links 2)
	287	for dir in $dirs; do
	288	db_index=$(basename $dir)
	289	echo "Building checkpoints: $ORIGIN_PATH/$db_index -> $DB_PATH/$db_index ..."
	290	$cmd_prefix $DB_BENCH_DIR/ldb checkpoint --checkpoint_dir=$DB_PATH/$db_index \
	291	--db=$ORIGIN_PATH/$db_index 2>&1
	292	done
	293	else
	294	# checkpoint cannot build in directory already exists
	295	$cmd_prefix rm -rf $DB_PATH
	296	echo "Building checkpoint: $ORIGIN_PATH -> $DB_PATH ..."
	297	$cmd_prefix $DB_BENCH_DIR/ldb checkpoint --checkpoint_dir=$DB_PATH \
	298	--db=$ORIGIN_PATH 2>&1
	299	fi
7c673cae FG	300	}
	301
	302	function multiply {
	303	echo "$1 * $2" \| bc
	304	}
	305
	306	# $1 --- name of the benchmark
	307	# $2 --- the filename of the output log of db_bench
	308	function update_report {
	309	main_result=`cat $2 \| grep $1`
	310	exit_on_error $?
	311	perc_statement=`cat $2 \| grep Percentile`
	312	exit_on_error $?
	313
	314	# Obtain micros / op
	315
	316	[[ $main_result =~ $MAIN_PATTERN ]]
	317	ops_per_s=${BASH_REMATCH[1]}
	318
	319	# Obtain percentile information
	320	[[ $perc_statement =~ $PERC_PATTERN ]]
	321	perc[0]=${BASH_REMATCH[1]} # p50
	322	perc[1]=${BASH_REMATCH[2]} # p75
	323	perc[2]=${BASH_REMATCH[3]} # p99
	324	perc[3]=${BASH_REMATCH[4]} # p99.9
	325	perc[4]=${BASH_REMATCH[5]} # p99.99
	326
	327	# Parse the output of the time command
	328	real_sec=`tail -3 $2 \| grep real \| awk '{print $2}'`
	329	user_sec=`tail -3 $2 \| grep user \| awk '{print $2}'`
	330	sys_sec=`tail -3 $2 \| grep sys \| awk '{print $2}'`
	331
	332	(printf "$DATA_FORMAT" \
	333	$COMMIT_ID $1 $REMOTE_USER_AT_HOST $NUM_MULTI_DB $NUM_KEYS $KEY_SIZE $VALUE_SIZE \
	334	$(multiply $COMPRESSION_RATIO 100) \
	335	$3 $4 $CACHE_SIZE \
	336	$MAX_BACKGROUND_FLUSHES $MAX_BACKGROUND_COMPACTIONS \
	337	$ops_per_s \
	338	$(multiply ${perc[0]} 1000) \
	339	$(multiply ${perc[1]} 1000) \
	340	$(multiply ${perc[2]} 1000) \
	341	$(multiply ${perc[3]} 1000) \
	342	$(multiply ${perc[4]} 1000) \
	343	$DEBUG \
	344	$real_sec \
	345	$user_sec \
	346	$sys_sec \
	347	>> $SUMMARY_FILE)
	348	exit_on_error $?
	349	}
	350
	351	function exit_on_error {
	352	if [ $1 -ne 0 ]; then
	353	echo ""
	354	echo "ERROR: Benchmark did not complete successfully."
	355	if ! [ -z "$2" ]; then
	356	echo "Failure command: $2"
	357	fi
	358	echo "Partial results are output to $RESULT_PATH"
	359	echo "ERROR" >> $SUMMARY_FILE
	360	exit $1
	361	fi
	362	}
	363
364	function checkout_rocksdb {
365	echo "Checking out commit $1 ..."
366
367	git fetch --all
368	exit_on_error $?
369
370	git checkout $1
371	exit_on_error $?
372	}
373
374	function build_db_bench_and_ldb {
375	echo "Building db_bench & ldb ..."
376
377	make clean
378	exit_on_error $?
379
11fdf7f2	380	DEBUG_LEVEL=0 PORTABLE=1 make db_bench ldb -j32
7c673cae FG	381	exit_on_error $?
	382	}
	383
	384	function run_remote {
	385	test_remote "$1"
	386	exit_on_error $? "$1"
	387	}
	388
	389	function test_remote {
	390	if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
	391	cmd="$SSH $REMOTE_USER_AT_HOST '$1'"
	392	else
	393	cmd="$1"
	394	fi
	395	eval "$cmd"
	396	}
	397
	398	function run_local {
	399	eval "$1"
	400	exit_on_error $?
	401	}
	402
	403	function setup_options_file {
	404	if ! [ -z "$OPTIONS_FILE" ]; then
	405	if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
	406	options_file="$DB_BENCH_DIR/OPTIONS_FILE"
	407	run_local "$SCP $OPTIONS_FILE $REMOTE_USER_AT_HOST:$options_file"
	408	else
	409	options_file="$OPTIONS_FILE"
	410	fi
	411	echo "--options_file=$options_file"
	412	fi
	413	echo ""
	414	}
	415
	416	function setup_test_directory {
	417	echo "Deleting old regression test directories and creating new ones"
	418
	419	run_remote "rm -rf $DB_PATH"
	420	run_remote "rm -rf $DB_BENCH_DIR"
	421	run_local "rm -rf $RESULT_PATH"
	422
	423	if ! [ -z "$WAL_PATH" ]; then
	424	run_remote "rm -rf $WAL_PATH"
	425	run_remote "mkdir -p $WAL_PATH"
	426	fi
	427
	428	run_remote "mkdir -p $DB_PATH"
	429
	430	run_remote "mkdir -p $DB_BENCH_DIR"
	431	run_remote "ls -l $DB_BENCH_DIR"
	432
	433	if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
	434	run_local "$SCP ./db_bench $REMOTE_USER_AT_HOST:$DB_BENCH_DIR/db_bench"
	435	run_local "$SCP ./ldb $REMOTE_USER_AT_HOST:$DB_BENCH_DIR/ldb"
	436	fi
	437
	438	run_local "mkdir -p $RESULT_PATH"
	439
	440	(printf $TITLE_FORMAT \
	441	"commit id" "benchmark" "user@host" "num-dbs" "key-range" "key-size" \
	442	"value-size" "compress-rate" "ops-per-thread" "num-threads" "cache-size" \
	443	"flushes" "compactions" \
	444	"ops-per-s" "p50" "p75" "p99" "p99.9" "p99.99" "debug" \
445	"real-sec" "user-sec" "sys-sec" \
446	>> $SUMMARY_FILE)
447	exit_on_error $?
448	}
449
450	function cleanup_test_directory {
451
452	if [ $DELETE_TEST_PATH -ne 0 ]; then
453	echo "Clear old regression test directories and creating new ones"
454	run_remote "rm -rf $DB_PATH"
455	run_remote "rm -rf $WAL_PATH"
456	if ! [ -z "$REMOTE_USER_AT_HOST" ]; then
457	run_remote "rm -rf $DB_BENCH_DIR"
458	fi
459	run_remote "rm -rf $1"
460	else
461	echo "------------ DEBUG MODE ------------"
462	echo "DB PATH: $DB_PATH"
463	echo "WAL PATH: $WAL_PATH"
464	fi
465	}
466
467	############################################################################
468
11fdf7f2	469	# shellcheck disable=SC2068
7c673cae	470	main $@