PostgreSQL on ECS 横向版本TPCB性能测试

本文对比的是PostgreSQL 9.4, 9.5, 9.6, 以及某基于9.4的国产数据库在ECS上的性能。

硬件环境和搭配

Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz
32核
128G 内存
500G 高效本地SSD  ,  15000 IOPS  ,  $PGDATA
700G * 2 高效SSD云盘  ,  20000 IOPS  ,  tbs1, tbs2

测试模型 TPCB
分别测试的数据量 10万乘以 1 10 100 500 1000 5000 10000
分别测试的并发数 1 2 4 8 16 32 64 128 256 512

环境部署

parted /dev/vdc
(parted) mklabel gpt
(parted) mkpart primary 1M 2049M
(parted) mkpart primary 2049M 710G
(parted) select /dev/vdb
Using /dev/vdb
(parted) rm 1
(parted) mkpart primary 1M 2049M
(parted) mkpart primary 2049M 710G
q

partx -a /dev/vdc
kpartx -a /dev/vdc
partprobe  /dev/vdc
partx -a /dev/vdb
kpartx -a /dev/vdb
partprobe  /dev/vdb

dmsetup remove vdc1
dmsetup remove vdc2

dmsetup remove vdb1
dmsetup remove vdb2

mkfs.xfs -f -b size=4096 -l logdev=/dev/vdc1,size=2047868928,sunit=16 -d agsize=134217728 /dev/vdc2
mkfs.xfs -f -b size=4096 -l logdev=/dev/vdb1,size=2047868928,sunit=16 -d agsize=134217728 /dev/vdb2

vi /etc/fstab
/dev/vdc2 /data02 xfs defaults,allocsize=16M,inode64,nobarrier,nolargeio,logbsize=262144,noatime,nodiratime,swalloc,logdev=/dev/vdc1 0 0
/dev/vdb2 /data01 xfs defaults,allocsize=16M,inode64,nobarrier,nolargeio,logbsize=262144,noatime,nodiratime,swalloc,logdev=/dev/vdb1 0 0

mkdir /data01
mkdir /data02
mkdir /data03

mount -a

mkdir -p /data01/pgdata
mkdir -p /data02/pgdata
mkdir -p /data03/pgdata
chown -R digoal:digoal /data*/*

还需要调整一些内核参数，LIMITS，本文略 .   

表空间目录

mkdir /data01/pgdata/9.4_tbs
mkdir /data02/pgdata/9.4_tbs
mkdir /data01/pgdata/9.5_tbs
mkdir /data02/pgdata/9.5_tbs
mkdir /data01/pgdata/9.6_tbs
mkdir /data02/pgdata/9.6_tbs
mkdir /data01/pgdata/sgrds_tbs
mkdir /data02/pgdata/sgrds_tbs
chown -R digoal:digoal /data*/*

配置环境变量

# vi env.sh  

#!/bin/bash 

if [ $# -ne 2 ]; then
  echo "$0 ver port"
else

export PS1="$USER@`/bin/hostname -s`-> "
export PGPORT=$2
export PGDATA=/data03/pgdata/pg_root$1
export LANG=en_US.utf8
export PGHOME=/home/digoal/pgsql$1
export LD_LIBRARY_PATH=$PGHOME/lib:/lib64:/usr/lib64:/usr/local/lib64:/lib:/usr/lib:/usr/local/lib:$LD_LIBRARY_PATH
export DATE=`date +"%Y%m%d%H%M"`
export PATH=$PGHOME/bin:$PATH:.
export MANPATH=$PGHOME/share/man:$MANPATH
export PGHOST=$PGDATA
export PGUSER=postgres
export PGDATABASE=postgres
alias rm='rm -i'
alias ll='ls -lh'
unalias vi

fi

初始化数据库集群

. ./env.sh 9.4 1920
initdb -D $PGDATA -E UTF8 --locale=C -U postgres

. ./env.sh 9.5 1921
initdb -D $PGDATA -E UTF8 --locale=C -U postgres

. ./env.sh 9.6 1922
initdb -D $PGDATA -E UTF8 --locale=C -U postgres

. ./env.sh sgrds 1923
initdb -D $PGDATA -E UTF8 --locale=C -U postgres

启动数据库

. ./env.sh 9.4 1920
pg_ctl stop -m fast
. ./env.sh 9.5 1921
pg_ctl stop -m fast
. ./env.sh 9.6 1922
pg_ctl stop -m fast
. ./env.sh sgrds 1923
pg_ctl stop -m fast

. ./env.sh 9.4 1920
nohup postgres -B 16GB \
-c port=$PGPORT \
-c max_connections=700 \
-c unix_socket_directories='.' \
-c listen_addresses='0.0.0.0' \
-c vacuum_cost_delay=0 \
-c vacuum_cost_limit=10000 \
-c bgwriter_delay=10ms \
-c bgwriter_lru_maxpages=1000 \
-c bgwriter_lru_multiplier=10.0 \
-c synchronous_commit=off \
-c full_page_writes=off \
-c wal_buffers=1900MB \
-c wal_writer_delay=10ms \
-c checkpoint_timeout=35min \
-c checkpoint_segments=1024 \
-c checkpoint_completion_target=0.9 \
-c random_page_cost=1.1 \
-c effective_cache_size=64GB \
-c log_destination='csvlog' \
-c logging_collector=on \
-c log_truncate_on_rotation=on \
-D $PGDATA \
>/dev/null 2>&1 &
sleep 2
psql -c "create tablespace tbs1 location '/data01/pgdata/$PGPORT_tbs';"
psql -c "create tablespace tbs2 location '/data02/pgdata/$PGPORT_tbs';"

. ./env.sh 9.5 1921
nohup postgres -B 16GB \
-c port=$PGPORT \
-c max_connections=700 \
-c unix_socket_directories='.' \
-c listen_addresses='0.0.0.0' \
-c vacuum_cost_delay=0 \
-c vacuum_cost_limit=10000 \
-c bgwriter_delay=10ms \
-c bgwriter_lru_maxpages=1000 \
-c bgwriter_lru_multiplier=10.0 \
-c synchronous_commit=off \
-c full_page_writes=off \
-c wal_buffers=1900MB \
-c wal_writer_delay=10ms \
-c checkpoint_timeout=35min \
-c max_wal_size=16GB \
-c checkpoint_completion_target=0.9 \
-c random_page_cost=1.1 \
-c effective_cache_size=64GB \
-c log_destination='csvlog' \
-c logging_collector=on \
-c log_truncate_on_rotation=on \
-D $PGDATA \
>/dev/null 2>&1 &
sleep 2
psql -c "create tablespace tbs1 location '/data01/pgdata/$PGPORT_tbs';"
psql -c "create tablespace tbs2 location '/data02/pgdata/$PGPORT_tbs';"

. ./env.sh 9.6 1922
nohup postgres -B 16GB \
-p $PGPORT \
-N 700 \
-c unix_socket_directories='.' \
-c listen_addresses='0.0.0.0' \
-c vacuum_cost_delay=0 \
-c vacuum_cost_limit=10000 \
-c bgwriter_delay=10ms \
-c bgwriter_lru_maxpages=1000 \
-c bgwriter_lru_multiplier=10.0 \
-c synchronous_commit=off \
-c wal_writer_flush_after=0 \
-c full_page_writes=off \
-c wal_buffers=1900MB \
-c wal_writer_delay=10ms \
-c checkpoint_timeout=35min \
-c max_wal_size=16GB \
-c checkpoint_completion_target=0.9 \
-c random_page_cost=1.1 \
-c effective_cache_size=64GB \
-c log_destination='csvlog' \
-c logging_collector=on \
-c log_truncate_on_rotation=on \
-D $PGDATA \
>/dev/null 2>&1 &
sleep 2
psql -c "create tablespace tbs1 location '/data01/pgdata/$PGPORT_tbs';"
psql -c "create tablespace tbs2 location '/data02/pgdata/$PGPORT_tbs';"

. ./env.sh sgrds 1923
nohup postgres -B 16GB \
-c port=$PGPORT \
-c max_connections=700 \
-c unix_socket_directories='.' \
-c listen_addresses='0.0.0.0' \
-c vacuum_cost_delay=0 \
-c vacuum_cost_limit=10000 \
-c bgwriter_delay=10ms \
-c bgwriter_lru_maxpages=1000 \
-c bgwriter_lru_multiplier=10.0 \
-c synchronous_commit=off \
-c full_page_writes=off \
-c wal_buffers=1900MB \
-c wal_writer_delay=10ms \
-c checkpoint_timeout=35min \
-c checkpoint_segments=1024 \
-c checkpoint_completion_target=0.9 \
-c random_page_cost=1.1 \
-c effective_cache_size=64GB \
-c log_destination='csvlog' \
-c logging_collector=on \
-c log_truncate_on_rotation=on \
-D $PGDATA \
>/dev/null 2>&1 &
sleep 2
psql -c "create tablespace tbs1 location '/data01/pgdata/$PGPORT_tbs';"
psql -c "create tablespace tbs2 location '/data02/pgdata/$PGPORT_tbs';"

测试脚本

# vi test.sh
#!/bin/bash

for s in 1 10 100 500 1000 5000 10000
do

################
. ~/env.sh 9.4 1920
pgbench -i -s $s --tablespace=tbs1 --index-tablespace=tbs2 --unlogged-tables

for i in 1 2 4 8 16 32 64 128 256 512
do
if [ $i -ge 16 ]; then
  j=16
else
  j=i
fi
pgbench -M prepared -n -r -c $i -j $i -T 120 > ./9.4.s_$s.i_$i.log 2>&1
done
pgbench -i -s 1
#################

################
. ~/env.sh 9.5 1921
pgbench -i -s $s --tablespace=tbs1 --index-tablespace=tbs2 --unlogged-tables

for i in 1 2 4 8 16 32 64 128 256 512
do
if [ $i -ge 16 ]; then
  j=16
else
  j=i
fi
pgbench -M prepared -n -r -c $i -j $i -T 120 > ./9.5.s_$s.i_$i.log 2>&1
done
pgbench -i -s 1
#################

################
. ~/env.sh 9.6 1922
pgbench -i -s $s --tablespace=tbs1 --index-tablespace=tbs2 --unlogged-tables

for i in 1 2 4 8 16 32 64 128 256 512
do
if [ $i -ge 16 ]; then
  j=16
else
  j=i
fi
pgbench -M prepared -n -r -c $i -j $i -T 120 > ./9.6.s_$s.i_$i.log 2>&1
done
pgbench -i -s 1
#################

################
. ~/env.sh sgrds 1923
pgbench -i -s $s --tablespace=tbs1 --index-tablespace=tbs2 --unlogged-tables

for i in 1 2 4 8 16 32 64 128 256 512
do
if [ $i -ge 16 ]; then
  j=16
else
  j=i
fi
pgbench -M prepared -n -r -c $i -j $i -T 120 > ./sgrds.s_$s.i_$i.log 2>&1
done
pgbench -i -s 1
#################

done

测试

chmod 500 test.sh
nohup ./test.sh >/dev/null 2>&1 &

测试结果

# vi result.sh

for s in 1 10 100 500 1000 5000 10000
do

echo "9.4 s $s"
for i in 1 2 4 8 16 32 64 128 256 512
do
cat 9.4.s_$s.i_$i.log | grep including | awk '{print $3}'
done

echo "9.5 s $s"
for i in 1 2 4 8 16 32 64 128 256 512
do
cat 9.5.s_$s.i_$i.log | grep including | awk '{print $3}'
done

echo "9.6 s $s"
for i in 1 2 4 8 16 32 64 128 256 512
do
cat 9.6.s_$s.i_$i.log | grep including | awk '{print $3}'
done

echo "sgrds s $s"
for i in 1 2 4 8 16 32 64 128 256 512
do
cat sgrds.s_$s.i_$i.log | grep including | awk '{print $3}'
done

done

. ./result.sh

瓶颈分析请结合压测过程中的主机资源开销(io,cpu,调用,内存,网络等)和PROFILE来判断。
数据量小的时候，高并发的情况下行级锁等待较多，因为数据量少，高并发的UPDATE的原因。