Hi,
I just wrote a thrift rpc service and run performance testing on the
programe. After stoping the test clients (2000 clients),
RES (in top command) will reduce to 84M, while 240M when 2000 clients
running. But after a few minutes, RES grows
back to 240M, and the inuse memory still low in pprof.
We run the program in CentOS release 6.5 (Final); thrift version is 0.9.3;
go version is 1.8.1.
Here is the pprof's output when the server just start:
(pprof) top
5121.71kB of 5121.71kB total ( 100%)
Dropped 9 nodes (cum <= 25.61kB)
Showing top 10 nodes out of 14 (cum >= 4097.62kB)
flat flat% sum% cum cum%
4097.62kB 80.01% 80.01% 4097.62kB 80.01% runtime.malg
512.05kB 10.00% 90.00% 512.05kB 10.00% util/workerpool.New
512.04kB 10.00% 100% 512.04kB 10.00% runtime.acquireSudog
0 0% 100% 512.05kB 10.00% main.main
0 0% 100% 512.05kB 10.00% qlyregister.
InitRegisterWorkerPool
0 0% 100% 1024.09kB 19.99% runtime.goexit
0 0% 100% 512.05kB 10.00% runtime.main
0 0% 100% 4097.62kB 80.01% runtime.mstart
0 0% 100% 4097.62kB 80.01% runtime.newproc.func1
0 0% 100% 4097.62kB 80.01% runtime.newproc1
and the top's output shows below:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
20553 root 20 0 245m 54m 3576 S 0.0 0.7 0:00.48 rpcregisterserv
After 2000 clients started,the pprof's result become:
(pprof) top
66094.31kB of 66094.31kB total ( 100%)
Dropped 137 nodes (cum <= 330.47kB)
Showing top 10 nodes out of 43 (cum >= 6270.55kB)
flat flat% sum% cum cum%
23553.62kB 35.64% 35.64% 23553.62kB 35.64% runtime.makechan
16385kB 24.79% 60.43% 41845.44kB 63.31% time.NewTimer
12848.11kB 19.44% 79.87% 12848.11kB 19.44% git.apache.org/thrift.git/lib/go
/thrift.(*tFramedTransportFactory).GetTransport
4609.83kB 6.97% 86.84% 4609.83kB 6.97% runtime.malg
2570.01kB 3.89% 90.73% 2570.01kB 3.89% github.com/garyburd/redigo/redis
.Dial
2048.16kB 3.10% 93.83% 2048.16kB 3.10% runtime.acquireSudog
1906.81kB 2.88% 96.71% 1906.81kB 2.88% runtime.addtimerLocked
1148.68kB 1.74% 98.45% 1148.68kB 1.74% runtime.allgadd
512.05kB 0.77% 99.23% 512.05kB 0.77% util/workerpool.New
512.04kB 0.77% 100% 6270.55kB 9.49% runtime.systemstack
and top's output after 2000 clients started:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
20553 root 20 0 668m 241m 4272 S 150.2 3.1 21:43.41
rpcregisterserv
After 2000 clients stop, GC already release memorys to system:
GC forced
gc 675 @1732.098s 3%: 0.072+19+0.17 ms clock, 0.28+0/19/50+0.69 ms cpu, 10->
10->10 MB, 20 MB goal, 4 P
scvg10: 155 MB released
scvg10: inuse: 61, idle: 158, sys: 220, released: 158, consumed: 61 (MB)
(pprof) top
9713.49kB of 9713.49kB total ( 100%)
Dropped 169 nodes (cum <= 48.57kB)
Showing top 10 nodes out of 21 (cum >= 3954.98kB)
flat flat% sum% cum cum%
4609.83kB 47.46% 47.46% 4609.83kB 47.46% runtime.malg
1906.81kB 19.63% 67.09% 1906.81kB 19.63% runtime.addtimerLocked
1536.12kB 15.81% 82.90% 1536.12kB 15.81% runtime.acquireSudog
1148.68kB 11.83% 94.73% 1148.68kB 11.83% runtime.allgadd
512.05kB 5.27% 100% 512.05kB 5.27% util/workerpool.New
0 0% 100% 1906.81kB 19.63% github.com/yosssi/gmq/mqtt/
client.(*Client).sendPackets
0 0% 100% 512.05kB 5.27% main.main
0 0% 100% 512.05kB 5.27% qlyregister.
InitRegisterWorkerPool
0 0% 100% 1906.81kB 19.63% runtime.addtimer
0 0% 100% 3954.98kB 40.72% runtime.goexit
But after 3 minutes, RES in top grown again:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
20553 root 20 0 668m 241m 4276 S 0.0 3.1 22:27.34 rpcregisterserv
Why RES become high again?
Thanks
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
