Single threaded async server in C++

[deepako...@gmail.com]

Hi,

I am planning to implement a service that has very low scale i.e. it would service only a handful of clients.

I want to keep resource usage to minimal and thus trying to use a single thread for all clients. After reading 

gRPC documentation it seems async model is the way to go. But when I tried greeter_async_server example

 in cpp, I see it creates multiple threads(18 in my case) although it uses a single thread to service all clients(which I want)

. 

Is there a way to avoid creation of so many threads in async model?

bash-4.2$ ps -ax | grep async

 1425 pts/5    Sl+    0:27 ./greeter_async_server

top - 15:15:02 up 153 days,  1:17, 13 users,  load average: 0.00, 0.00, 0.04

Threads:  18 total,   0 running,  18 sleeping,   0 stopped,   0 zombie

%Cpu(s):  0.0 us,  0.0 sy,  0.0 ni, 99.9 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st

KiB Mem : 49457112 total, 45066340 free,  3105740 used,  1285032 buff/cache

KiB Swap:  2097148 total,  2095108 free,     2040 used. 45603564 avail Mem

  PID USER      PR  NI    VIRT    RES    SHR S %CPU %MEM     TIME+ COMMAND

 1425 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.00 greeter_async_s

 1426 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:12.08 greeter_async_s

 1428 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:01.04 greeter_async_s

 1429 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.81 greeter_async_s

 1430 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.99 greeter_async_s

 1431 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:01.09 greeter_async_s

 1432 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.77 greeter_async_s

 1433 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:01.02 greeter_async_s

 1434 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.99 greeter_async_s

 1435 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:01.08 greeter_async_s

 1436 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.83 greeter_async_s

 1437 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:01.06 greeter_async_s

 1438 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.91 greeter_async_s

 1439 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.79 greeter_async_s

 1440 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:01.06 greeter_async_s

 1444 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.83 greeter_async_s

 1445 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.75 greeter_async_s

 1446 deojha    20   0  244532   7716   5164 S  0.0  0.0   0:00.76 greeter_async_s

Uses single thread to service all clients.

top - 15:15:42 up 153 days,  1:18, 13 users,  load average: 0.22, 0.05, 0.05

Threads:  18 total,   0 running,  18 sleeping,   0 stopped,   0 zombie

%Cpu(s):  6.3 us,  7.5 sy,  0.0 ni, 85.7 id,  0.0 wa,  0.0 hi,  0.5 si,  0.0 st

KiB Mem : 49457112 total, 45056008 free,  3110176 used,  1290928 buff/cache

KiB Swap:  2097148 total,  2095108 free,     2040 used. 45593760 avail Mem

  PID USER      PR  NI    VIRT    RES    SHR S %CPU %MEM     TIME+ COMMAND

 1425 deojha    20   0  244640   8308   5548 S 19.6  0.0   0:02.21 greeter_async_s

 1426 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:12.09 greeter_async_s

 1428 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:01.04 greeter_async_s

 1429 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.81 greeter_async_s

 1430 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.99 greeter_async_s

 1431 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:01.09 greeter_async_s

 1432 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.77 greeter_async_s

 1433 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:01.02 greeter_async_s

 1434 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.99 greeter_async_s

 1435 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:01.08 greeter_async_s

 1436 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.83 greeter_async_s

 1437 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:01.06 greeter_async_s

 1438 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.91 greeter_async_s

 1439 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.79 greeter_async_s

 1440 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:01.06 greeter_async_s

 1444 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.83 greeter_async_s

 1445 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.75 greeter_async_s

 1446 deojha    20   0  244640   8308   5548 S  0.0  0.0   0:00.76 greeter_async_s

Regards,

Deepak

[Sree Kuchibhotla]

Hi Deepak,

grpc core internally creates two sets of thread pools :

- Timer thread pool (to execute timers/alarms): Max of 2 threads. Typically just one.

- Executor thread pool:  A dedicated thread pool (with max threads of 2*number of cores on your machine) that handle executing closures.=

So this is what you are perhaps seeing.  Currently we have not exposed a way to configure these thread pool sizes but we might add it in future.

thanks,

-Sree

[Deepak Ojha]

Thanks Sree!

Just to confirm by closure, you mean shutting down idle connections?

Regards,

Deepak

--

Thanks & Regards

Deepak Ojha

[Sree Kuchibhotla]

Hi Deepak,

By closure I meant grpc_closure i.e the callback functions which contain most of the logic inside grpc core.

thanks,

Sree