Complete Win32 non-blocking implementation by hcorrada · Pull Request #58 · rstudio/httpuv

hcorrada · 2016-01-30T19:41:27Z

Using ideas from Simon Urbanek's 'background' draft package that implements the Rhttpd strategy. Implementation description in src/httpuv.cpp

Fully implement solution for running the libuv default loop without blocking in win32. It is based on Simon Urbanek's 'background' draft package that implements a version of the Rhttpd strategy. Implementation details are in src/httpuv.cpp file.

jcheng5 · 2016-02-11T20:44:34Z

-          TerminateThread(server_thread, 0);
-        server_thread = 0;
+        if (GetExitCodeThread(thread, &ts) && ts == STILL_ACTIVE)
+          TerminateThread(thread, 0);


This seems really bad--it's almost never a good idea to forcibly terminate a thread from the outside, with the possible exception of when the entire process is about to shut down. It can leave all sorts of state corrupted. The usual approach is to set a flag that the other thread will check, and if it's important for this thread to not continue until it knows the other thread is dead, to wait/join the other thread (in Windows it looks like WaitForSingleObject will work).

Thanks. I'll take a look, these lines are straight from 'background' and Rhttpd, so I'll pass along whatever we end up using here to them.

jcheng5 · 2016-02-11T20:47:52Z

Thanks for all the hard work. I know this stuff takes a lot of effort.

I don't fully understand the comment in httpuv.cpp; is the libuv loop run on the main thread, or the background thread?

jcheng5 · 2016-02-11T20:50:31Z

I may also be spending more time on this in the future. It turns out that even in the non-daemonized case, having the libuv loop running on the main thread really hurts I/O performance a lot when R is also busy--it's enough of a problem that some geospatial related features in Leaflet are not worth shipping until we resolve it.

hcorrada · 2016-02-12T13:12:25Z

The libuv ends up running in the main thread, through this message passing mechanism. The background thread is only running the 'polling' loop. Again, straight from 'background' and Rhttpd.

What do you have in mind? With the existing 'InputHandler' and this win32 message passing mechanisms, the libuv loop still runs in the main R thread. Perhaps use libuv's thread mechanism directly?

Ok, thanks, I'll add a better solution to thread stopping in the next day or so.

jjallaire · 2016-02-12T13:27:58Z

If the background thread is running a message loop then you should just be
able to "PostMessage" (async message pass) a custom WM_ message telling it
to terminate (you just then handle that message within the thread by call
ExitThread.

On Fri, Feb 12, 2016 at 8:12 AM, hcorrada notifications@github.com wrote:

The libuv ends up running in the main thread, through this message passing
mechanism. The background thread is only running the 'polling' loop. Again,
straight from 'background' and Rhttpd.

What do you have in mind? With the existing 'InputHandler' and this win32
message passing mechanisms, the libuv loop still runs in the main R thread.
Perhaps use libuv's thread mechanism directly?

Ok, thanks, I'll add a better solution to thread stopping in the next day
or so.

—
Reply to this email directly or view it on GitHub
#58 (comment).

hcorrada · 2016-03-21T19:38:30Z

@jcheng5 Check this thread exit strategy. Let me know if it can be improved. THanks!

jcheng5 · 2016-03-23T17:23:58Z

+    int ntimes = 10;
+    int i = 0;
+
+    while (i < ntimes && GetExitCodeThread(thread, &ts) && ts == STILL_ACTIVE) {


I think you just want WaitForSingleObject(hThread, millis) here.

hcorrada · 2016-03-25T15:39:46Z

With WaitForSingleObject I can take that out. I'll submit another commit soon.

hcorrada · 2016-03-30T12:49:02Z

Hey @jcheng5, thanks for the pointer!

hcorrada · 2016-05-31T16:19:34Z

Hi, @jcheng5, can this be pulled in?

paul-shannon · 2016-12-20T20:26:00Z

Hi Hector,

@hcorrada - I cannot tell if your good work, getting Windows httpuv async communications, R <-> browser, made it into release. Could you let me know?

hcorrada · 2017-01-13T19:08:33Z

Don't think so... @jcheng5 maybe take a look after rstudio conference?

paul-shannon · 2017-01-13T19:58:06Z

Thanks, Hector. Here’s hoping! - Paul

…

On Jan 13, 2017, at 11:08 AM, hcorrada ***@***.***> wrote: Don't think so... @jcheng5 maybe take a look after rstudio conference? — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

jcheng5 · 2017-04-06T17:41:19Z

Sorry for the long, long delay on this. I just felt really uncomfortable with the Windows implementation using threads and still do. I believe I may have found a much cleaner solution using SetTimer though, this doesn't need a background thread at all:
https://github.com/jcheng5/later/blob/master/src/later_win32.cpp

I also need to dive deeply back into httpuv to support asynchronous I/O. I'll try to prioritize Win32 support for daemonized httpuv as well, and getting it to a place where I'm comfortable calling it non-experimental.

paul-shannon · 2017-04-06T18:04:52Z

Hi Joe, This is great news! Daemonized httpuv on all platforms would be a real boon. - Paul

…

On Apr 6, 2017, at 10:41 AM, Joe Cheng ***@***.***> wrote: Sorry for the long, long delay on this. I just felt really uncomfortable with the Windows implementation using threads and still do. I believe I may have found a much cleaner solution using SetTimer though, this doesn't need a background thread at all: https://github.com/jcheng5/later/blob/master/src/later_win32.cpp I also need to dive deeply back into httpuv to support asynchronous I/O. I'll try to prioritize Win32 support for daemonized httpuv as well, and getting it to a place where I'm comfortable calling it non-experimental. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

hcorrada · 2017-04-06T18:33:41Z

Thanks Joe! That would be awesome!

wch · 2018-04-13T19:26:51Z

I think this is no longer necessary because of #106.

jcheng5 and others added 2 commits January 30, 2016 14:28

Import utils::packageVersion

e6056e1

Non-blocking win32 implementation

b6ec849

Fully implement solution for running the libuv default loop without blocking in win32. It is based on Simon Urbanek's 'background' draft package that implements a version of the Rhttpd strategy. Implementation details are in src/httpuv.cpp file.

jcheng5 reviewed Feb 11, 2016
View reviewed changes

Exit thread on WIN32 elegantly

687bc8b

jcheng5 reviewed Mar 23, 2016
View reviewed changes

Use 'WaitForSingleObject' on thread exit

55308a0

wch closed this Apr 13, 2018

Uh oh!

Conversation

hcorrada commented Jan 30, 2016

Uh oh!

jcheng5 Feb 11, 2016

Choose a reason for hiding this comment

Uh oh!

hcorrada Feb 12, 2016

Choose a reason for hiding this comment

Uh oh!

jcheng5 commented Feb 11, 2016

Uh oh!

jcheng5 commented Feb 11, 2016

Uh oh!

hcorrada commented Feb 12, 2016

Uh oh!

jjallaire commented Feb 12, 2016

Uh oh!

hcorrada commented Mar 21, 2016

Uh oh!

jcheng5 Mar 23, 2016

Choose a reason for hiding this comment

Uh oh!

hcorrada commented Mar 25, 2016

Uh oh!

hcorrada commented Mar 30, 2016

Uh oh!

hcorrada commented May 31, 2016

Uh oh!

paul-shannon commented Dec 20, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hcorrada commented Jan 13, 2017

Uh oh!

paul-shannon commented Jan 13, 2017 via email

Uh oh!

jcheng5 commented Apr 6, 2017

Uh oh!

paul-shannon commented Apr 6, 2017 via email

Uh oh!

hcorrada commented Apr 6, 2017

Uh oh!

wch commented Apr 13, 2018

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

paul-shannon commented Dec 20, 2016 •

edited

Loading