I think I figured out why Cog some times terminates when downloading large files on OSX; An oversleep which is interrupted can be returned as a t_time struct with tv_sec -1 and tv_nsec close to 999999999.
The while check is nanosleep (&naptime, &naptime) == -1 && naptime.tv_sec > 0 1|| naptime.tv_nsec > MINSLEEPNS
or so, so when an interrupt is encountered which has tv_sec = -1, you will get the invalid argument error.
rewriting it to naptime.tvsec > -1 && naptime.tv_nsec > MINSLEEPNS && nanosleep(&naptime, &naptime) == -1
I no longer had crashes in 99% of hte cases when evaluating the following test:
HTTPSocket httpGetDocument: 'http://www.squeaksource.com/Pharo/Morphic-stephane_ducasse.334.mcz'.
I've attached the sqUnixHeartbeat.c.
Cheers, Henry
Thanks Henrik!!
On Fri, Aug 13, 2010 at 7:38 AM, Henrik Johansen < henrik.s.johansen@veloxit.no> wrote:
I think I figured out why Cog some times terminates when downloading large files on OSX; An oversleep which is interrupted can be returned as a t_time struct with tv_sec -1 and tv_nsec close to 999999999.
The while check is nanosleep (&naptime, &naptime) == -1 && naptime.tv_sec > 0 1|| naptime.tv_nsec > MINSLEEPNS
or so, so when an interrupt is encountered which has tv_sec = -1, you will get the invalid argument error.
rewriting it to naptime.tvsec > -1 && naptime.tv_nsec > MINSLEEPNS && nanosleep(&naptime, &naptime) == -1
I no longer had crashes in 99% of hte cases when evaluating the following test:
HTTPSocket httpGetDocument: ' http://www.squeaksource.com/Pharo/Morphic-stephane_ducasse.334.mcz'.
I've attached the sqUnixHeartbeat.c.
Cheers, Henry
Thank you! I can confirm that this solves the last remaining Seaside issue.
I simulated some heavy parallel requests on a Cog-Seaside image and it doesn't crash anymore.
Lukas
On 13 August 2010 21:20, Eliot Miranda eliot.miranda@gmail.com wrote:
Thanks Henrik!!
On Fri, Aug 13, 2010 at 7:38 AM, Henrik Johansen henrik.s.johansen@veloxit.no wrote:
I think I figured out why Cog some times terminates when downloading large files on OSX; An oversleep which is interrupted can be returned as a t_time struct with tv_sec -1 and tv_nsec close to 999999999.
The while check is nanosleep (&naptime, &naptime) == -1 && naptime.tv_sec > 0 1|| naptime.tv_nsec > MINSLEEPNS
or so, so when an interrupt is encountered which has tv_sec = -1, you will get the invalid argument error.
rewriting it to naptime.tvsec > -1 && naptime.tv_nsec > MINSLEEPNS && nanosleep(&naptime, &naptime) == -1
I no longer had crashes in 99% of hte cases when evaluating the following test:
HTTPSocket httpGetDocument: 'http://www.squeaksource.com/Pharo/Morphic-stephane_ducasse.334.mcz'.
I've attached the sqUnixHeartbeat.c.
Cheers, Henry
On 16 August 2010 16:48, Lukas Renggli renggli@gmail.com wrote:
Thank you! I can confirm that this solves the last remaining Seaside issue.
I simulated some heavy parallel requests on a Cog-Seaside image and it doesn't crash anymore.
Lukas, could you share details how you simulating heavy load? I know there are some tools on linux to flood server with HTTP requests , but i never used it myself.
Lukas
On 13 August 2010 21:20, Eliot Miranda eliot.miranda@gmail.com wrote:
Thanks Henrik!!
On Fri, Aug 13, 2010 at 7:38 AM, Henrik Johansen henrik.s.johansen@veloxit.no wrote:
I think I figured out why Cog some times terminates when downloading large files on OSX; An oversleep which is interrupted can be returned as a t_time struct with tv_sec -1 and tv_nsec close to 999999999.
The while check is nanosleep (&naptime, &naptime) == -1 && naptime.tv_sec > 0 1|| naptime.tv_nsec > MINSLEEPNS
or so, so when an interrupt is encountered which has tv_sec = -1, you will get the invalid argument error.
rewriting it to naptime.tvsec > -1 && naptime.tv_nsec > MINSLEEPNS && nanosleep(&naptime, &naptime) == -1
I no longer had crashes in 99% of hte cases when evaluating the following test:
HTTPSocket httpGetDocument: 'http://www.squeaksource.com/Pharo/Morphic-stephane_ducasse.334.mcz'.
I've attached the sqUnixHeartbeat.c.
Cheers, Henry
-- Lukas Renggli www.lukas-renggli.ch
Lukas, could you share details how you simulating heavy load? I know there are some tools on linux to flood server with HTTP requests , but i never used it myself.
I used 'ab', the Apache HTTP server benchmarking tool. That works out of the box on any machine with Apache installed and additionally gives a nice performance report. In this case however I just wanted to generate load to crash the VM (which I didn't manage).
For this test I used the following code using a single session I created beforehand in the web browser.
ab -n 1000 -c 10 "http://127.0.0.1:8080/tests/functional/WAHtml5ElementsTest?_s=As2WXbb5pqm18x..."
Another easy thing to do is to fork multiple instances of 'wget' and tell it to randomly browse through the links of a web application.
wget --recursive --no-parent --delete-after "http://127.0.0.1:8081/examples/multicounter"
Each of the started processes then clicks through an individual session. It is important to disable the toolbar, otherwise 'wget' starts to mess around with your image (through the configuration).
Previously I've also used JMeter. This tool is much more controllable and can be scripted to click through an application in a predefined way, but it also means a lot more work to setup.
Lukas
On 16 August 2010 17:32, Igor Stasenko siguctua@gmail.com wrote:
On 16 August 2010 16:48, Lukas Renggli renggli@gmail.com wrote:
Thank you! I can confirm that this solves the last remaining Seaside issue.
I simulated some heavy parallel requests on a Cog-Seaside image and it doesn't crash anymore.
Lukas, could you share details how you simulating heavy load? I know there are some tools on linux to flood server with HTTP requests , but i never used it myself.
Lukas
On 13 August 2010 21:20, Eliot Miranda eliot.miranda@gmail.com wrote:
Thanks Henrik!!
On Fri, Aug 13, 2010 at 7:38 AM, Henrik Johansen henrik.s.johansen@veloxit.no wrote:
I think I figured out why Cog some times terminates when downloading large files on OSX; An oversleep which is interrupted can be returned as a t_time struct with tv_sec -1 and tv_nsec close to 999999999.
The while check is nanosleep (&naptime, &naptime) == -1 && naptime.tv_sec > 0 1|| naptime.tv_nsec > MINSLEEPNS
or so, so when an interrupt is encountered which has tv_sec = -1, you will get the invalid argument error.
rewriting it to naptime.tvsec > -1 && naptime.tv_nsec > MINSLEEPNS && nanosleep(&naptime, &naptime) == -1
I no longer had crashes in 99% of hte cases when evaluating the following test:
HTTPSocket httpGetDocument: 'http://www.squeaksource.com/Pharo/Morphic-stephane_ducasse.334.mcz'.
I've attached the sqUnixHeartbeat.c.
Cheers, Henry
-- Lukas Renggli www.lukas-renggli.ch
-- Best regards, Igor Stasenko AKA sig.
Thanks a lot!
On 16 August 2010 19:22, Lukas Renggli renggli@gmail.com wrote:
Lukas, could you share details how you simulating heavy load? I know there are some tools on linux to flood server with HTTP requests , but i never used it myself.
I used 'ab', the Apache HTTP server benchmarking tool. That works out of the box on any machine with Apache installed and additionally gives a nice performance report. In this case however I just wanted to generate load to crash the VM (which I didn't manage).
For this test I used the following code using a single session I created beforehand in the web browser.
ab -n 1000 -c 10 "http://127.0.0.1:8080/tests/functional/WAHtml5ElementsTest?_s=As2WXbb5pqm18x..."
Another easy thing to do is to fork multiple instances of 'wget' and tell it to randomly browse through the links of a web application.
wget --recursive --no-parent --delete-after "http://127.0.0.1:8081/examples/multicounter"
Each of the started processes then clicks through an individual session. It is important to disable the toolbar, otherwise 'wget' starts to mess around with your image (through the configuration).
Previously I've also used JMeter. This tool is much more controllable and can be scripted to click through an application in a predefined way, but it also means a lot more work to setup.
Lukas
On 16 August 2010 17:32, Igor Stasenko siguctua@gmail.com wrote:
On 16 August 2010 16:48, Lukas Renggli renggli@gmail.com wrote:
Thank you! I can confirm that this solves the last remaining Seaside issue.
I simulated some heavy parallel requests on a Cog-Seaside image and it doesn't crash anymore.
Lukas, could you share details how you simulating heavy load? I know there are some tools on linux to flood server with HTTP requests , but i never used it myself.
Lukas
On 13 August 2010 21:20, Eliot Miranda eliot.miranda@gmail.com wrote:
Thanks Henrik!!
On Fri, Aug 13, 2010 at 7:38 AM, Henrik Johansen henrik.s.johansen@veloxit.no wrote:
I think I figured out why Cog some times terminates when downloading large files on OSX; An oversleep which is interrupted can be returned as a t_time struct with tv_sec -1 and tv_nsec close to 999999999.
The while check is nanosleep (&naptime, &naptime) == -1 && naptime.tv_sec > 0 1|| naptime.tv_nsec > MINSLEEPNS
or so, so when an interrupt is encountered which has tv_sec = -1, you will get the invalid argument error.
rewriting it to naptime.tvsec > -1 && naptime.tv_nsec > MINSLEEPNS && nanosleep(&naptime, &naptime) == -1
I no longer had crashes in 99% of hte cases when evaluating the following test:
HTTPSocket httpGetDocument: 'http://www.squeaksource.com/Pharo/Morphic-stephane_ducasse.334.mcz'.
I've attached the sqUnixHeartbeat.c.
Cheers, Henry
-- Lukas Renggli www.lukas-renggli.ch
-- Best regards, Igor Stasenko AKA sig.
-- Lukas Renggli www.lukas-renggli.ch
vm-dev@lists.squeakfoundation.org
-
Eliot Miranda -
Henrik Johansen -
Igor Stasenko -
Lukas Renggli