On the graph: "APED Restart" is a cron killing script ...

On Nov 15, 9:37 pm, Mariusz <[email protected]> wrote:
> Hi,
>
> http://p.web-album.org/5f/9e/5f9e86e29e1612c2fcaa373b6ea02b25a,4,0.png
> Here is the graph from my munin's ape plugin.
>
> I pushing messages (~2500 msg/day, about 1KB each POST) to ape server
> (APE 1.0) via PHP [Curl] to my server-side command. I use only
> methods:
> Ape.config()
> Ape.getChannelByName()
> chan.pipe.sendRaw();
>
> I think there is a big problem ...
>
> On Nov 15, 5:42 pm, <[email protected]> wrote:
>
> > Hey,
>
> > Are you experiencing the same issue ?
> > Can you provide an testcase and/or some informations about your
> > configuration (number of online user, etc...)
>
> > Thanks ;)
>
> > Anthony
>
> > On Mon, 15 Nov 2010 08:24:14 -0800 (PST), Doug Molineux
>
> > <[email protected]> wrote:
> > > I was wondering, if there was any progress on this issue?
>
> > > Thanks,
> > > Doug
>
> > > On Nov 9, 12:10 pm, <[email protected]> wrote:
> > >> Hey,
>
> > >> I just finished my websocket benchmaker tools.
> > >> I connected 10k users, make them join a public channel. Everything
> > looks
> > >> good :/ Nomemoryleak.
> > >> I'll dig more again.
>
> > >> Btw, can you give me an IM address where we can talk ?
>
> > >> On Tue, 9 Nov 2010 12:47:04 -0500, Felix Filozov <[email protected]>
> > >> wrote:
>
> > >> > Here are the results of my tests
> > >> > Test 1 - Joins to private
> > >> > channel====================================12K
> > >> > joins within 5 minutes - 4% ofmemoryusage.
> > >> > Test 2 - Joins to public channel
> > >> > =====================================12K joins within 5 minutes - 45%
> > >> > of
> > >> >memoryusage.
> > >> > Clearly, joining a private channel consumes lessmemory. I reran Test
> > 1
> > >> > again just to make sure, and the results were consistent. 
> > >> > However, after both tests completed,memorywas not released. Throwing
> > >> > morememoryat it is not a problem, but restarting it could be an
> > issue.
> > >> > Hopefully these results help.
> > >> > - Felix
> > >> > On Mon, Nov 8, 2010 at 9:30 PM,  wrote:
> > >> >  Ok, thanks for the informations.
> > >> >  Can you please try the same test on a private channel ? (channel
> > with
> > >> > "*"
> > >> >  as first character).
>
> > >> >  With my benchmark I successfuly connected 30k users with about 150MB
> > >> > of
> > >> >  memory.
> > >> >  BTW I used a long polling transport (which is a lot more expensive
> > >> > than
> > >> >  WebSocket).
>
> > >> >  Let me write a websocket tools and run the same bench.
>
> > >> >  I keep you in touch.
>
> > >> >  On Mon, 8 Nov 2010 20:05:47 -0500, Felix Filozov
> > >> >  wrote:
> > >> >  > I only have access to the server side, and, using a JOIN hook, I
> > >> >  > determined that there have been 9000+ JOINs within 5 minutes.
> > >> >  > I'm pretty sure the browser receive a successful CHANNEL raw,
> > >> > because
> > >> > in
> > >> >  > my previous tests that was the case.
> > >> >  > On Mon, Nov 8, 2010 at 7:40 PM,  wrote:
> > >> >  >  I see.
>
> > >> >  >  Do you have any rate of browser succesfully receiving the
> > >> > "CHANNEL"
> > >> > and
> > >> >  >  "JOIN" raws on joining ?
>
> > >> >  >  Thanks
>
> > >> >  >  Anthony
>
> > >> >  >  On Mon, 8 Nov 2010 19:22:14 -0500, Felix Filozov
> > >> >  >  wrote:
> > >> >  >  > I'm using BrowserMob to spawn thousands of browsers, with in 5
> > >> >  > minutes,
> > >> >  >  > that visit a URL. Each browser is open for at most 4 seconds.
> > >> >  > Meaning,
> > >> >  >  > within 5 minutes there's many browsers opening and closing.
> > >> >  >  >
> > >> >  >  > The URL contains a Flex application, which connects to APE
> > >> > (jaeger
> > >> >  >  > branch) using a WebSocket.
> > >> >  >  >
> > >> >  >  > If you want, I can direct the traffic to you.
> > >> >  >  >
> > >> >  >  > On Mon, Nov 8, 2010 at 6:58 PM,  wrote:
> > >> >  >  >  I've just tryed what you've said :
> > >> >  >  >
> > >> >  >  >  - Make 10k user Joining a public channel
> > >> >  >  >  - ???
> > >> >  >  >  - Timeout for all user (leading to a LEFT for each user)
> > >> >  >  >
> > >> >  >  >  Memoryconsuption is still about few MB on my machine.
> > >> >  >  >
> > >> >  >  >  How do you run the bench ?
> > >> >  >  >  Can you send me the tools/script you are using ?
> > >> >  >  >
> > >> >  >  >  Thanks
> > >> >  >  >
> > >> >  >  >  Anthony
> > >> >  >  >
> > >> >  >  >  On Mon, 8 Nov 2010 13:11:37 -0500, Felix Filozov
> > >> >  >  >  wrote:
> > >> >  >  >  > I just ran a test of 9000+ joins (they joined for a few
> > >> > seconds
> > >> >  > then
> > >> >  >  >  > left) to APE within 5 minutes, and thememoryconsumption
> > >> > went up
> > >> >  > to
> > >> >  >  >  38%.
> > >> >  >  >  > It stayed there even after everybody left.
> > >> >  >  >  >
> > >> >  >  >  > On Fri, Nov 5, 2010 at 9:06 PM, Felix Filozov  wrote:
> > >> >  >  >  >  I had my units wrong. It's 2.4%, not 2.4MB. 
> > >> >  >  >  > I will do an exact test, where I only test connection
> > >> > attempts. I
> > >> >  >  >  > believe this is where the leak occurs, because message
> > >> >  >  >  > passing occurred infrequently during my previous tests.
> > >> >  >  >  >
> > >> >  >  >  > On Fri, Nov 5, 2010 at 8:38 PM,  wrote:
> > >> >  >  >  >  There are several reason where thememoryisn't restored
> > >> > back :
> > >> >  >  >  >
> > >> >  >  >  >  - Spidermonkey Garbage collector
> > >> >  >  >  >  - APE allocate a pool of socket for concurent users with a
> > >> >  > "n*2"
> > >> >  >  > schema
> > >> >  >  >  >  :
> > >> >  >  >  >
> > >> >  >  >  >  Basememoryis 2 sockets slot :
> > >> >  >  >  >
> > >> >  >  >  >  - 2 users connect => nomemoryallocation
> > >> >  >  >  >  - 1 more user connects : Basememory* 2 (4 slot reserved)
> > >> >  >  >  >  - 2 more user connect (5 active users) => Basememory* 2
> > >> > (8
> > >> >  > slot
> > >> >  >  >  >  reserved)
> > >> >  >  >  >  - And so forth.
> > >> >  >  >  >
> > >> >  >  >  >  This scheme avoid APE to malloc/free too often.
> > >> >  >  >  >
> > >> >  >  >  >  The fact is , does yourmemoryincrease because of message
> > >> >  > sent, or
> > >> >  >  >  >  because of user connection (there is amemoryleak if your
> > >> >  > 2.4MB
> > >> >  >  > was
> > >> >  >  >  > caused
> > >> >  >  >  >  by the first).
> > >> >  >  >  >
> > >> >  >  >  >  On Fri, 5 Nov 2010 19:20:32 -0400, Felix Filozov
> > >> >  >  >  >  wrote:
> > >> >  >  >  >  > I just tried the jaegerape branch, and the result is the
> > >> >  > same. It
> > >> >  >  >  > climbs
> > >> >  >  >  >  > to 2.4MB of usage, and stays there.
> > >> >  >  >  >  > I'll be more than willing to direct the traffic to one of
> > >> >  > your
> > >> >  >  >  >  > installations.
> > >> >  >  >  >  >
> > >> >  >  >  >  >  On Fri, Nov 5, 2010 at 4:17 PM,  wrote:
> > >> >  >  >  >  >  Hey,
> > >> >  >  >  >  >
> > >> >  >  >  >  >  Your APE is compiled using some compile optimization
> > >> > (-O2),
> > >> >  >  > leading
> > >> >  >  >  > to
> > >> >  >  >  >  >  some invalid valgrind output.
> > >> >  >  >  >  >  BTW, can you please check if you still have the same
> > >> >  > behaviour
> > >> >  >  >  > using
> > >> >  >  >  >  > the
> > >> >  >  >  >  >  "jaegerape" branch :
> > >> >  >  >  >  >
> > >> >  https://github.com/APE-Project/APE_Server/commits/jaegerape[3]
> > >> >  > [3] [3]
> > >> >  >  > [4] [2]
> > >> >  >  >  >  >
> > >> >  >  >  >  >  Thanks
> > >> >  >  >  >  >
> > >> >  >  >  >  >  Anthony
> > >> >  >  >  >  >
> > >> >  >  >  >  >  On Thu, 4 Nov 2010 17:29:56 -0400, Felix Filozov
> > >> >  >  >  >  >  wrote:
> > >> >  >  >  >  >  > Here's some Valgrind output I captured
> > >> >  >  >  >http://pastebin.com/uDcp43Xf[4] [4] [4] [5]
> > >> >  >  >  >  > [4] [1]
> > >> >  >  >  >  >  >
> > >> >  >  >  >  >  > On Thu, Nov 4, 2010 at 3:25 PM, Felix Filozov
> > >> >  wrote:
> > >> >  >  >  >  >  >  I just used the Move demo to do a test on APE 1.0
> > >> > and
> > >> >  > APE
> > >> >  >  > Git
> > >> >  >  >  >  > (latest
> > >> >  >  >  >  >  > APE in Git).
> > >> >  >  >  >  >  >
> > >> >  >  >  >  >  > I had thousands of connection attempts to APE 1.0,
> > >> > and
> > >> >  >memory
> > >> >  >  >  >  > increased
> > >> >  >  >  >  >  > to 4MB and it stayed there even after all those
> > >> >  > connections
> > >> >  >  > were
> > >> >  >  >  >  > broken.
> > >> >  >  >  >  >  >
> > >> >  >  >  >  >  > Similarly, I had thousands of connection attempts to
> > >> > APE
> > >> >  > Git,
> > >> >  >  > and
> > >> >  >  >  >  >memory
> > >> >  >  >  >  >  > increased to 2MB and it stayed there as well.
> > >> >  >  >  >  >  >
> > >> >  >  >  >  >  > I don't know ifmemoryis immediately deallocated, or
> > >> >  > later
> > >> >  >  > on.
> > >> >  >  >  > If
> > >> >  >  >  >  > the
> > >> >  >  >  >  >  > former, it doesn't seem to work.
> > >> >  >  >  >  >  >
> > >> >  >  >  >  >  > On Thu, Nov 4, 2010 at 2:28 PM, Johnathan Leppert
> > >> >  >  wrote:
> > >> >  >  >  >  >  >  What version are you running?
> > >> >  >  >  >  >  >
> > >> >  >  >  >  >  > On Thu, Nov 4, 2010 at 2:25 PM, Ecrofom  wrote:
> > >> >  >  >  >  >  >   I believe I'm experiencing the same thing. After
> > >> > about
> > >> >  > 24
> > >> >  >  >  > hours, I
> > >> >  >  >  >  >  >  noticed ape was using over 300MB of ram. This
> > >> > didn't
> > >> >  > happen
> > >> >  >  >  > before
> > >> >  >  >  >  > but
> > >> >  >  >  >  >  >  I did change how I was using ape.
> > >> >  >  >  >  >  >
> > >> >  >  >  >  >  >  Before: The default standard setup as shown in the
> > >> > chat
> > >> >  >  > demo on
> > >> >  >  >  >  > your
> > >> >  >  >  >  >  >  site.Memoryusage was pretty much constant and
> > >> > low.
> > >> >  >  >  >  >  >
> > >> >  >  >  >  >  >
>
> ...
>
> read more »

-- 
You received this message because you are subscribed to the Google
Groups "APE Project" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/ape-project?hl=en
---
APE Project (Ajax Push Engine)
Official website : http://www.ape-project.org/
Git Hub : http://github.com/APE-Project/

Reply via email to