c++ - pselect blocks even though data is available for read on socket -

-

I am experiencing an intermittent delay while reading from POSIS Socket (RHEL 6x86_64C ++ ICPC).

My code is designed so that the user can provide a full time-limit (a relative timeout) that Many call can be used for recv. I ask PECEL to make sure that the data is available for reading before reading.

This usually works as expected (waiting for data, but not exceeding the deadline, no noticeable delay can be offered when available to retrieve the data ). However, I have a user who can get his application from time to time (~ 50% of the time) in a state where ~ 400-500 ms for selection block though data is available on the socket. If I look at / proc / net / tcp, then I can see that the data is available in the RX queue and I can slowly read the queued data after seeing the application. If I leave to choose the call and only call books, then the behavior is the same (but at least delay indicates that RRV is also blocking unnecessarily) When the application comes in this situation It stays in this way (experiences constant delay in each Pse Strait / REV).

I have spent many hours here and on other sites. This is the closest problem I could have found, but there was no resolution ...

Has anyone played this kind of behavior before? What I'm at a loss to do is I used the code to confirm that this is where the delay is happening (EDIT: We have actually validated that the whole method below was slow, someone There was no special system call.) It seems that there is a kernel / OS problem but I'm not sure where to see me. Here's the code ... 

  // Protected Balls Message: Wait (int socket, constant timepak and deadline) {// If deadline has not been provided (deadline.tv_sec = = 0 & amp; amp; amp; amp; & amp; deadline.tv_nsec == 0) {back true; } // Make sure that we have not already crossed the timeline deadline; Clock_gettime (CLOCK_REALTIME, and current time); If (VirtualClock :: Semietime (current time, deadline)> = 0) {LOG_WARNING ("Waiting for data to finish waiting"); M_timedOut = True; return false; } // Get timeout timespec get timeout; Mmateset (& amp; expiration, 0, size (timeout)); Timeout.tv_nsec = VirtualClock :: nsecs (current time, deadline); VirtualClock :: fixtime (timeout); // Wait for data fs_set descSet; FD_ZERO (& descSet); FD_SET (socket, and disassetset); Favored Results = Pixels (socket + 1, and disassette, faucet, tap, and timeout, tap); If (result == -1) {m_error = errno; LOG_ERROR ("Failed to wait for data:% d,% s", m_error, strerror (m_error)); return false; } And if (Results == 0 ||! FD_ISSET (socket, and disassette)) {LOG_WARNING ("Time to wait for data to be found"); M_timedOut = True; return false; } Back true; }

VirtualClock is a time-related utility class that is used to compare here / fix-up times (i.e. not starting any delays). I appreciate any information on this behavior.

This was actually not a problem with any system call. We used the strokes to find out and were seeing multiple calls for clock_gettime, a programming error was seen in another (third) review of the calling code, which resulted in the result of the code called in the context of corrupt stack data. This helps by a faulty API design on my behalf, which results in corruption of timeframe.

I was allowing the user to pass in the context of configuring the server configured config (including the last time related data). My server was saving the reference instead of copying the class object. The user made an example of his server class on the heap, a server on the stack was passed in conferencing (in context), resulting in non-determining garbage in the configuration when the method exits and ServerConfig exits from the scope. This is the old code and I have prevented this type of thing from burning at other places, but it has slipped through one.

Then the lesson for me has been learned: For references provided by the user, from time to time reconsidering optimization (instead of just making a copy, I was hanging on a reference, the whole reason) , And when you see non-deterministic behaviors like this (I think that when I suspect the builds have been lifted but this time it has not happened). Besides, Strace is a great tool ... I've seen others use it, but now I'm comfortable using it myself.

Thanks for the comment and sorry for the false alarms.


Comments

Post a Comment

Popular posts from this blog

HTML/CSS - Automatically set height width from background image? -

-
Long time ... I do not think it's possible, but I'm shocked before! I tag anchor tags, all of which are background images, all 300px wide but their height is all the difference, is it really set to do without typing altitude? Is it to set the dimensions of BG URL? Thank you! I do not think people understand - my mistake is to hurry the question. Here is the code as an example: # ago -1 {width: 300px; Height: 410px; Background: url ('/ image-1.jpg');} # East -2 {width: 300px; Height: 420px; Background: url ('/ image-2.jpg');} # Pre-3 {width: 300px; Height: 430px; Background: url ('/ image-3.jpg');} # Pre-4 {width: 300px; Height: 440px; Background: url ('/ image-3.jpg');} & lt; A href = "#" id = "ex-1" & gt; & Lt; / A & gt; & Lt; A href = "#" id = "ex-2" & gt; & Lt; / A & gt; & Lt; A href = "#" id = "pre-3" & gt; & Lt; / A & gt...
Read more

php - Mysql Show Process - Sleep Commands and what to do -

-
I am thinking that there is something that I can do better so that my process list should start with the command with MySQL Be clear Currently I am looking for a lot: 17325 user_a localhost db_1 sleep 1132 NULL 17464 user_a localhost db_1 sleep 1124 NULL 17983 user_a localhost db_1 sleep 1078 null 18113 user_a localhost db_1 sleep 1068 NULL 18207 user_a localhost db_1 sleep 1060 NULL 18231 user_a localhost db_1 sleep 1058 NULL 18353 user_a localhost db_1 sleep 1047 NULL 18447 user_a localhost db_1 sleep 1040 null 18489 user_a localhost db_1 sleep 1036 NULL 23408 user_a localhost db_1 sleep 637 NULL Something in my phpscript is that I want to clean it Can i The script looks like: Select the ID from $ q = 'db_1 where fkId = 2'; $ Ar = mysql_query ($ q); ($ Line = mysql_fetch_array ($ qr)) {echo ($ row ['id'];} Thanks in advance! I'm guessing that when you have 10 threads, you're constantly using the connectio...
Read more

c - What is the address of buf (the local variable in the main function)? -

-
I had exercise with programming in c or classes and I followed the question: Is Buf (local variable in main function)? Either enter the answer in hexadecimal format (like 0x 8 "digits" 0-9 or A-F, 0xbfff0014) or in decimal format. Note here that we want the address of the buff, not its content. The code looks like this: int main (int argc, char * argv []) {while (1) {four buff [1024] = {0}; Int r; .... I then started the code and put the built code in the buff [1024] line and in Typed Gdb p & amp; Buf and it gives me results: 0xbffff0f0 But when I use it in this quiz it gives me the result that this value is wrong. My question is this: Is this the name of the Buff variable (P & E)? Or if not, why not? Exercise is done on the virtualbox machine, so I think everyone should be the same address There is no correct numerical answer, the address may vary from one time to another, it is a programming drill that you say the following three...
Read more