Re: [PERFORM] [pgsql-hackers-win32] scalability issues on win32

Attached patch solves the problem for me.Didn't see anything of the
problem you described about hangs in my implementation. Probably because
mine exits at the first sign of a linebreak.

//Magnus

>-----Original Message-----
>From: [email protected] 
>[mailto:[email protected]] On Behalf Of 
>Claudio Natoli
>Sent: den 17 december 2004 01:31
>To: Bruce Momjian
>Cc: Merlin Moncure; Tom Lane; Win 32 hackers PGSQL
>Subject: Re: [PERFORM] [pgsql-hackers-win32] scalability 
>issues on win32
>
>
>
>Hi Bruce and all,
>
>sorry guys, I won't have a chance to get to this until after 
>the Christmas break.
>
>If one of the other usual suspects (Dave, Andrew, Magnus, 
>Merlin) can claim it sooner, please do!
>
>Cheers,
>Claudio
>
>
>
>> -----Original Message-----
>> From: Bruce Momjian [mailto:[email protected]]
>> Sent: Friday, 17 December 2004 11:25 AM
>> To: Claudio Natoli
>> Cc: 'Merlin Moncure'; Tom Lane; Win 32 hackers PGSQL
>> Subject: Re: [PERFORM] [pgsql-hackers-win32] scalability 
>> issues on win32
>> 
>> 
>> 
>> Claudio, I need this completed so we can get it into CVS.  
>> Without it a
>> 8.0.0 version string will fail to compare.
>> 
>> --------------------------------------------------------------
>> -------------
>> 
>> Claudio Natoli wrote:
>> > 
>> > Hi Merlin,
>> > 
>> > > On a probably unrelated note:
>> > > Can some other win32 hacker try the following?
>> > > Download the latest cvs sources from the server and edit 
>> configure,
>> > > replacing all instances of '8.0beta5' with '8.0' etc. as is 
>> > > will be when released.  Following that do a full 
>> make/make clean.  After 
>> > > that try and start he server with pg_ctl.exe and tell me 
>> if it succeeds or
>> > not.
>> > 
>> > I can confirm this failure.
>> > 
>> > (in my case at least) pg_ctl is not seeing the full version 
>> string of the
>> > postmaster in find_other_exec. This is due to the single 
>> call to ReadFile in
>> > pipe_read_line, which is not returning the entire version 
>> string (losing all
>> > characters beyond the last decimal point in the version string). A
>> > subsequent call to ReadFile picks up the remainder of the 
>> string. I don't
>> > see that we can rely on a single call returning the entire 
>> version string.
>> > 
>> > If someone can run with this, I've attached initial looping 
>> changes for
>> > pipe_read_line. However I'm not sure of the correct 
>> terminating condition
>> > (gets into a call to ReadFile that never returns; the 
>> WaitForSingleObject
>> > appears insufficient as a precondition for ReadFile) and 
>> just don't have
>> > time to complete it right now.
>> > 
>> > Cheers,
>> > Claudio
>> >


---------------------------(end of broadcast)---------------------------
TIP 3: if posting/reading through Usenet, please send an appropriate
      subscribe-nomail command to [email protected] so that your
      message can get through to the mailing list cleanly

Index: port/exec.c
===================================================================
RCS file: /projects/cvsroot/pgsql/src/port/exec.c,v
retrieving revision 1.33
diff -c -r1.33 exec.c
*** port/exec.c	27 Nov 2004 22:44:15 -0000	1.33
--- port/exec.c	19 Dec 2004 15:42:04 -0000
***************
*** 486,516 ****
  					  &si,
  					  &pi))
  	{
- 		DWORD		bytesread = 0;
-
  		/* Successfully started the process */

  		ZeroMemory(line, maxsize);

! 		/* Let's see if we can read */
! 		if (WaitForSingleObject(childstdoutrddup, 10000) != WAIT_OBJECT_0)
  		{
! 			/* Got timeout */
! 			CloseHandle(pi.hProcess);
! 			CloseHandle(pi.hThread);
! 			CloseHandle(childstdoutwr);
! 			CloseHandle(childstdoutrddup);
! 			return NULL;
  		}

! 		/* We try just once */
! 		if (ReadFile(childstdoutrddup, line, maxsize, &bytesread, NULL) &&
! 			bytesread > 0)
  		{
  			/* So we read some data */
! 			int			len = strlen(line);
  			retval = line;

  			/*
  			 * If EOL is \r\n, convert to just \n. Because stdout is a
  			 * text-mode stream, the \n output by the child process is
--- 486,531 ----
  					  &si,
  					  &pi))
  	{
  		/* Successfully started the process */
+ 		char *lineptr;

  		ZeroMemory(line, maxsize);

! 		/* Read at least one line from the pipe */
! 		for (lineptr = line;lineptr < line+maxsize-1;)
  		{
! 			DWORD		bytesread = 0;
!
! 			/* Let's see if we can read */
! 			if (WaitForSingleObject(childstdoutrddup, 10000) != WAIT_OBJECT_0)
! 				/* Got timeout, but perhaps we got a line already */
! 				break;
!
! 			if (!ReadFile(childstdoutrddup, lineptr, maxsize-(lineptr-line), &bytesread, NULL))
! 				break; /* Error occurred, but perhaps we got a line already */
!
! 			lineptr += strlen(lineptr);
!
! 			if (!bytesread)
! 				break; /* EOF */
!
! 			if (strchr(line, '\n'))
! 				break; /* One or more lines read */
  		}

! 		if (lineptr != line)
  		{
  			/* So we read some data */
! 			int			len;
  			retval = line;

+ 			/* If we got more than one line, cut off after the first \n */
+ 			lineptr = strchr(line,'\n');
+ 			if (lineptr)
+ 				*(lineptr+1) = 0;
+
+ 			len = strlen(line);
+
  			/*
  			 * If EOL is \r\n, convert to just \n. Because stdout is a
  			 * text-mode stream, the \n output by the child process is