Great, please report back if/when you discover the cause.
Meanwhile doing "man 2 flock" yields info about ENOLCK: "The kernel ran out
of memory for allocating lock records." This concerns your new server. The
old one does not support locking, or so it seems (EOPNOTSUPP).
b.
On 28 March 2011 23:15, Rob Adams <rad...@circlepix.com> wrote:
> Bostjan,
>
> Thanks again for the tip. At this point, it looks like there is an issue
> with my main file server. It's weird that the older servers were working
> with it fine though (by instantly failing on lock), so I originally thought
> it might be a php issue. I think I'm making progress on it now though.
>
> -- Rob
>
>
>
> On 3/28/2011 12:08 PM, Rob Adams wrote:
>
>> On 3/26/2011 11:07 AM, Bostjan Skufca wrote:
>>
>>> Can you strace it's execution and see where your delay is comming
>>> from? If
>>> you are using apache, make it create just one child and strace that
>>> one when
>>> you generate a request.
>>>
>>
>> Thanks for the advice. I installed strace on the new server, and it has
>> a thirty second delay on the flock call (opening the session file):
>>
>> open("/path/to/session/sess_SESSIONID", O_RDWR|O_CREAT, 0600) = 20
>> flock(20, LOCK_EX) = -1 ENOLCK (No locks available)
>> fcntl64(20, F_SETFD, FD_CLOEXEC) = 0
>> fstat64(20, {st_mode=S_IFREG|0644, st_size=13, ...}) = 0
>> pread64(20, "cpixUser|b:1;", 13, 0) = 13
>> gettimeofday({1301331125, 622954}, NULL) = 0
>> gettimeofday({1301331125, 623046}, NULL) = 0
>> brk(0x989d000) = 0x989d000
>> gettimeofday({1301331125, 623301}, NULL) = 0
>> pwrite64(20, "cpixUser|b:1;", 13, 0) = 13
>> close(20) = 0
>>
>> This has the correct session file name (ie
>> "/path/to/session/sess_SESSIONID").
>>
>> So then I installed strace on one of the older servers, to see what it
>> was doing, and this is what I see where it is opening the session file:
>>
>> open("\"M\235\240GU\21", O_RDWR|O_CREAT, 0600) = 9
>> flock(9, LOCK_EX) = -1 EOPNOTSUPP (Operation not supported)
>> fcntl(9, F_SETFD, FD_CLOEXEC) = 0
>> fstat(9, {st_mode=S_IFCHR|040, st_rdev=makedev(113, 1114152), ...}) = 0
>> lseek(9, 0, SEEK_SET) = 0
>> read(9, "cpixUser|b:1;"..., 13) = 13
>> gettimeofday({540876916, 1078020164}, {3221773391, 671636288}) = 0
>> sigprocmask(SIG_BLOCK, NULL, []) = 0
>> lseek(9, 0, SEEK_SET) = 0
>> write(9, "cpixUser|b:1;"..., 13) = 13
>> close(9) = 0
>>
>> I don't know why it has the session file as this: "\"M\235\240GU\21".
>>
>> So it looks like the flock(?) is taking 30 seconds to timeout on the new
>> server, and failing instantly on the old ones. On the new server, I have
>> a message in /var/log/messages that reads:
>>
>> Mar 28 10:38:34 webServer kernel: lockd: server mainFileServer not
>> responding, timed out
>>
>> I get this message evertime I call session_start(). I have tried
>> reloading the new server from scratch (CentOS 5.5, php 5.2.17, apache
>> 2.2) and it is still doing the same thing.
>>
>>
>>
>>
>
> --
> PHP General Mailing List (http://www.php.net/)
> To unsubscribe, visit: http://www.php.net/unsub.php
>
>
