[Xitami-Dev] Re: XITAMI CRASH ERROR-URGENT!

[Shivakumar Gopalakrishnan]

Other than actually fixing the code itself, other options are starting the process as a service and set the option to restart itself on failure.
   
  Do not know whether you are trying this already.
   
  Kumar
  

Vishal Kumar Nalla <vishalkumar.nalla at sirvisetti.com> wrote:
  Hi,

Xitami 2.5c is crashing at the following events,can anybody give me a resolution as why this is
happening:

Abort at xisuper:Clean-Up-Temporary-Files: (Ready-To-Go, Not-Busy-Event)
Abort at smttran:Regenerate-Alarm-Event: (Getting-Event, Alarm-Event)
Abort at smttime:Generate-Response-Events:set alarm request (Expect-Event, Timer-Event)

At these three events Xitami is crashing on my system,the above information i have gathered from
the crash log.This is a high priority issue for us, as we dont have options to go for other
webservers.Kindly assist/provide a resolution for the above mentioned crash details.

References to source code for the above events:

smttime:
/************************ GENERATE RESPONSE EVENTS ***********************/

MODULE generate_response_events (THREAD *thread)
{
static long
cur_date = 0, /* Check if clock went backwards */
cur_time = 0;
TIMEREQ
*request; /* Pointer to request in list */
long
sig_date, /* When do we signal */
sig_time, /* the next alarm? */
delay_days, /* Which is in how many days */
delay_csecs; /* and how many centiseconds */

/* Handle a backwards clock adjustment by estimating the difference */
/* and applying this to all time requests. */
smt_set_step ("handle clock adjustment");
if (date_is_future (cur_date, cur_time))
{
date_diff (cur_date, cur_time,
date_now (), time_now (),
&delay_days, &delay_csecs);
for (request = requests.next;
request != (TIMEREQ *) &requests;
request = request-> next)
{
past_date (&request-> exp_date, &request-> exp_time,
delay_days, delay_csecs);
}
}
cur_time = time_now ();
cur_date = date_now ();
sig_date = 0; /* No alarm requests, yet */
sig_time = 0;

/* First pass - generate response events and expire old requests */
smt_set_step ("expire old requests");
for (request = requests.next;
request != (TIMEREQ *) &requests;
request = request-> next)
{
if (request-> exp_date < cur_date
|| (request-> exp_date == cur_date
&& request-> exp_time <= cur_time))
{
send_alarm_reply (request); /* Send back an ALARM to caller */
if (request-> cycles == 1) /* Expire if single-shot */
{
request = request-> prev;
request_destroy (request-> next);
}
/* Else set next expiry time to closest future point... under
* normal circumstances we add the delay date and time. If the
* delay is very short (too short), or the clock was moved
* forwards, this may not be enough. We then take the current
* date and time and add the delay to that.
*/
else
{
future_date (&request-> exp_date, &request-> exp_time,
request-> days, request-> csecs);

if (request-> exp_date < cur_date
|| (request-> exp_date == cur_date
&& request-> exp_time < cur_time))
{
request-> exp_date = cur_date;
request-> exp_time = cur_time;
future_date (&request-> exp_date, &request-> exp_time,
request-> days, request-> csecs);
}
if (request-> cycles > 1)
request-> cycles--;
}
}
}
/* Second pass - calculate time for next signal */
smt_set_step ("calculate next signal");
for (request = requests.next;
request != (TIMEREQ *) &requests;
request = request-> next)
{
/* Keep track of the oldest expiry time, for next timer signal */
if (sig_date > request-> exp_date
|| (sig_date == request-> exp_date
&& sig_time > request-> exp_time)
|| sig_date == 0)
{
sig_date = request-> exp_date;
sig_time = request-> exp_time;
}
}
/* If we had an alarm request, go set it off */
smt_set_step ("set alarm request");
if (sig_date)
{
date_diff (sig_date, sig_time,
cur_date, cur_time,
&delay_days, &delay_csecs);
if (delay_days > 24) /* 24 days are less than 2^31 */
delay_days = 24;

/* Try to set a process alarm (SIGALRM signal) after the desired
* delay. If this fails -- usually because we're on a crippleOS
* -- then we substitute a real timer by a slow SMT loop. This is
* okay on crippleOS'es since these usually don't worry too much
* if one process hogs the CPU.
*/
if (!process_alarm (delay_days * 86400000L + delay_csecs * 10))
SEND (&thread-> queue-> qid, "_TIMER", "");
}
}

static void
send_alarm_reply (TIMEREQ *request)
{
byte
*msg_body; /* Message returned to caller */
int
msg_size; /* Size of formatted msg_body */

/* Calculate size required for message body */
msg_size = exdr_write (NULL, SMT_TIME_REPLY,
request-> body_size, request-> body_data);

/* Now allocate and format the buffer */
msg_body = mem_alloc (msg_size);
msg_size = exdr_write (msg_body, SMT_TIME_REPLY,
request-> body_size, request-> body_data);
event_send (
&request-> reply_to, /* Send to specified queue */
NULL, /* No queue for reply */
"TIME_ALARM", /* Name of event to send */
msg_body, /* Event body contents */
msg_size, /* Event body size */
NULL, NULL, NULL, /* No response events */
0); /* No timeout */

mem_free (msg_body); /* And we can free the buffer */
}


/* -------------------------------------------------------------------------
* request_destroy
*
* Destroys the specified request.
*/

static void
request_destroy (TIMEREQ *request)
{
mem_free (request-> body_data);
node_destroy (request);
}


smttran:

/************************* REGENERATE ALARM EVENT ************************/

MODULE regenerate_alarm_event (THREAD *thread)
{
struct_smt_time_reply
*reply; /* Reply from timer */

tcb = thread-> tcb; /* Point to thread's context */
get_smt_time_reply (thread-> event-> body, &reply);
if (streq ((char *) reply-> tag_data, "COMMIT"))
the_next_event = commit_event;
else
the_next_event = empty_event;

free_smt_time_reply (&reply);

}

xisuper:

/************************ CLEAN UP TEMPORARY FILES ***********************/

MODULE clean_up_temporary_files (THREAD *thread)
{
SYMBOL
*symbol;

/* We get the list of temp file directories (potentially a different one
* for each virtual host) from the http_dirs symbol table. We delete
* temporary files older than one day.
*/
for (symbol = http_dirs-> symbols; symbol; symbol = symbol-> next)
cleanup_directory (symbol-> name, "pipe*", 1);
}


/************************* TERMINATE THE THREAD **************************/

MODULE terminate_the_thread (THREAD *thread)
{
the_next_event = terminate_event;
}


-- 
Thanks
Vishal Kumar Nalla

On Sun, November 19, 2006 00:29, Roman Stepanov wrote:
> Sorry, but i didn't receive any attachments with your mail.
> About setting the time - works fine, without any crashes.
> I have a fresh copy of xitami installed from http://www.xitami.com/xiopen_2_5.zip
> If your copy has been downloaded from another source, please send me url.
>
>
>
> Saturday, November 18, 2006, 8:13:59 PM, you wrote:
>
>> Hi Everyone,
>
>> Can you please look into these log files and let me know what exactly is happening from
>> Xitami.This is haulting our process,kindly provide a solution as soon as possible.
>
>
>
>
>
>
>
>





-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.imatix.org/pipermail/xitami-dev/attachments/20070406/8b0d5ad7/attachment.htm