All lsphp processes die - users get "no data"

[Karl]

Hi,

We have an issue on one of our servers running Litespeed Web Server. Every so often all lsphp5 processes will die and the server effectively stops responding until they come back again.

This is on 4.2.4, it's a busy server so the debug log grows incredibly quickly (15,000 lines covers about 4s of activity). So it's hard to see if anything is going on in there, as there's no pattern to it - unless we know what to look for in the log.

There's nothing in the normal error log, nothing logged with an info, warning or error status in the log either. There's no core dumps, the main litespeed process does not die.

Thanks,

Karl

 

[mistwang]

Are you using PHP suEXEC daemon mode? Sounds like something wrong with shared opcode cache.
4.2.4 should log the reason why PHP processes die if they are crashed, just set log level at "info".
To let PHP dump core files, you need to set "LSAPI_ALLOW_CORE_DUMP" env to lsphp5 external app.
http://www.litespeedtech.com/php-litespeed-sapi.html

 

[Karl]

All Opcode caches are disabled, the system isn't recording any segfaults - it was on PHP 5.2.17 (< 10 per day, but that didn't tally with the "no data" responses), but with 5.3.27 (customer can't use anything newer) no segfaults are recorded.

 

[NiteWave]

one possible reason, is lsphp not built properly, please just access phpinfo.php, and watch stderr.log or error.log, if any error message there.

 

[NiteWave]

or simply run at command line:
#/usr/local/lsws/fcgi-bin/lsphp5 -V

see if any error message

 

[Karl]

Hi,

No error message. For 98% of requests it runs fine, then something will cause it to kill every single lsphp5 process.

Thanks,

 

[mistwang]

can you truncate error_log and stderr.log, set log level at "INFO", restart LSWS, wait for "no data" happen. record the time when it happens, send the log to bug@litespeed... for analysis.