Bug 50213

From Bugzilla Helper:
User-Agent: Mozilla/4.0 (compatible; MSIE 5.5; Windows NT 5.0)

Description of problem:
Either php4 or apache is crashing, and not leaving any log trail behind, 
except for weird mysql errors (see below).  Symptom: web browser will 
attempt to load a page hit for at least 30 minutes before giving up with 
error.  Symptom: a custom java app client that hits the web site for 
content does the same.  So far all the reports occur only on SSL protected 
https pages, though this might be a red herring as most page hits in this 
site are SSL.  At least 3 different php4 pages have shown this behaviour.  
Also maybe a red herring as those 3 pages are the most used.


How reproducible:
Sometimes

Steps to Reproduce:
1. hit the SSL pages a lot
2. eventually, very rarely, the page will not load, delaying nearly 
forever as described above.
3.
	

Actual Results:  The page will not load, IE's world will spin for at least 
30 mins, or our java app waiting for the web response will hang.


Expected Results:  Page should load almost immediately.

Additional info:

Recently moved a production web server onto a fresh RH7.1 box.  After 2 
weeks of running there have built up a list of reports of this weird 
client-side problem.  There were no reports of this type of problem on our 
previous RH6.2 PHP3 server.

Versions:
php-4.0.4pl1-9
php-mysql-4.0.4pl1-9
apache-1.3.19-5
mysql-3.23.36-1
mysql-server-3.23.36-1

The apache access log will register the page hit right away as normal for 
one of these problematic hits.  No errors seem to be associated with these 
hits in the apache error log.

The only logged clue I get is mysterious mysql log errors:

010728  0:44:26  Aborted connection 533 to db: 'warehouseone' 
user: 'warehouseone' host: `localhost' (Got an error reading communication 
packets)
010728  0:44:27  Aborted connection 527 to db: 'warehouseone' 
user: 'warehouseone' host: `localhost' (Got an error reading communication 
packets)
010728  0:51:04  Aborted connection 537 to db: 'warehouseone' 
user: 'warehouseone' host: `localhost' (Got an error reading communication 
packets)
010728  0:58:32  Aborted connection 536 to db: 'warehouseone' 
user: 'warehouseone' host: `localhost' (Got an error reading communication 
packets)
010728  2:48:20  Aborted connection 535 to db: 'warehouseone' 
user: 'warehouseone' host: `localhost' (Got an error reading communication 
packets)
010728  3:01:36  Aborted connection 534 to db: 'warehouseone' 
user: 'warehouseone' host: `localhost' (Got an error reading communication 
packets)

There's consistently 100-200 of these errors in the log each day.  This 
could indicate that many problems per day, many of which may be happening 
to end users in which case I would not hear about them.  The reports I get 
are from internal users of the site and my own experiences.

When this bug happened to me personally today, I noted the time I started 
loading the page.  A mysql error did not show up in the log until 1.5 
hours later, and then there were many.  These mysql log errors seem to 
show up in bursts.  The mysql logs don't give me a way to correlate an 
entry with an exact apache hit, but I found it interesting there were no 
errors logged at all for 1.5 hours after my hit attempt.

There are no errors showing up in the php error log either.

Our scripts heavily use mysql and are quite complex.