Hi Jay,
not sure if this is connected, but we've had similar problems with
ncWMS/thredds some years ago
http://www.unidata.ucar.edu/mailing_lists/archives/thredds/2010/msg00069.html
WMS-clients will request many maps/tiles at once and this will give high
server load if not cached. This might lead to canceled client requests.
In addition, tomcat6 with ncWMS had a strange bug with requests
cancelled by the client, leading to a server-crash at the end.
We solved this by adding a apache 'mod-cache' in front of tomcat and
made tomcat deliver all pages with cache-headers keeping pictures for 7
days in the cache. Server-load dropped nicely due to the cache, and
tomcat doesn't get any more 'client abort exceptions' since those are
swollowed by the cache.
Heiko
On 2013-12-11 23:33, Jay Alder wrote:
Hi, we’ve recently released a web application that uses TDS for mapping,
which is getting a lot of traffic. At one point the server stopped
responding altogether, which is a major problem. A quick restart of
tomcat got it going again, so I’m starting to dig into the logs. We
normally get the GET / request complete behavior, but occasionally we’ll
have:
GET …url…
GET …url…
GET …url…
GET …url…
GET …url…
GET …url…
GET …url…
GET …url…
meanwhile having a 100% CPU spike (with 12 CPUs) for a minute or more
request compete
request compete
request compete
request cancelled by client
request cancelled by client
request compete
request compete
While watching the logs the few times I’ve seen this occur it seems to
pull out of it ok. However the time the server failed, requests were
never returned. From the logs, requests came in for roughly 40 minutes
without being completed. Unfortunately do to the high visibility we
started to get emails from users and the press about the application no
longer working.
Has anyone experienced this before and/or can you give guidance on how
to diagnose or prevent this?
Here are some config settings:
CentOS 5.7
Java 1.6
TDS 4.3.17
only WMS is enabled
Java -Xmx set to 8Gb (currently taking 5.3, the dataset is 600 Gb of
30-arcsecond grids for the continental US, 3.4 Gb per file)
For better or worse we are configured to use 2 instances of TDS to keep
the catalogs and configuration isolated. I’m not sure if this matters,
but I didn’t want to omit it. Since it is a live server I can’t easily
change to the preferred proxy configuration.
I am trying not to panic yet. However, if the server goes unresponsive
again, staying calm may no longer be an option.
Jay Alder
US Geological Survey
Oregon State University
104 COAS Admin Building
Office Burt Hall 166
http://ceoas.oregonstate.edu/profile/alder/
_______________________________________________
thredds mailing list
thredds@xxxxxxxxxxxxxxxx
For list information or to unsubscribe, visit:
http://www.unidata.ucar.edu/mailing_lists/
--
Dr. Heiko Klein Tel. + 47 22 96 32 58
Development Section / IT Department Fax. + 47 22 69 63 55
Norwegian Meteorological Institute http://www.met.no
P.O. Box 43 Blindern 0313 Oslo NORWAY
