[mapserver-users] MapCache errors when writing new tiles in a cluster environment

[John Taranu]

We are having some problems with MapCache running in a multi-server cluster
environment.  

 

Occasionally, requests fail for brand new, uncached tiles from a vertical
assembly layer (i.e. from other WMS layers as sources).   I can confirm this
from the server logs:

 

Jan  9 18:55:33 server2 httpd[28518]: [error] [client 127.0.0.1] failed to
create file /home/caches/NewLayer/gmaps/17/000/036/125/000/082/973.xxx: No
such file or directory

Jan  9 18:55:33 server2 httpd[28518]: [error] [client 127.0.0.1] failed to
create file /home/caches/NewLayer/gmaps/17/000/036/125/000/082/973.xxx: No
such file or directory

 

Only one tile request is being made, but the failure is logged twice for
some reason.  It is returned as a blank image with an HTTP 500 error code to
the client. 

 

This error does not appear to be occurring at the tile locking phase, but
rather at the point of writing the tile back into the cache directory.

 

Our setup uses multiple web servers running mapcache under a load balancer.
The disk cache directory (and the lock file directory) is shared as an NFS
mount.  All servers are running the latest version of MapCache from Github
as an Apache module on CentOS 5.6.  

 

This error still occurs when only using one webserver, but it occurs much
less frequently.  It occurs more frequently when the servers are under a
higher load.

 

Has anyone experienced similar problems?

 

---------------------------------------

John Taranu

Senior Web Developer

First Base Solutions Inc

Markham, Ontario

email: jtaranu at firstbasesolutions.com

web:  <http://www.firstbasesolutions.com/> firstbasesolutions.com

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/mapserver-users/attachments/20130111/4613f7c5/attachment.html>