MulticastKeepaliveHeartbeatReceiver uses invalid and hardcoded size for MTU resulting in failure to replicate caches

Type: Bug
Status: New
Priority: 2 Major
Resolution:
Component/s: ehcache-core
Labels:

Assignee: drb
Reporter: shartrec
Created: June 26, 2014
Votes: 1
Watchers: 2
Updated:
June 29, 2014
Resolved:

Description

We have a set up where we replicate > 100 caches and in some installs are seeing {noformat} 2014-06-26 11:42:15,737 Multicast Heartbeat Receiver Thread ERROR [sf.ehcache.distribution.PayloadUtil] Could not ungzip. Heartbeat will not be working. Not in GZIP format 2014-06-26 11:42:15,738 Multicast Heartbeat Receiver Thread DEBUG [sf.ehcache.distribution.MulticastKeepaliveHeartbeatReceiver] rmiUrls received {noformat} Digging into the code for PayloadUtil it appears the caclulartion is * Wrong - it is set at 1500, but is uses as total payload size, whereas the MTU size includes the payload and any headers that may be in the packet. * Hard coded. In various networks, such as virtualized or over a VPN the MTU size may in fact be considerably smaller than the ethernet default of 1500.

Comments

Trevor Campbell 2014-06-26

This is related to EHC-424

Chris Fuller 2014-06-29

Another point on the hardcoding… it works to the other extreme too. If you have fiber between your nodes, then the MTU is over 4300. Forcing fragmentation at 1500 doesn’t really have any merit.