Sudden frequent node crashes

Allen Landsidel landsidel.allen at
Thu Oct 8 14:58:08 EDT 2015

Riak 2.1.1
FreeBSD 9.1.
Servers are all virtualized on VMWare ESX 5.5.  Each node is given ~300G 
of storage, 4G of RAM, and 2 core SMP.
Storage is via FC SAN.
Access to the cluster from clients is strictly over the HTTP interface 
and is funneled through haproxy.
The cluster was five nodes using leveldb with a ring size of 16.

The cluster is setup this way so that we could start small and add nodes 
as needed, with a small memory footprint, rather than preallocating a 
ton of memory and disk space that we may never use.  The cluster has 
been running fine for the past month.

Today the cluster has begun experiencing strange repeated failures.  No 
changes in any of the clients have been made, so the problem seems as 
though it must be in the data coming in.

I started a fresh two-node cluster and shortly after data begins flowing 
into the cluster, nodes begin crashing.

The first report in the crash log is below.  I'm far from an expert at 
understanding these errors, but "invalid_unicode" seems telling.

If more from the crash log is needed, I can provide it.  Once this first 
error comes in, the node goes offline and errors just continue to roll 
in to the log file, in what looks like an attempt to automatically 
restart the node.

** Generic server <0.2136.0> terminating
** Last message in was {vnodeid,10000}
** When Server state == {state,"/var/db/riak/kv_vnode/0",0,<0.2135.0>,2}
** Reason for termination ==
2015-10-08 18:36:18 =CRASH REPORT====
     initial call: riak_kv_vnode_status_mgr:init/1
     pid: <0.2136.0>
     registered_name: []
     exception exit: 
     ancestors: [<0.2135.0>,riak_core_vnode_sup,riak_core_sup,<0.169.0>]
     messages: []
     links: [<0.2135.0>]
     dictionary: []
     trap_exit: false
     status: running
     heap_size: 610
     stack_size: 27
     reductions: 569

More information about the riak-users mailing list