recvstats – master node daemon that receives status messages from compute nodes
recvstats [-p port] [-N initial-max-num-nodes] [-f]
The recvstats daemon is part of the
beostat package. It executes
on the master node, receives periodic per-node status data sent by each
cluster node’s sendstats daemon, and makes the data available to
various commands and services on the master node.
The recvstats daemon parses the received data to ensure basic
validity. The exact content and format of the sendstats messages is
version specific, though it typically includes a unique node number
identifying the sender, plus the dynamic values of the following
proc file system files:
The recvstats daemon stores the incoming data in shared memory in
/dev/shm filesystem, which should be readable by everyone. If
that filesystem doesn’t exist or the file permissions are not set
correctly, then recvstats and the consumers of that data will not
function correctly. Note: most consumers of this recvstats data
access it using various commands (e.g.,
Scyld IMF) or use the
abstracted library interface.
The recvstats daemon is started by the ClusterWare service,
and the sendstats daemon is started for BProc nodes by the node
The following options are available to the recvstats daemon.
- -p port
Override the default listen port of 5545. Only use this option if there is a conflict with the default port, and use the same non-default port when executing sendstats on each cluster node.
- -N initial-max-num-nodes
Start recvstats with an explicit non-default guess about how many cluster nodes will be sending data. If a node number above this initial maximum sends data, then recvstats dynamically expands the shared memory structure to accommodate it.
Start the recvstats daemon on the master node:
[root@cluster ~] # /usr/bin/recvstats -p 5545 -N `beoconfig nodes`