Installation instructions:

1. Copy python_modules/infiniband.py to {libdir}/ganglia/python_modules/
2. Copy conf.d/infiniband.pyconf to /etc/ganglia/conf.d
3. The gmond Ganglia daemon will need read/write access to the InfiniBand devices in order to perform these queries. To make this happen, you will need gmond to be running as some user other than nobody (e.g., ganglia). For example, create an InfiniBand group and add ganglia:

* ``groupadd --system infiniband`` * ``usermod --append --groups infiniband ganglia`` * ``chgrp -R infiniband /dev/infiniband/`` * ``chmod 660 /dev/infiniband/umad* /dev/infiniband/issm*`` * To make these changes permanent, you may need custom udev rules. Create a file named ``/etc/udev/rules.d/90-ib.rules`` with the contents: KERNEL=="umad*", NAME="infiniband/%k", MODE="0660", GROUP="infiniband" KERNEL=="issm*", NAME="infiniband/%k", MODE="0660", GROUP="infiniband" KERNEL=="ucm*", NAME="infiniband/%k", MODE="0666", GROUP="infiniband" KERNEL=="uverbs*", NAME="infiniband/%k", MODE="0666", GROUP="infiniband" KERNEL=="ucma", NAME="infiniband/%k", MODE="0666", GROUP="infiniband" KERNEL=="rdma_cm", NAME="infiniband/%k", MODE="0666", GROUP="infiniband" * Ensure your gmond daemon is set to run as the ``ganglia`` username. Set ``user = ganglia`` in the file ``/etc/ganglia/gmond.conf`` * ``service gmond restart`` or ``systemctl restart gmond`` 

By default, all metrics that we could detect for each InfiniBand port are collected. The device type (e.g., mlx4), device number and port number will be appended to the end of each metric. For example:

• ib_port_multicast_xmit_packets_mlx4_0_port1
• ib_port_multicast_xmit_packets_mlx4_0_port2

The following metrics have been implemented:

• ib_excessive_buffer_overrun_errors
• ib_link_downed
• ib_link_error_recovery
• ib_local_link_integrity_errors
• ib_port_rcv_constraint_errors
• ib_port_rcv_data
• ib_port_rcv_errors
• ib_port_rcv_packets
• ib_port_rcv_remote_physical_errors
• ib_port_rcv_switch_relay_errors
• ib_port_unicast_xmit_packets
• ib_port_unicast_rcv_packets
• ib_port_multicast_xmit_packets
• ib_port_multicast_rcv_packets
• ib_port_xmit_constraint_errors
• ib_port_xmit_data
• ib_port_xmit_discards
• ib_port_xmit_packets
• ib_symbol_error
• ib_vl15_dropped
• ib_rate