This module contains an implementation of the protobuf protocol of FastForward. Protobuf protocol is exposed on port 19091 UDP.
A basic configuration would look like the following.
input:
- type: protobuf
The protocol frames messages as UDP datagrams.
Each message is a frame prefixed with the version and the length of the entire frame.
| version | length | data |
| 4 | 4 | * |
The version field designates which version of the protocol is in use. This determines the structure of the data field.
The length field designates how long the entire frame is supposed to be, since UDP can crop the message, this is used for detecting buffer underruns.
Messages in the data field are serialized according to the protobuf protocol.
This serialization in this module piggy-backs from the implementation of ffwd-client-java.
Large UDP payloads can be silently dropped if the socket buffer is not adequately sized. This usually manifests by having gaps in metrics.
By default net.core.rmem_max is set very low on most operating systems. Setting net.core.rmem_max to 26214400 which is ~25MB allows for bursts of metrics to be processed by FFWD without being dropped.
When ffwd creates the socket for the plugin receiveBufferSize should be set to the same size as the max size allowed by the kernel.
- type: protobuf
protocol:
type: udp
receiveBufferSize: 26214400
For more information about udp buffers check out… https://medium.com/@CameronSparr/increase-os-udp-buffers-to-improve-performance-51d167bb1360
You can monitor for errors on the socket by checking /proc/net/udp6 or /proc/net/udp depending on how you setup the listening socket. The last column is a counter of errors on the socket. Usually any errors on the socket are due to the buffer being full and packets being dropped.
The port, part of the ip:port combo, is hex encoded and can be decoded with python.
cat /proc/net/udp6
1222: 0100630A: 0100630A:4A38 01 00000000:00000000 00:00000000 00000000 0 0 36387 2 ffff931c5727ac00 0
print int("0x4A38", 0)
19000
print int("0x4A93", 0)
19091