An adapter between slipstream telemetry and honeycomb events
Honeycomb is an observability tool with a sweet query interface and distributed tracing support.
At NFIBrokerage, we ❤️ honeycomb because it enables us to debug issues in fractions of time (and with far more certainty) than log ingestion services, and lets us understand our system in real-time.
For example, it was gloriously fast to debug this issue: NFIBrokerage/slipstream#12 because of the telemetry support added in 0.3 and this exporter library.
SlipstreamHoneycomb currently captures the only telemetry which Slipstream emits: connection messages.
These connection messages are fairly low-level and are very quick (usually a matter of microseconds). Below is an example screenshot of a trace in our Honeycomb:
The long purple bar is the connection, which was open for ~150s in this case.
That came from the [:slipstream, :connection, :connect, :stop]
event. The
other spans of this connection are [:slipstream, :connection, :handle, :stop]
. From top to bottom, we see from the example:
- the initial
handle_continue(:connect, ..
- the gun connection coming up (
{:gun_up, conn, :http}
) - the gun upgrade which is decoded into the
%Slipstream.Events.ChannelConnected{}
event - a JoinTopic command from the client
- a
:gun_ws
message that decoded into a TopicJoinSucceeded event - a big push of data from the remote server (decoded to MessageReceived event)
and then much later (~30s), we see:
- a SendHeartbeat command, followed by
- the HeartbeatAcknowledged
(x4). And finally, the parent process exits with a ParentProcessExited event.
def deps do
[
{:slipstream_honeycomb, "~> 0.1"}
]
end
Check out the docs here: https://hexdocs.pm/slipstream_honeycomb