Diffusion™ metrics provide
information about the server, client sessions, topics and log events. Diffusion can provide metrics in three main ways:
via the web console, via JMX-compatible MBeans and via Prometheus.
Methods of accessing metrics
There are multiple ways to access the metrics. As of Diffusion 6.3, the same information is available
through each access method.
Note: In previous versions of Diffusion, metrics were sometimes referred to as "statistics".
- Web console metrics
- The metrics are available through the Diffusion web console. This is the most
convenient way to access metrics for development and testing purposes, but does not support aggregating metrics across multiple servers or
recording and retrieving historical data.
JMX or Prometheus access are more suitable for production systems.
- MBeans for JMX
- Diffusion registers MBeans with the Java Management Extensions (JMX) service. This enables monitoring of the metrics using the JMX tools that are available from
a range of vendors.
- Diffusion provides endpoints for the
Prometheus monitoring system.
To use Prometheus, your Diffusion server needs to have a Commercial with Scale & Availability license,
or an evaluation license such as the Community Evaluation license. See License types for more information.
The metrics can be accessed in the following recommended ways:
- As MBeans, using a JMX tool, such as VisualVM or JConsole. See the table below for
MBean interfaces. For
more information, see Using Java VisualVM or Using JConsole.
- Using the Diffusion management console. For more
information, see Diffusion management console.
- As Prometheus endpoints at http://localhost:8080/metrics, provided
you have a suitable license.
If not accessing from the same machine as the Diffusion server,
replace localhost with the IP address or hostname.
Collecting custom metrics using metric collectors
A metric collector is a way to collect metrics for a particular set of topics or sessions, configured by you.
You can use the Diffusion web console or JMX to define
metric collectors. See Configuring metrics for details.
Collected metrics are published to the console, JMX and optionally via Prometheus.
Counters and gauges
Metrics are divided into counters and gauges.
- Counter metric
- A counter is a cumulative metric, which reports a value since
the server was started. A counter metric will always go up over a server's lifetime. For example,
the total number of bytes received by the server is a counter.
- Gauge metric
- A gauge is a metric which reports the current value of a metric. A gauge value can go up or down.
For example, the number of connected sessions is a gauge.
This section describes the built-in metrics that are always available, aside from any metric collectors you may have created.
Metrics are not persisted between server restarts. Restarting the server will set all counter metrics back to zero.
The following is a list of all the top level statistics and their attributes.
Table 1. Metrics provided by Diffusion
||Number of log events for a given ID code and severity level (levels are error, warn,
info, debug, trace).
||Data received from the network in bytes.
||Data sent to the network in bytes.
||Number of connected sessions.
||Session data received from the network in bytes.
||Session data received from the network in messages.
||Number of open sessions.
||Session data sent to the network in bytes.
||Session data sent to the network in messages.
||Peak number of sessions.
||Total sessions opened.
||Current number of topics.
||Total number of topics.
||The value data stored by the topics, in bytes.
||Number of direct subscriptions to the topics.
||Number of sessions subscribed to one or more topics.
||Number of updates sent to subscribers.
||Data sent to subscribers, before message compression, in bytes.
||Data sent to subscribers, after message compression, in bytes.
||Number of updates to a topic that provide a full value.
||Number of updates to a topic that provide a partial value.
||On each change of topic value, this metric increases by the size of the new value.
||On each change of topic value, this metric increases by the size of an internal delta representing the difference the previous and new values.
Delta compression ratio
value_bytes and delta_bytes can be used to capture the theoretical delta compression ratio of the application data flowing through the topics.
Both the console and the JMX MBean perform this calculation. The ratio is a value between 0 and 1. The closer the ratio is to 1,
the more benefit the application data will obtain from delta streaming. If value_bytes is 0, there have been no updates,
so the delta compression ratio is reported as zero. Otherwise it is calculated as:
1 - delta_bytes / value_bytes
Delta streaming is enabled for subscriptions by default, but can be disabled on a per-topic basis using the PUBLISH_VALUES_ONLY topic property.
If delta streaming is enabled, a stable set of subscribers remain connected, and no session has a significant backlog (so conflation is not applied),
the following relationship should hold:
subscriber_update_bytes ≅ delta_updates x subscribers
Delta streaming can also be used to update topic values. If the delta compression ratio is high, but delta_updates is zero
(or low, relative to value_updates), consider whether your application can use the stateful update stream API to take advantage of delta streaming.
Log metrics record information about server log events. Separate metrics are kept for each unique pair of log code and log severity level that has been logged.
The log severity levels are: error, warn, info, debug, trace.
A JMX MBean is created for each pair of log code and log severity that has been logged at least once.
Here is an example MBean name: com.pushtechnology.diffusion:type=LogMetrics,server="server_name",level=warncode=PUSH-12345
Session metrics versus network metrics
The network inbound_bytes and outbound_bytes metrics include bytes that are not counted by the equivalent session metrics.
The session metrics include bytes from transport framing and all session traffic (including additional HTTP traffic from long polling).
The network metrics include all bytes included in the session metrics as well as non-session bytes such as:
- TLS overhead
- Web server traffic (for example, browsers downloading the web console pages)
- Rejected connection attempts
Metrics in the Publisher API
Publisher metrics, client instance metrics, and topic instance metrics have all been removed.
Consequently the PublisherStatistics, ClientStatistics and TopicStatistics interfaces provide
no information. These interfaces are deprecated and will be removed in a future release.
Limited server metrics are still available through the Publisher API using the ServerStatistics interface.
more information, see the Java API