InfluxData line-protocol flavor

Overview

ClusterCockpit uses the InfluxData line-protocol for transferring messages between its components. The line-protocol is a text-based representation of a metric/event with a value, time and describing tags. All metrics/events have the following format (if written to stdout):

<measurement>,<tag set> <field set> <timestamp>

Where <tag set> and <field set> are comma-separated lists of key=value entries. In a mind-model, think about tags as indices in the database for faster lookup and the <field set> as values.

We are using the tag set to add metadata information and the field for the payload.

Remark: In the first iteration, we only sent metrics (number values) but we extended the specification to messages with different purposes. The below text was changed accordingly. The update is backward-compatible, for metrics (number values), nothing has changed.

Message categories

There exist the following line line-protocol message flavors:

Metric: The field key is value, measurement = metric name
Event: The field key is event, Events are actionable informations, measurement = event subtype (job, phases, ?? ), Additional tag function=
Log message: The field key is log. Log messages are purely informational, measurement = [ccb, ccms, ccmc, ccem, ccnc], Additional tag loglevel
Control message: The field key is control, measurement = knob name (rapl, freq, prefetcher, topology, config), Additional tags: method=[get,put]

Messaging

ClusterCockpit uses the NATS messaging network, with the option to support other messaging frameworks in the future. To distinguish between different message types and easily filter the types an application is interested in the following subject hierarchy tree is used:

<cluster name>. |
                --- metrics
                |
                --- events.[job]
                |
                --- log.[ccb, ccms, ccmc, ccem, ccnc]
                |
                --- control.[get,put]

Points generic for all message categories

In ClusterCockpit we limit the flexibility of the InfluxData line-protocol slightly. The idea is to keep the format usable by different components.

Each message is identifiable by the measurement (= metric name), the hostname, the type and, if required, a type-id.

Mandatory tags per message

hostname
type
- node
- socket
- die
- memoryDomain
- llc
- core
- hwthread
- accelerator
type-id for further specifying the type like CPU socket or HW Thread identifier

Although no type-id is required if type=node, it is recommended to send type=node,type-id=0.

Optional tags depending on the message

In some cases, optional tags are required like filesystem, device or version. While you are free to do that, the ClusterCockpit components in the stack above will recognize stype (= "sub type") and stype-id. So filesystem=/homes should be better specified as stype=filesystem,stype-id=/homes.

Message types

There exist different message types in the ClusterCockpit ecosystem, all specified using the InfluxData line-protocol.

Metrics

Identification: value=X field with X being a number

While the measurements (metric names) can be chosen freely, there is a basic set of measurements which should be present as long as you navigate in the ClusterCockpit ecosystem

flops_sp: Single-precision floating point rate in Flops/s
flops_dp: Double-precision floating point rate in Flops/s
flops_any: Combined floating point rate in Flops/s (often (flops_dp * 2) + flops_sp)
cpu_load: The 1m load of the system (see /proc/loadavg)
mem_used: The amount of memory used by applications (see /proc/meminfo)
ipc: instructions-per-cycle metric
mem_bw: Main memory bandwidth (read and write) in MByte/s
cpu_power: Power consumption of the whole CPU package
mem_power: Power consumption of the memory subsystem
clock: CPU clock in MHz
...

For the whole list, see job-data schema

Events

Identification: event="X" field with "X" being a string The name (measurement) of the event message can further specialize the purpose (similar to REST endpoints), e.g. start_job, and stop_job for events of type job.

Example start job event: TBD

Controls

Identification:

control="X" field with "X" being a string
method tag is either GET or PUT

Logs

Identification: log="X" field with "X" being a string