[receiver/hostmetrics/processscraper]: memory metric names do not adhere to semantic conventions

[andrzej-stencel]

What happened?

Description

The memory metrics reported by the hostmetrics receiver's process scraper do not adhere to the semantic conventions:

• process.memory.physical_usage should be process.memory.usage according to the spec
• process.memory.virtual_usage should be process.memory.virtual according to the spec

Steps to Reproduce

Run the collector with hostmetrics receiver enabled and the process scraper configured (see example config below).

Expected Result

The following metrics should be reported for memory usage:

• process.memory.usage
• process.memory.virtual

Actual Result

The following metric names are reported for memory usage:

• process.memory.physical_usage
• process.memory.virtual_usage

Collector version

v0.60.0

Environment information

OS: Any (tested on Ubuntu 20.04)

OpenTelemetry Collector configuration

exporters:
  logging:
    loglevel: debug

receivers:
  hostmetrics:
    collection_interval: 3s
    scrapers:
      process:

service:
  pipelines:
    metrics:
      receivers:
      - hostmetrics
      exporters:
      - logging

Log output

Metric #2
Descriptor:
     -> Name: process.memory.physical_usage
     -> Description: The amount of physical memory in use.
     -> Unit: By
     -> DataType: Sum
     -> IsMonotonic: false
     -> AggregationTemporality: AGGREGATION_TEMPORALITY_CUMULATIVE
NumberDataPoints #0
StartTimestamp: 2022-09-20 10:07:12 +0000 UTC
Timestamp: 2022-09-20 10:07:13.425679646 +0000 UTC
Value: 528384
Metric #3
Descriptor:
     -> Name: process.memory.virtual_usage
     -> Description: Virtual memory size.
     -> Unit: By
     -> DataType: Sum
     -> IsMonotonic: false
     -> AggregationTemporality: AGGREGATION_TEMPORALITY_CUMULATIVE
NumberDataPoints #0
StartTimestamp: 2022-09-20 10:07:12 +0000 UTC
Timestamp: 2022-09-20 10:07:13.425679646 +0000 UTC
Value: 10870784

Additional context

No response

[github-actions]

Pinging code owners: @dmitryax. See Adding Labels via Comments if you do not have permissions to add labels yourself.

[kentquirk]

In the meeting today, people were talking about putting the switch behind a feature flag.

What's wrong with sending the metric under both a new name and an old one for a while, and putting the old one behind a feature flag that initially defaults to ON?