Breaking changes for 8.0

[graphaelli]

Consolidated list of breaking changes we should consider for the next major.

• replace timestamp.us and span.start with @timestamp (from Use @timestamp instead of timestamp.us and switch to type date_nanos #1678)
• add date_range field for trace events
• Revisit error fields and grouping logic (from Revisit error fields and grouping logic #1769)
• Healthcheck improvements (from meta Improve healthcheck endpoint #2632)
• bug fixes in beater package that are breaking changes (TODO comments in code)
• ensure every route only accepts dedicated METHOD and put logic for it directly into the mux.go file when creating routes.
• start filling in event.* fields (not breaking)
• reject events with an empty transaction type (Server accepts and records empty transaction type #3141)
• use xxHash for grouping key calculation (Investigate switching error grouping key calculation to use something faster apm-data#146)
• consider json-schema changes after separating RUM decoding, (eg: optional timestamp)
• require transaction.name - already happens by convention, should formalize since UI doesn't show no-name transactions
• remove page.url and page.referer (Index page URL as ECS field #3857)
• aligning db fields in ECS is potentially a breaking change for span.db, see ECS#514
• store some metrics as aggregate_metric_double Store various metrics (sum,count,etc) as aggregate_metric_double #4612
• create data stream specific ingest pipelines
• ILM policy to include delete / frozen tier

---

Discarded for 8.0:

- [ ] Remove null values from Json Spec (from #1410)
- [ ] Add labels and deprecate tags on Intake API (from #1835)
- [ ] Move onboarding index to system/dot-prefix or saved object
- [ ] Multiple listener support (from #1405)
- [ ] Synchronous upload of Sourcemaps or consider to move upload endpoint to Kibana (from #2862 )
- [ ] move pipeline configuration into index template / use final_pipeline (Moving to Fleet.)
- [ ] restructure apm-server.yml file to follow along with filebeat input structure as suggested (Jaeger/OTLP will be served on the same port as the Elastic APM agent protocol)

---

• Cross out rather than deleting abandoned ideas
• Check off implemented ideas
• Reopen/reuse existing issues if proceeding with that idea

[jalvz]

I won't create an issue for this, but by 8.0 we should look also into consistency responses. It seems weird that a failed auth returns {"error": "invalid token"}, but wrong method returns a whole

{
  "accepted": 0,
  "errors": [
    {
      "message": "only POST requests are supported"
    }
  ]
}

Particularly because a 405 obviously won't cause any events to be ingested

[graphaelli]

@axw posting your notes here for safe keeping. Regarding options for moving todate_nanos @timestamp:

• @timestamp as date_nanos with a default format of epoch_millis and a runtime field for <foo>.timestamp.us until we can move fully onto @timestamp.
• UI uses the fields API to query @timestamp with format strict_date_optional_time_nanos, and also queries <foo>.timestamp.us and uses that if it exists. We would stop mapping and indexing the latter, this would just be for backwards compatibility

[axw]

For discussion, some things I'd like to look at for 8.0:

• remove index template, ILM policy, pipeline setup & related config -- move all setup to integration package
• remove old index selector support, always create data streams (Introduce standalone data stream support, deprecate Elasticsearch setup #5914)
• remove dedicated Jaeger port (now that we have it muxed on 8200)
• align standalone config defaults with integration package
• switch apm-server.sampling.keep_unsampled default to false
• switch @timestamp to date_nanos (maybe add timestamp.us runtime field for backwards compatibility)
• replace <event>.duration.us with event.duration (maybe add runtime fields for backwards compatibility)

[axw]

Another thing just popped up while I was looking into our dependence on "classic indices": the original source mapping storage implementation. APM Server indexes and searches sourcemaps in apm-<version>-sourcemap.

Users can now upload sourcemaps to Kibana, but currently the only way APM Server can receive them is through Fleet. If we are going to maintain support for sourcemapping in standalone APM Server, we will need to have APM Server query Kibana for sourcemaps when running with data streams enabled.

[axw]

Remove apm-server.instrumentation.* config -- replaced by instrumentation.* many releases ago.

[simitt]

closing this issue as we created tracking issues and labeled them 8.0 for everything we plan to implement.