Counter and UpDownCounter temporality should be CUMULATIVE

[ocelotl]

The temporalities specified here and here should be CUMULATIVE instead of DELTA.

[jmacd]

This makes sense to me. The reason these should be CUMULATIVE here and now, in the present code base, is that the file opentelemetry/sdk/metrics/export/processor.py converts the input measurements (which are deltas, from these two instrument kinds) into the output aggregations (which are cumulatives) using a call like batch_value.merge(aggregator).

[aabmass]

In the current metrics implementation, it will be DELTA if the stateful is false, so I think the temporality needs to be derived from stateful (related #1007). The code Josh mentioned resets the batch map in when not stateful, so the merging doesn't happen and you get DELTAs:

opentelemetry-python/opentelemetry-sdk/src/opentelemetry/sdk/metrics/export/processor.py

Lines 56 to 62 in c4950a3

	def finished_collection(self):
	"""Performs certain post-export logic.
	For processors that are stateless, resets the batch map.
	"""
	if not self.stateful:
	self._batch_map = {}

opentelemetry-python/opentelemetry-sdk/src/opentelemetry/sdk/metrics/export/processor.py

Lines 81 to 86 in c4950a3

	batch_value = self._batch_map.get(key)
	if batch_value:
	# Update the stored checkpointed value if exists. The call to merge
	# here combines only identical records (same key).
	batch_value.merge(aggregator)

[aabmass]

IIRC, it should be configurable. In Go, it defaults to cumulative but can be configured to delta, kind of like stateful.

[lzchen]

@aabmass @jmacd

We currently have stateful to be configurable. I believe Go has implemented in such a way that the pipeline decides on whether a metric is stateful or not? stateful is also not mentioned anywhere in the sdk spec, so I don't know if we want to depend on something that potentially might not exist?

[ocelotl]

This PR fixes the aggregation temporality for the Counter and UpDownCounter instruments which is an improvement, @aabmass, maybe this can be merged, and the changes related to stateful can be fixed in the PR for #1007?

[aabmass]

@ocelotl SGTM