Closed pyohannes closed 1 year ago
Some minor, late pieces of feedback:
Create
being created separatedly from Publish
: Is Create
that expensive that it may need it's own Span
? Is this operation covering something other than creating a local process object?Settlement
span? Publish
is a batch operation? That is, many messages are produced but on the other end we only consume them if they are delivered individually. No strong opinion yet but was curious about this specific case.@carlosalberto about your questions above:
For the single-parent-child relationship (receiving a single span): this recommendation, are we ok with this recommendation even when Publish is a batch operation? That is, many messages are produced but on the other end we only consume them if they are delivered individually. No strong opinion yet but was curious about this specific case.
We already have a diagram covering that case, and it's a valid one. I started a discussion about it on Slack, but for visibility we intend to add a small clarification to the text so it covers your suggestion/question
Before:
After
Do you think that addresses that point?
@joaopgrassi Thanks, yes, it does. I only would like, in this case, a small note mentioning that this (optional) parenting results in a trace including all the messages published in the batch plus children (which may not be so obvious). So it's not distracting, maybe put it as a foot page note or so?
(Non blocking by the way - I just want to raise the attention on what may not be obvious and the effects of enabling the direct parenting, going forward).
@carlosalberto
a small note mentioning that this (optional) parenting results in a trace including all the messages published in the batch plus children (which may not be so obvious)
hum, if each message has its own create context, then it won't be the same trace. I guess your scenario would only happen if the publish span is used as the creation context for all the messages. In that case yes, then the consumer will all be under the same trace.
OR did you mean that when each message has its own create context, by looking at the trace you might also see other messages, even though you didn't process them?
Not sure how we should communicate this now 🤔
A create context will usually not be a root context, hence all messages published in a batch will usually belong to the same trace (NB: This is different from bach-receive). This could even be added as a requirement
A create context will usually not be a root context, hence all messages published in a batch will usually belong to the same trace
This one @joaopgrassi ;)
Following up on WG call for prototypes, I'm documenting parts of this OTEP implemented on Azure messaging SDKs (including conflicts or extra features):
process
and not delivery in all cases (full processing handler or just a delivery callback), will need to update@lmolkova Is there / will there be public documentation on this Azure support for OpenTelemetry and how to use it with non-AppInsights backends?
@lmolkova Is there / will there be public documentation on this Azure support for OpenTelemetry and how to use it with non-AppInsights backends?
sure, all the docs are here:
We try not to add many attributes until semcov stabilizes at least a bit, but will definitely do more. Any feedback or issues are very welcome.
This OTEP aims at defining consistent conventions about what spans to create for messaging scenarios, and at defining how those spans relate to each other. Instrumentors should be able to rely on a consistent set of conventions, as opposed to deducing conventions from a set of examples.
This was split from OTEP open-telemetry/opentelemetry-specification#192, which became too comprehensive.
Fixes https://github.com/open-telemetry/opentelemetry-specification/issues/1085.