Add more tests, fix compliance to semantics #236
Conversation
alertedsnake
requested review from
a team,
toumorokoshi and
hectorhdzg
and removed request for
a team
|
(And now replying properly with my real github account) Confirmed a slight issue - I'm setting the span.status.description field to the error description, but it seems that's expected to be type: detail in the case of an exception, which is a bit harder to do with gRPC, since any abort is raised as a bare Exception even when it isn't technically. I suppose I could make this end up something like StatusCode.NAME: description (with name and description set appropriately). Anyone know where the canonical doc on that format is? This shows up because of this line here: https://github.com/open-telemetry/opentelemetry-python-contrib/blob/master/exporter/opentelemetry-exporter-datadog/src/opentelemetry/exporter/datadog/exporter.py#L252 I did notice #154 and was pondering tackling that as well, so if this is considered an error, I can fix it there instead :) |
|
Ah, I know what to do, hang on. |
|
I should mention that the lint test runs fine locally... |
I think this is just an incorrect assumption that was made because the default exception handling logic in the SDK is to format the exception that way. Really exceptions should be recorded as events, so the datadog_exporter should actually use that to extract it instead. The python code already does this. I'd recommend for now just modifying the datadog_exporter to not make that assumption, and filing a bug against it so it can be fixed in the future. What do you think? |
regarding this, have you updated your version of PIP? There's a new version that now has strict ordering and also results in detecting conflicting dependencies consistency. I see the linting failing because of install conflicts, at the CI probably updates to the latest version of pip every time. I see this in the tox lint output: I think you just need to rebase your commit to fix that. your branch has 16.dev0 as the prometheus writer version. head of the main branch looks to be 0.17.dev0. |
Oh! That's a good clue, thanks for catching that, I'll try it. |
Good call. I've already started working on the Datadog exporter to add metrics (#154), so I'll file an issue and then fix it while I'm in there. |
For open-telemetry#139 - some of these things weren't as clear before.
In the real world, the interceptor is always going to be passed a HandlerCallDetails object with the `method` field set to something valid... previously these tests set this to an empty string, but that's not what happens in a real case. So I've updated the tests to do that right, but also made sure that the span processing allows this to be unset, just in case someone was expecting this behavior to be present.
Now checking the span attributes we've added in the Interceptor. Also fixed some cut & paste errors that resulted in not verifying all spans as intended.
Added a test for handling gRPC aborts, which was previously missing. Updated the `net.*` attributes, so that `net.peer.ip` is always set, and `net.peer.name` is set only if it's `localhost`, since otherwise we'd need to do DNS lookups, and the spec doesn't require that. Added tests to verify the `net.*` shows up properly.
It's really nice to have tests to catch this stuff now.
alertedsnake
force-pushed
the
fix-semantics
branch
from
f312ad0 to
f3abdbc
Compare
Ooh boy, got my work cut out for me :) |
Description
A whole bunch of reworking the tests
I've updated tests to use proper handler names - In the real world, the interceptor is always going to be passed a HandlerCallDetails object with the
methodfield set to something valid... previously these tests set this to an empty string, but that's not what happens in a real case. So I've updated the tests to do that right, but also made sure that the span processing allows this to be unset, just in case someone was expecting this behavior to be present.I've added more checks into the existing tests, now we're checking that span attributes are set as expected.
Fixed a few bugs in the tests where things which were intended to be checked, actually were not.
And finally, I've added a test to catch the gRPC
abort()and verify the error status in the span.Updated status codes to be compliant with specifications
The previous PR I made had the gRPC status codes using the English names for the status, but the spec has been clarified to require the integer value.
Also, the specs say to not use
span.set_status()for a non-error condition, so I've added a check for that in a few places.Added additional attributes
Also as the spec has been clarified, I've added the
rpc.methodandrpc.serviceattributes.Fixes #173
Fixes #139
Type of change
Please delete options that are not relevant.
How Has This Been Tested?
As most of this is either test changes or changes covered by new or existing tests, I've just run a bunch of tests :)
Does This PR Require a Core Repo Change?
Checklist:
See contributing.md for styleguide, changelog guidelines, and more.