CLI commands that gets redirected to Shelter always yields a 202 "queued" status whereas if the request is handled by the API it would get processed on the spot and return whatever its supposed (eg 404, 200, etc). The API would return useful information to the CLI caller, however with Shelter the user would always see 202 and would not know the true response of the call ,potentially leading to incorrect workflow in the CI or lead to confusions for the user.
For example, this user did not know why their workflow was not working, turns out they forgot to do create-commit. In the logs all they saw was
Process Upload Completion complete when in reality the upload-complete command should have returned "Commit SHA not found" which the user would have been able to self diagnose fix instead of raising a support ticket.
Notice the initial responses to upload-complete command was 202, then a few log lines later those were processed and showed an error that is now lost.
https://cloudlogging.app.goo.gl/UG5kfb5uVax3MKyZ8
Related thought:
Could the use of Shelter for CLI commands also cause race conditions with a user's typical "create-commit -> ... -> upload" workflow?
"create-commit" -> queued/202
"upload" (some variant of upload command) -> queued/202
CI shows all good
upload message gets consumed first -> errors out with commit not found
create commit message gets consumed -> a commit created
the system ends in an expected state
Not sure if this can happen with the current setup (topics, subscribers, FIFO policy, etc)
CLI commands that gets redirected to Shelter always yields a 202 "queued" status whereas if the request is handled by the API it would get processed on the spot and return whatever its supposed (eg 404, 200, etc). The API would return useful information to the CLI caller, however with Shelter the user would always see 202 and would not know the true response of the call ,potentially leading to incorrect workflow in the CI or lead to confusions for the user.
For example, this user did not know why their workflow was not working, turns out they forgot to do create-commit. In the logs all they saw was Process Upload Completion complete when in reality the upload-complete command should have returned "Commit SHA not found" which the user would have been able to self diagnose fix instead of raising a support ticket.
https://sentry.zendesk.com/agent/tickets/129874
Notice the initial responses to upload-complete command was 202, then a few log lines later those were processed and showed an error that is now lost. https://cloudlogging.app.goo.gl/UG5kfb5uVax3MKyZ8
Related thought: Could the use of Shelter for CLI commands also cause race conditions with a user's typical "create-commit -> ... -> upload" workflow?
Not sure if this can happen with the current setup (topics, subscribers, FIFO policy, etc)