Repository navigation
Fix CLI connection display reporting wrong is_extra_encrypted flag - #67561
Conversation
|
Congratulations on your first Pull Request and welcome to the Apache Airflow community! If you have any issues or are unsure about any anything please check our Contributors' Guide
|
71c4113 to
040fe0a
Compare
`airflow connections get` showed the password's encryption status under both `is_encrypted` and `is_extra_encrypted` keys, because the display mapper sourced both from `Connection.is_encrypted`. These flags are independent — the extra field (which usually holds API keys, certs) can be encrypted while the password is not, or vice versa. Operators checking the readout could be misled about whether `extra` is encrypted at rest. Adds a regression test that exercises both `ConnectionDisplayMapper` methods (full_details, masked_sensitive) across the divergent input combinations.
040fe0a to
ffdc3a5
Compare
Pedrinhonitz
left a comment
There was a problem hiding this comment.
Thanks for the adjustment, it looks good to me.
|
@eladkal - can you take a peek at this one? |
potiuk
left a comment
There was a problem hiding this comment.
LGTM — clear bug: is_extra_encrypted was sourced from conn.is_encrypted, so the two independent flags always reported the same value. Fix is correct and the parametrized test covers both mappers × both flag combinations. (Minor: since this is a user-visible CLI output fix, a bugfix newsfragment would've been fine to keep — but not worth blocking.) Thanks!
Drafted-by: Claude Code (Opus 4.8); reviewed by @potiuk before posting
|
@potiuk - why was |
My mistake - I thought I already merged that one. |
|
Awesome work, congrats on your first merged pull request! You are invited to check our Issue Tracker for additional contributions. |
…pache#67561) `airflow connections get` showed the password's encryption status under both `is_encrypted` and `is_extra_encrypted` keys, because the display mapper sourced both from `Connection.is_encrypted`. These flags are independent — the extra field (which usually holds API keys, certs) can be encrypted while the password is not, or vice versa. Operators checking the readout could be misled about whether `extra` is encrypted at rest. Adds a regression test that exercises both `ConnectionDisplayMapper` methods (full_details, masked_sensitive) across the divergent input combinations.
airflow connections getshowed the password's encryption status under bothis_encryptedandis_extra_encryptedkeys, because the display mapper sourced both fromConnection.is_encrypted. These flags are independent — the extra field (which usually holds API keys, certs, etc.) can be encrypted while the password is not, or vice versa. Operators checking the readout could be misled about whetherextrais encrypted at rest.The fix changes
is_extra_encryptedto source fromConnection.is_extra_encryptedin bothConnectionDisplayMapper.full_detailsandConnectionDisplayMapper.masked_sensitive. A parametrized regression test exercises both methods across the two divergent flag combinations to prevent re-introducing the bug.Was generative AI tooling used to co-author this PR?
Generated-by: Claude Code (Opus 4.7) following the guidelines