bitflicker64 commented on code in PR #3449:
URL: https://github.com/apache/hugegraph/pull/3449#discussion_r4458407777


##########
helm/hugegraph/templates/hubble-deployment.yaml:
##########
@@ -286,6 +296,8 @@ spec:
               value: {{ if $vermeerEnabled }}{{ printf "http://%s:%v"; (include 
"hugegraph.vermeer.masterName" .) .Values.vermeer.master.httpPort | quote }}{{ 
else }}""{{ end }}
             - name: HG_HUBBLE_VERMEER_TOKEN_FILE
               value: {{ if $vermeerEnabled 
}}"/var/run/hugegraph/vermeer/token"{{ else }}""{{ end }}
+            - name: HG_HUBBLE_TEXT2GQL_MASTER_KEY_FILE
+              value: "/hubble-data/text2gql-master.key"

Review Comment:
   **`fsGroup` can make the relocated key unreadable after Pod replacement**
   
   Moving the key onto the `/hubble-data` volume puts it under kubelet's 
volume-ownership handling. `values.yaml` tells non-root deployments with 
persistence to set `hubble.podSecurityContext.fsGroup`. With the default 
`fsGroupChangePolicy: Always`, kubelet recursively adds group read/write 
(`0660`) to files on supported volumes each time the volume is mounted.
   
   Hubble creates the key as `0600`, and 
`Text2GqlSecretStore.secureAttributes()` rejects any group bit (`allowed = 
{OWNER_READ, OWNER_WRITE}`). So the first save works. After the next Pod 
replacement the key becomes `rw-rw----`, and both decrypt and encrypt return 
503 `secret-unavailable`, even though the PVC and the UID were kept. Before 
this PR the key lived in the container filesystem, so `fsGroup` never touched 
it.
   
   Suggested fix: restore owner-only mode in the startup wrapper before Hubble 
starts, for example `[[ -f "${HG_HUBBLE_TEXT2GQL_MASTER_KEY_FILE}" ]] && chmod 
600 "${HG_HUBBLE_TEXT2GQL_MASTER_KEY_FILE}"`. Alternatively, document 
`fsGroupChangePolicy: OnRootMismatch`. Please add a test or a manual check for 
the `fsGroup` case.
   
   Note: this comes from reading the source and the documented kubelet 
behaviour. It was not run on a real cluster, and CSI drivers that manage 
permissions themselves may behave differently.



##########
helm/hugegraph/templates/hubble-deployment.yaml:
##########
@@ -267,6 +272,11 @@ spec:
               if [[ "${FOUND_SERVER_HOST}" == false ]]; then
                 printf 'server.host=0.0.0.0\n' >>"${TMP}"
               fi
+              # The image default (data/text2gql-master.key, relative to 
/hubble)
+              # is outside the data volume; keep the key next to the H2 
database.

Review Comment:
   **Document the one-time API key re-entry on upgrade**
   
   For an existing release, the ciphertext in H2 was encrypted with a key in 
the old container filesystem, at `/hubble/data/text2gql-master.key`. That key 
cannot be migrated and is gone once the upgrade replaces the Pod. Hubble does 
not create a key at startup or on decrypt (`masterKey(false)` throws 
`unavailable()`). So the UI keeps showing a saved key, but generation fails 
with 503. Saving other Text2GQL settings without re-entering the key also 
fails, because the save path decrypts the stored value.
   
   Losing the key on Pod replacement was already the behaviour before this PR. 
However, this PR changes deployment behaviour, and the repository's `AGENTS.md` 
asks for matching docs in the same PR. The PR description also says nothing 
operational changes. Please add a short note to the Hubble section or 
`docs/operations.md`: after upgrading, re-enter the Text2GQL API key once, and 
from then on it survives Pod replacement while persistence is enabled. 
Optionally, also mention that the key does not survive Pod replacement with 
`persistence.enabled=false`.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to