bitflicker64 commented on PR #3132:
URL: https://github.com/apache/hugegraph/pull/3132#issuecomment-5410435969

   **Checkpoint (2026-08-25)**
   
   The PD raft IP whitelist is now disabled in-cluster through the upstream 
`raft.ip-whitelist.enabled` switch, and the interim `wait-for-pd-dns` init 
container is gone. The gate only ever sequenced around the first-boot half of 
the problem; the switch removes the rejection mechanism itself, so both the 
install race and the pod-IP-recycle rejection are addressed at the cause.
   
   Re-tested end to end on Kind against a PD image carrying the switch (details 
in the description):
   
   - **Zero** `Blocked connection` and zero `Could not resolve allowlist entry` 
lines across the entire campaign, every namespace, every PD.
   - **Pod-IP recycle now passes.** A deleted PD returned on a changed IP with 
a stable hostname and peers accepted it without a single blocking line, against 
39 and 63 blocking lines and a permanent Raft zombie previously.
   - Two deterministic installs plus a zero-override default install, all clean 
on the first attempt.
   - Fault battery — leader partition, PD follower/leader crash, PD and Store 
majority loss, PVC reattach, rolling restart under write load, auth on every 
Server replica — with acknowledged writes durable throughout.
   - Upgrade from the previous revision with data in place rolled PD and Server 
only, kept Store and Hubble untouched, and produced no block window; the manual 
remediation the gate-era upgrade needed is obsolete.
   - Secret rotation rolls Server via the `checksum/auth` annotation, and the 
documented first-run path works as written.
   
   Also in this push: the chart CI's `JAVA_OPTS` assertions matched a 
quote-terminated shard-count prefix and broke once the whitelist flag was 
appended, and the README still described the removed init container in its 
Limitations and Upgrading sections. Both are fixed, and the new value is 
documented in the values table.
   
   Still open and unchanged: the `/v1/health` quorum-signal review thread. 
Gating the Store wait on `/v1/members` needs PD REST authentication, which 
current builds reject for every credential, so that choice still depends on the 
PD-side direction.


-- 
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