[
https://issues.apache.org/jira/browse/HDDS-15780?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18139796#comment-18139796
]
Siyao Meng commented on HDDS-15780:
-----------------------------------
A few more items for the IPv6 docs, found while checking master for gaps:
* *Rack awareness:* Hadoop's {{TableMapping}} only matches IPv6 entries in
Java's full form ({{2001:db8:0:0:0:0:0:10}}), not the compressed form
({{2001:db8::10}}). {{ScriptBasedMapping}} passes the full form to the script.
Until HADOOP-20072 is fixed, write topology tables in the full form or use host
names.
* *Dual-stack secure clusters:* the RPC client binds to the first address of
the principal host, which is IPv4 by default. With IPv6 service addresses, set
{{ipc.client.bind.wildcard.addr=true}} (or see HDDS-17728). With service
authorization on, IPv6 addresses also need PTR records that map to the
principal host.
* *Client Hadoop versions:* the Hadoop 3.4.4 bump (HDDS-16137) only covers
Ozone's own processes and CLI. Applications that use the ozonefs jars (Spark,
Hive, MapReduce) and the YARN ResourceManager when it renews delegation tokens
run their own Hadoop, and need the same fixes (HADOOP-19958, HADOOP-19961 and
the others) to use IPv6 service addresses. The {{ozonefs-hadoop2}} jar will not
get them. Document the minimum client Hadoop version, or recommend DNS names
for service addresses.
* *Datanode address family:* a datanode has one IP address, recorded by SCM
from the connection the datanode registers on. On a dual-stack cluster, clients
of the other family cannot reach it unless
{{hdds.datanode.use.datanode.hostname=true}} and the host name has both A and
AAAA records.
> Document IPv6 (dual-stack and IPv6-only) deployment for Ozone
> -------------------------------------------------------------
>
> Key: HDDS-15780
> URL: https://issues.apache.org/jira/browse/HDDS-15780
> Project: Apache Ozone
> Issue Type: Task
> Reporter: Siyao Meng
> Priority: Major
>
> Publish operator guidance for the supported IPv4, dual-stack, and IPv6-only
> profiles after the functional and qualification tasks are complete.
> The documentation must cover:
> * Existing IPv4 defaults and explicit IPv6 listener configuration.
> * Host-only, host-and-port, URI authority, and HTTP Host-header formats.
> * The distinction between bind and advertised addresses.
> * Wildcard listener use and rejection of wildcard, link-local, or scoped
> advertised addresses.
> * Operating-system IPV6_V6ONLY behavior and why binding to :: does not always
> provide IPv4 reachability.
> * Stable DNS, forward and reverse DNS, Kerberos principal selection, and TLS
> DNS or IP subject alternative names.
> * The ofs bracketed-literal form and the o3fs requirement to use an OM
> service ID or DNS name.
> * S3 path-style access through a literal and virtual-host-style access
> through DNS with an AAAA record.
> * The supported non-rolling software upgrade followed by IPv4 to dual-stack
> to IPv6-only migration.
> * SCM and OM Ratis migration with stable DNS or stored IP literals, including
> rollback requirements.
> * Datanode UUID retention, pipeline replacement, bounded migration batches,
> and hdds.datanode.use.datanode.hostname.
> * The minimum supported Hadoop and Ratis versions and any remaining
> limitations.
> Depends on the functional, migration, dependency-release, and qualification
> tasks under HDDS-15763.
> Parent epic: HDDS-15763.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
