[KYUUBI #7712] Support IPv6 host:port parsing in Kyuubi service discovery

[maomaodev]

Why are the changes needed?

Fix #7712. Kyuubi currently cannot discover or connect to server/engine endpoints in IPv6-only clusters. Both the server side (DiscoveryClient.parseInstanceHostPort) and the standalone JDBC client side (ZooKeeperHiveClientHelper) parse a host:port string by naively splitting on ":". This breaks for IPv6 because IPv6 addresses themselves contain :.

The correct parsing behavior already exists upstream in Hive (org.apache.hadoop.hive.common.IPStackUtils). This patch mirrors that logic in a new lightweight IPStackUtils utility under kyuubi-util, so it can be shared between the server side and the standalone JDBC client (which only depends on kyuubi-util).

How was this patch tested?

• Added IPStackUtilsTest (11 test cases) aligned with Hive's own IPStackUtilsTest, covering IPv4, IPv6 with brackets, IPv6 without brackets, hostname, invalid port, missing port, missing host and port range validation.
• Extended DiscoveryClientSuite with IPv6 cases (with and without brackets) and converted it from an orphan trait to a runnable class.

Was this patch assisted by generative AI tooling?

Assisted-by: DeepSeek-V4-Pro

[pan3793]

Please prefix the title with [KYUUBI #7712].

A few more places have the same issue and should be fixed in this PR:

• kyuubi-ha/src/main/scala/org/apache/kyuubi/ha/client/zookeeper/ZookeeperDiscoveryClient.scala L319: when kyuubi.ha.zookeeper.publish.configs=true, an IPv6 instance is split on the first colon and publishes a broken host/port. Please use IPStackUtils.getHostAndPort too.
• kyuubi-server/src/main/scala/org/apache/kyuubi/server/ui/JettyServer.scala L59: getServerUri (and the advertisedHost branch in KyuubiRestFrontendService.connectionUrl) emits unbracketed IPv6, which breaks the batch kyuubiInstance / InternalRestClient path in IPv6-only clusters. Please fix these with IPStackUtils as well.
• kyuubi-hive-jdbc/src/main/java/org/apache/kyuubi/jdbc/hive/Utils.java L409 (also L422, L536): authority strings concatenate host without brackets, so the rewritten jdbcUriString is malformed for IPv6. Please use IPStackUtils.formatIPAddressForURL when building them.

[pan3793]

getHostAndPort throws IllegalArgumentException only, catch it instead of Exception.

[pan3793]

Port Hive's concatHostPort and use it here, it also validates host and port.

[maomaodev]

@pan3793 Good catch! Fixed all three, and swept the codebase for the same pattern.

Your three:

• ZookeeperDiscoveryClient.addConfsToPublish: split(":", 2) → IPStackUtils.getHostAndPort.
• JettyServer (start / getServerUri) & KyuubiRestFrontendService.connectionUrl (advertisedHost): → IPStackUtils.concatHostPort.
• Utils.java L409/L422/L536: authority building → IPStackUtils.concatHostPort (which internally calls formatIPAddressForURL + validates host/port).

Also fixed (same bug elsewhere):

• KyuubiTrinoFrontendService.connectionUrl (advertisedHost branch).
• ServiceNodeInfo: the instance string published to service discovery.
• EngineRef: Data Agent JDBC URL.
• KyuubiConnection.getServerHttpUrl: HTTP URL construction.
• KyuubiConfFileParser.addHosts: hosts property fed to beeline JDBC URL builder.

Related:

• Ported Hive's concatHostPort (validates host/port) into IPStackUtils with UT mirroring Hive.
• Narrowed catch (Exception) in DiscoveryClient to catch (IllegalArgumentException) per your earlier point.

Intentionally left alone (all non-functional, happy to revisit if you disagree):

• KyuubiSessionImpl.scala: 5 s"$host:$port" occurrences, all in warn / info / error / logSessionInfo messages — display only, no parsing downstream.
• KyuubiConnection.java: LOG.warn("Failed to connect to " + host + ":" + port) and logZkDiscoveryMessage("Connected to " + host + ":" + port) — same, log-only.
• TBinaryFrontendService.scala / KyuubiTHttpFrontendService.scala: s"Failed to initialize frontend service on $serverAddr:$portNum" — exception message string, not consumed by any code path.
• EmbeddedZookeeper.scala: s"...binding to $host:$clientPort" — exception message, real ZK connect string comes from getConnectString.
• EngineUIProxyServlet.scala L87: X-Forwarded-Context header value is a path prefix, not a URL authority.
• HiveSiteHS2ConnectionFileParser.java L215: Hive-side compatibility layer using getHostName() (returns hostname string, not a numeric address).