[Bug] Hive query on Paimon table fails with 'Proxy user is not supported' when Kerberos proxy user is used

[sohurdc]

Search before asking

• I searched in the issues and found nothing similar.

Paimon version

1.3.1

Compute Engine

paimon-hive-connector-3.1, flink-1.17.1

Minimal reproduce step

---

labels: bug, hive, kerberos

Describe the bug

When a Paimon table is written through Flink and then queried from Hive via HiveServer2, the query fails with:

SELECT * FROM paimon_sink LIMIT 1;
Error: Error while compiling statement: FAILED: UnsupportedOperationException Proxy user is not supported (state=42000,code=40000)

Environment

• Paimon version: 1.3.1
• Hive connector: paimon-hive-connector-3.1
• Hive access mode: HiveServer2 (JDBC / Beeline) with proxy user (hive.server2.proxy.user=bdwh)
• Kerberos: enabled

Root cause analysis

The error is thrown from org.apache.paimon.security.KerberosLoginProvider:

public void doLogin() throws IOException {
    if (principal != null) {
        LOG.info(
                "Attempting to login to KDC using principal: {} keytab: {}", principal, keytab);
        UserGroupInformation.loginUserFromKeytab(principal, keytab);
        LOG.info("Successfully logged into KDC");
    } else if (!isProxyUser(UserGroupInformation.getCurrentUser())) {
        LOG.info("Attempting to load user's ticket cache");
        UserGroupInformation.loginUserFromSubject(null);
        LOG.info("Loaded user's ticket cache successfully");
    } else {
        throwProxyUserNotSupported();
    }
}

The Kerberos login logic currently has three branches:

1. principal is non-null -> UserGroupInformation.loginUserFromKeytab(principal, keytab) (the Spark / Flink way).
2. principal is null and the current user is not a proxy user -> UserGroupInformation.loginUserFromSubject(null) (load the ticket cache).
3. principal is null and the current user is a proxy user -> throws Proxy user is not supported.

The problem is that Hive cannot log into Kerberos the same way Spark does — Hive does not support passing --keytab directly. Instead, an already-Kerberos-authenticated HiveServer2 acts as a proxy on behalf of the business user, with the caller specifying hive.server2.proxy.user=bdwh (e.g. via Beeline).

In this scenario, principal is null and the current UGI is a proxy user, so Paimon falls into branch 3 and throws Proxy user is not supported, which makes the whole query fail before it can even read the Paimon table.

Proposed fix

For the proxy-user case, Paimon should do nothing (the Kerberos credentials are already provided by the HiveServer2 delegation / proxy mechanism) instead of throwing.

public void doLogin() throws IOException {
    if (principal != null) {
        LOG.info(
                "Attempting to login to KDC using principal: {} keytab: {}", principal, keytab);
        UserGroupInformation.loginUserFromKeytab(principal, keytab);
        LOG.info("Successfully logged into KDC");
    } else if (!isProxyUser(UserGroupInformation.getCurrentUser())) {
        LOG.info("Attempting to load user's ticket cache");
        UserGroupInformation.loginUserFromSubject(null);
        LOG.info("Loaded user's ticket cache successfully");
    } else {
        // Proxy user: do nothing, credentials are provided by HiveServer2 proxy.
        // throwProxyUserNotSupported();
    }
}

Note: an earlier attempt replaced the throw with UserGroupInformation.loginUserFromSubject(null), but that can cause TGT issues; the final fix is to simply do nothing for proxy users.

Verification

After applying the fix and rebuilding the Hive connector:

mvn -DskipTests -Dspotless.check.skip=true clean package -pl paimon-hive -am
# produces: paimon-hive-connector-3.1-1.3.1.jar

The query succeeds:

0: jdbc:hive2://dmc014011.venus.sohurdc.com:1> SELECT count(*) FROM paimon_sink LIMIT 1;
+---------+
| _c0     |
+---------+
| 154046  |
+---------+
1 row selected (75.586 seconds)

Ranger authorization also works as expected (the proxy user still needs the required privileges):

SELECT * FROM paimon_sink LIMIT 1;
Error: Error while compiling statement: FAILED: HiveAccessControlException Permission denied: user [bdwh] does not have [SELECT] privilege on [bdwh/paimon_sink/*] (state=42000,code=40000)

Expected behavior

When Kerberos proxy user is used (as in HiveServer2), Paimon should not throw Proxy user is not supported and should allow queries to proceed.

Additional context

Related code reference: paimon-common/src/main/java/org/apache/paimon/security/KerberosLoginProvider.java

What doesn't meet your expectations?

FAILED: UnsupportedOperationException Proxy user is not supported

Anything else?

No response

Are you willing to submit a PR?

• I'm willing to submit a PR!