EINTERNAL

EINTERNAL

Error message

hdfsGetHosts(path=%s, start=%"PRId64", length=%"PRId64"): BlockLocation#getHosts returned NULL

What it means

hdfsGetHosts walks the file's BlockLocation[] and calls BlockLocation#getHosts over JNI for each block. Java normally returns an array (possibly empty), so a null return signals a broken implementation or bad JVM state; libhdfs reports EINTERNAL with the path and the start/length range in the message.

Source

Thrown at hadoop-hdfs-project/hadoop-hdfs-native-client/src/main/native/libhdfs/hdfs.c:3786

        if (jthr || !jFileBlock) {
            ret = printExceptionAndFree(env, jthr, PRINT_EXC_ALL,
                "hdfsGetHosts(path=%s, start=%"PRId64", length=%"PRId64"):"
                "GetObjectArrayElement(%d)", path, start, length, i);
            goto done;
        }
        
        jthr = invokeMethod(env, &jVal, INSTANCE, jFileBlock,
                JC_BLOCK_LOCATION, "getHosts",
                "()[Ljava/lang/String;");
        if (jthr) {
            ret = printExceptionAndFree(env, jthr, PRINT_EXC_ALL,
                "hdfsGetHosts(path=%s, start=%"PRId64", length=%"PRId64"):"
                "BlockLocation#getHosts", path, start, length);
            goto done;
        }
        jFileBlockHosts = jVal.l;
        if (!jFileBlockHosts) {
            fprintf(stderr,
                "hdfsGetHosts(path=%s, start=%"PRId64", length=%"PRId64"):"
                "BlockLocation#getHosts returned NULL", path, start, length);
            ret = EINTERNAL;
            goto done;
        }
        //Figure out no of hosts in jFileBlockHosts, and allocate the memory
        jNumBlockHosts = (*env)->GetArrayLength(env, jFileBlockHosts);
        blockHosts[i] = calloc(jNumBlockHosts + 1, sizeof(char*));
        if (!blockHosts[i]) {
            ret = ENOMEM;
            goto done;
        }

        //Now parse each hostname
        for (j = 0; j < jNumBlockHosts; ++j) {
            jHost = (*env)->GetObjectArrayElement(env, jFileBlockHosts, j);
            jthr = (*env)->ExceptionOccurred(env);
            if (jthr || !jHost) {

View on GitHub (pinned to 2add963021)

Solutions

Align the libhdfs native library version with the hadoop jars of the same distribution
Generate CLASSPATH with `hadoop classpath` from the matching installation
Treat EINTERNAL from this call as non-retryable and report the file and range from the message

Defensive patterns

Strategy: fallback

Try / catch

if (hdfsGetHosts(fs, path, start, length, &hosts) == -1) {
    /* EINTERNAL: JVM returned null; treat as unavailable, do not retry the same call */
    hosts = NULL;  /* proceed without locality info */
}

Prevention

Keep libhdfs.so and hadoop jars from the same distribution
Generate CLASSPATH with `hadoop classpath` from the matching install
Design locality-aware callers to degrade gracefully when host info is unavailable

When it happens

Trigger: Version mismatch between the native libhdfs and the hadoop-hdfs jars on CLASSPATH; custom BlockLocation implementations; JVM state corrupted by earlier pending exceptions.

Common situations: Client machines with stale Hadoop jars; data-locality probes (HBase-style region placement) in mixed-version clusters; shaded Hadoop classes inside embedding frameworks.

Related errors

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/cb255f0aac4bbaf9. Report an issue: GitHub.