[HADOOP-16114] NetUtils#canonicalizeHost gives different value for same host

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 2.7.6, 3.1.2
• Fix Version/s: 3.1.3
• Component/s: net
• Release Note: The above patch will resolve the race condition
• Assignee: Praveen Krishna
• Reporter: Praveen Krishna
• Created: 15/Feb/19 08:18
• Updated: 07/Mar/19 11:23
• Resolved: 07/Mar/19 11:10

Description

In NetUtils#canonicalizeHost uses ConcurrentHashMap#putIfAbsent to add an entry to the cache

  private static String canonicalizeHost(String host) {
    // check if the host has already been canonicalized
    String fqHost = canonicalizedHostCache.get(host);
    if (fqHost == null) {
      try {
        fqHost = SecurityUtil.getByName(host).getHostName();
        // slight race condition, but won't hurt
        canonicalizedHostCache.putIfAbsent(host, fqHost);
      } catch (UnknownHostException e) {
        fqHost = host;
      }
    }
    return fqHost;
}

 

If two different threads were invoking this method for the first time (so the cache is empty) and if SecurityUtil#getByName()#getHostName gives two different value for the same host , only one fqHost would be added in the cache and an invalid fqHost would be given to one of the thread which might cause some APIs to fail for the first time `FileSystem#checkPath` even if the path is in the given file system. It might be better if we modify the above method to this

 

  private static String canonicalizeHost(String host) {
    // check if the host has already been canonicalized
    String fqHost = canonicalizedHostCache.get(host);
    if (fqHost == null) {
      try {
        fqHost = SecurityUtil.getByName(host).getHostName();
        // slight race condition, but won't hurt
        canonicalizedHostCache.putIfAbsent(host, fqHost);
        fqHost = canonicalizedHostCache.get(host);
      } catch (UnknownHostException e) {
        fqHost = host;
      }
    }
    return fqHost;
}

 

So even if other thread get a different host name it will be updated to the cached value.

Issue Links

• links to: GitHub Pull Request #493