[HADOOP-1265] TaskTracker won't bind to localhost

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 0.12.3
• Fix Version/s: 0.15.0
• Component/s: ipc

Environment

The firewall looks (something) like this:
iptables -P INPUT ACCEPT
iptables -P FORWARD ACCEPT
iptables -P OUTPUT ACCEPT
iptables -Z

1. all local connections are OK
   iptables -A INPUT --destination 127.0.0.1 --in-interface lo -j ACCEPT
   iptables -A OUTPUT --destination 127.0.0.1 --out-interface lo -j ACCEPT

1. accept all traffic that originates internally
   iptables -A INPUT --match state --state RELATED,ESTABLISHED -j ACCEPT
   iptables -A OUTPUT --match state --state RELATED,ESTABLISHED -j ACCEPT

1. <snip: open some ports>

1. default target is drop for all unspecified packets
   iptables -P INPUT DROP
   iptables -P FORWARD DROP
   iptables -P OUTPUT DROP

• Assignee: Unassigned
• Reporter: Eelco Lempsink
• Created: 17/Apr/07 22:39
• Updated: 24/Oct/07 21:42
• Resolved: 11/Oct/07 17:21

Description

Connecting to the TaskTracker with the default configuration fails with the firewall settings as mentioned in the Environment. This means the job seems to start, but then will hang with all tasks at 0%

Also, setting mapred.tasktracker.dns.interface to 'lo' has no effect. I would expect this bound the TaskTracker in such way that local connections would be made.

To make it work, I have to explicitly poke a hole in the firewall:

1. hadoop
   iptables -A INPUT --protocol tcp --destination-port 50050 -j ACCEPT
   iptables -A OUTPUT --protocol tcp --destination-port 50050 -j ACCEPT

While in practise a Hadoop will often run on a cluster (so the firewall has to be opened anyway), I don't think this should be the default behaviour, because it is highly confusing.

Issue Links

• relates to: HADOOP-1202 Jobtracker & Namenode can only bind to one interface