Mailing-List: contact notifications-help@accumulo.apache.org; run by ezmlm
Precedence: bulk
Reply-To: jira@apache.org
Date: Tue, 7 Jun 2016 21:28:21 +0000 (UTC)
From: "Dave Marion (JIRA)" <jira@apache.org>
To: notifications@accumulo.apache.org
Message-ID: <JIRA.12976447.1465306852000.19943.1465334901039@Atlassian.JIRA>
In-Reply-To: <JIRA.12976447.1465306852000@Atlassian.JIRA>
References: <JIRA.12976447.1465306852000@Atlassian.JIRA> <JIRA.12976447.1465306852295@arcas>
Subject: [jira] [Comment Edited] (ACCUMULO-4331) Make port configuration and
 allocation consistent across services
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Tue, 07 Jun 2016 21:28:23 -0000


    [ https://issues.apache.org/jira/browse/ACCUMULO-4331?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15319456#comment-15319456 ] 

Dave Marion edited comment on ACCUMULO-4331 at 6/7/16 9:27 PM:
---------------------------------------------------------------

So, based on the comments here, it seems the consensus is to

1. Keep zero as a valid port option, which means any ephemeral port
2. Keep tserver.port.search, but maybe change it such that it performs a 1-up attempt instead of random
3. Allow the user to specify a range of ports to be used (M-N). [new feature]
4. If user specifies a range of ports, and tserver.port.search is disabled, any port in that range is still valid.
5. If user specifies a range of ports, and tserver.port.search is enabled, we must try all the ports in the range before searching


was (Author: dlmarion):
So, based on the comments here, it seems the consensus is to

1. Keep zero as a valid port option, which means any ephemeral port
2. Keep tserver.port.search, but maybe change it such that it performs a 1-up attempt instead of random
3. Allow the user to specify a range of ports to be used (M-N). [new feature]
4. If user specifies a range of ports, and tserver.port.search is disabled, any port in that range is still valid.

> Make port configuration and allocation consistent across services
> -----------------------------------------------------------------
>
>                 Key: ACCUMULO-4331
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-4331
>             Project: Accumulo
>          Issue Type: Bug
>    Affects Versions: 1.8.0
>            Reporter: Dave Marion
>             Fix For: 1.8.0
>
>
> There was some discussion in ACCUMULO-4328 about ports, so I decided to track down how the client ports are configured and allocated. Issues raised in the discussion were:
>  1. The port search feature was not well understood
>  2. Ephemeral port allocation makes it hard to lock servers down (e.g. iptables)
> Looking through the code I found the following properties allocate a port number based on conf.getPort(). This returns the port number based on the property and supports either a single value or zero. Then, in the server component (monitor, tracer, gc, etc) this value is used when creating a ServerSocket. If the port is already in use, the process will fail.
> {noformat}
> monitor.port.log4j
> trace.port.client
> gc.port.client
> monitor.port.client
> {noformat}
> The following properties use TServerUtils.startServer which uses the value in the property to start the TServer. If the value is zero, then it picks a random port between 1024 and 65535. If tserver.port.search is enabled, then it will try a thousand times to bind to a random port.
> {noformat}
> tserver.port.client
> master.port.client
> master.replication.coordinator.port
> replication.receipt.service.port
> {noformat}
> I'm proposing that we deprecate the tserver.port.search property and the value zero in the property value for the properties above. Instead, I think we should allow the user to specify a single value or a range (M-N). 


--
This message was sent by Atlassian JIRA
(v6.3.4#6332)